Proceedings of The International Association of Forensic Linguists [PDF]

www.ncjrs.gov/pdffiles1/nij/grants/219559.pdf, finding that over a quarter of federal capital appeals were dismissed on ...... creating the first written constitution in History, further adding to Montesquieu's description the concept of 'checks ..... Montesquieu ([1748] 1973) Do espírito das leis [The spirit of laws]. São Paulo: Abril ...

5 downloads 12 Views 12MB Size

Recommend Stories


Proceedings of The International Association of Forensic Linguists' Tenth Biennial Conference
Before you speak, let your words pass through three gates: Is it true? Is it necessary? Is it kind?

proceedings of the 41st international association for danube research conference
In the end only three things matter: how much you loved, how gently you lived, and how gracefully you

Proceedings of the - ICA2016 [PDF]
Sep 9, 2016 - Libro digital, PDF. Archivo Digital: descarga y online. ISBN 978-987-24713-6-1. 1. Acústica. 2. Acústica Arquitectónica. 3. Electroacústica. I. Miyara,. Federico II. Miyara, Federico ...... large amounts of calibrated soft porous mi

Proceedings of the 9th international clay conference
Seek knowledge from cradle to the grave. Prophet Muhammad (Peace be upon him)

PROCEEDINGS OF THE 6th INTERNATIONAL MANAGEMENT CONFERENCE
Don't fear change. The surprise is the only way to new discoveries. Be playful! Gordana Biernat

Adjunct Proceedings of the 8th International Conference
The happiest people don't have the best of everything, they just make the best of everything. Anony

Proceedings of the International Conference on Data Engineering and [PDF]
We have done experiment on 822 different documents in which 522 prepared in Text file format and 300 in PDF (Portable Document Format). Each document containing at least five Indian Languages and more than 800 words. The documents belonged to differe

Proceedings of the International Citrus Canker and [PDF]
Nov 7, 2005 - Second International Citrus Canker and Huanglongbing Research Workshop 3. TUESDAY NOVEMBER 8, 2005: CURRENT STATUS OF CITRUS CANKER. 7:30 – 8:00 AM Registration. 8:00 – 8:15 AM Welcome/Introduction Andy LaVigne, CEO Florida Citrus M

Proceedings of the International Conference on Algebra 2010: ... [PDF]
An ADL L is called a k-ADL,” if to each a e L, (al” = (a']" for some ac' e L. An ADL L is a k-ADL if and only if to each a e L, there exists y e L such that a. A y = 0 and a V y is dense. An ADL L is called normal” if every prime ideal contains

Idea Transcript


Proceedings of The International Association of Forensic Linguists’ Tenth Biennial Conference

Edited by Samuel Tomblin, Nicci MacLeod, Rui Sousa-Silva and Malcolm Coulthard Editorial Assistant Andrea Nini

ISBN: 978 1 85449 432 0

Published in 2012 by the Centre for Forensic Linguistics Centre for Forensic Linguistics Aston University Aston Triangle Birmingham B4 7ET, UK.

© Copyright remains solely with individual authors

Contents Editors’ introduction

6

Part 1: The discourse of forensic contexts Language struggle and power abuse through diglossia in a murder trial Giorgos Georgiou

10

The construction of admissions of fault through American rules of evidence: speech, silence, and significance in the legal creation of liability Janet Ainsworth

21

I object, or did I? Mel Greenlee

35

Improving the comprehensibility of U.S. pattern jury instructions Bethany K. Dumas

49

Garment, or upper-garment? A matter of interpretation? Eva Ng

58

Appreciate others’ beauty as well as one’s own: a contrastive multimodal discourse analysis of two courtroom trials Yuan Chuanyou

73

Language and the construction of a positive identity among inmates in Kenyan jails Emmanuel Satia

92

Norm-enacting activity as an object of study in forensic linguistics: propositions and first impression João Pedro Pádua

104

Non-adversarial justice: the changing role of courtroom participants in an Indigenous sentencing court Natalie Stroud

115

Part 2: Linguistic investigation and evidence Sincere and deceptive statements in Italian criminal proceedings Tommaso Fornaciari and Massimo Poesio

126

The false report during an emergency call: using discourse analysis to detect deceit Marty Laforest

139

The rake’s progress: linguistic strategies for deception Isabel Picornell

153

-3-

An historical forensic linguistic analysis of contested letters in the Forrest Reid collection Yvonne McGivern

169

Investigating formulaic language as a marker of authorship Samuel Tomblin

177

The use of sequences of linguistic categories in forensic written text comparison revisited Núria Bel, Sheila Queralt Estevez, Maria S. Spassova and M. Teresa Turell

192

Whose Tweet? Authorship analysis of micro-blogs and other short-form messages Nicci MacLeod and Tim Grant

210

‘There are letters for you all on the sideboard’: what can linguists learn from multiple suicide-note writers? Jess J. Shapero and Susan A. Blackwell

225

Did he have an accent? Forensic speaker descriptions of unknown voices Mark Griffiths

245

The study of inter- and intra-speaker variation towards an index of idiolectal similitude Núria Gavaldà-Ferré

262

Computer-mediated investigative interviews: a potential screening tool for the detection of insider threat Marisa Jenkins and Coral J. Dando

272

Imaginative leaps in trademark law Ron R. Butters

283

Part 3: Language analysis for the determination of origin Arabic varieties and LADO: how can LADO deal with variance? Judith Rosenhouse

290

‘Faked’ or truthful second language proficiency: assessing claims Margaret van Naerssen

306

Part 4: The relationship between media reports and crime Corruption in the Spanish news: verbalizing crime for public opinion María Ángeles Orts and Ángela Almela Wikiforensics and the transcoding of innocence and guilt in the Meredith Kercher case Maria Bortoluzzi and Lieve Gies

-4-

315

325

Part 5: Ethical and moral issues in forensic linguistics Statutory interpretation, morality, and the text Lawrence M. Solan

338

Retiring President’s closing address: ethics, best practices, and standards Ronald R. Butters

351

Ethics and method in forensic linguistics Lawrence M. Solan

362

-5-

Editors’ introduction In July, 2011, Aston University’s Centre for Forensic Linguistics hosted the 10 th Biennial Conference of The International Association of Forensic Linguists. The conference was attended by over 130 delegates from approximately 31 countries, and, as with previous conferences, represented the diverse range of topics related to forensic linguistics and the language of the law that makes our field so stimulating. This collection represents a selection of papers that were presented at the conference. Since its formation in 1994, The International Association of Forensic Linguists has continued to thrive and now, 18 years later, what was once an emerging field, “too scattered to justify one’s naming of a subject area ‘forensic linguistics’” (French & Coulthard, 1994: vii) has become an established subdiscipline of applied linguistics as evidenced through a range of introductions to the field (e.g., Coulthard & Johnson, 2007; Gibbons, 2003; Gibbons & Turell, 2008; Solan & Tiersma, 2005) and handbooks (e.g., Coulthard & Johnson, 2010; Tiersma & Solan, 2012), let alone a plethora of edited collections and monographs, far too many to single out only a representative few. We hope that this collection of papers will add to the growing wealth of literature and demonstrate the often creative ways in which members of our community have developed research methods and new perspectives for interpreting data. What makes this collection of papers most exciting is the range of authors represented—from IAFL Presidents, past and present, established scholars, professionals and arguably most importantly, students. Our community stands out as a group of scholars, practitioners and professionals who take pride in sharing the details of our work and fostering our students in an effort to sustain our field and promote growth. It is in this spirit that we have produced this electronic version of conference proceedings, so that we have a written record of our tenth biennial conference. The papers have been organised around five main themes: the discourse of forensic contexts, including the courtroom and beyond; linguistic investigation and evidence; language analysis for the determination of origin (LADO); the relationship between media reports and crime; and ethical and moral issues in forensic linguistics. The first six papers in this collection all deal with issues surrounding discourse in the courtroom. Giorgos Georgiou opens with his analysis of court transcripts from a murder trial in Cyprus, where a diglossic situation exists. Georgiou argues that code-switching between the dialect, Cypriot Greek, and the Standard, Standard Modern Greek, indicates a lack of power and is associated with social status. Georgiou demonstrates the ways in which codeswitching is related to language struggle and argues that inequality arises as a result. The rules of evidence in law are of central importance to Janet Ainsworth and in her paper, she examines two common law evidentiary rules, revealing normative and ideological assumptions about how language is used. Mel Greenlee explores the problem of objecting during a trial and how an ineffective objection precludes a future appeal on that issue. Greenlee draws on a small set of Californian cases to examine the language of trial objections. She also considers how the reviewing courts interpreted those objections. Bethany Dumas describes her work with the Tennessee Judicial Conference Committee on Pattern Jury Instructions and outlines how linguists can assist with the comprehensibility of pattern jury instructions. In this paper, Dumas outlines and exemplifies five steps which she argues are necessary for the production and revision of pattern jury instructions. Two papers follow which deal specifically with the adversarial courtroom. Eva Ng’s research explores the problem of ambiguity created by polysemous words for court interpreters during an adversarial trial in Hong Kong. She focusses specifically on the context of an interpreter-mediated rape trial. Yuan Chuanyou contrasts the inquisitorial and adversarial trial systems and describes his analysis of two courtroom extracts, one from a -6-

Chinese courtroom and one from an American. His examines in particular participants’ gestures, movements and speech leading him to conclude that there are differences which are attributable to the different trial systems. The next three papers move the focus on forensic discourse analysis away from the courtroom and into other forensic-related contexts. Using critical discourse analysis, Emmanuel Satia describes the ways in which a group of prison inmates in Kenya construct a positive identity when writing to a religious leader. João Pedro Pádua presents his research on parliamentary contexts, bringing to attention the ways in which legal norms are created and enacted; an area that so far has received very little attention. Natalie Stroud then presents her research on courtroom participants in an Indigenous sentencing court in Australia. Stroud adopts an interdisciplinary approach to compare the role of participants in a conventional court with a non-adversarial Indigenous sentencing court. Our focus then turns to linguistic investigation and evidence. Tommaso Fornaciari and Massimo Poesio begin by describing their research on authentic Italian trial hearings where they train statistical models to classify testimonies as either sincere or deceptive. Marty Laforest presents a model for deception detection in the context of calls to the emergency services whilst Isabel Picornell offers a slightly different perspective to deception; rather than seeing texts as either deceptive or truthful, she views deceptive communication as a combination of both truthful and deceptive elements, and therefore focuses on the strategies used by deceivers as they move between these camps. Moving onto authorship analysis, Yvonne McGivern describes her work on the Forest Reid Papers and particularly two disputed letters. She assesses the probability of their authorship and raises questions about methodology. Samuel Tomblin offers research on a previously unexplored marker of authorship, formulaic language, and assesses the potential for this marker as a new tool in the forensic linguists’ tool kit. Núria Bel, Sheila Queralt Estevez, Maria S. Spassova and M. Teresa Turell describe a series of studies on Spanish language data, which explore the discriminatory capacity of tag sequences for application to authorship attribution tasks. Nicci MacLeod and Tim Grant describe their research into the authorship of a different type of text—micro-blogs. Finally on the theme of authorship, Jess Shapero and Susan Blackwell report their research on suicide notes by comparing intra- and inter-author variation through the use of a semantic tagger. Mark Griffiths presents his research on earwitness accounts of voices and proposes a preliminary audiofit typology for eliciting voice and accent data from non-linguists. Núria Gavaldà Ferré describes her work on individual’s idiolectal style, focussing on certain phonological variables from the variety of Southern British English. Marisa Jenkins and Coral Dando focus on secure organisations and the risk of insider attacks from employees. They argue that computer-mediated interviews may be a viable substitution for the face-toface interview in this investigative context. Ron Butters concludes this section by exploring the notion of ‘imaginativeness’ in relation to trademark disputes. Two LADO papers are included: Judith Rosenhouse deals specifically with LADO as it relates to Arabic and the associated difficulties, which leads her to propose the development of automated software as an additional tool in the determination of origin. Margaret van Naerssen then describes her research, which assesses the ability of non-native speakers of English to ‘fake’ their proficiency level. There then follow two papers which explore the relationship between the media and crime. María Ángeles Orts and Ángela Almela investigate a 500,000 word corpus of Spanish news articles to examine the words used by the Spanish news media to describe corrupt practices. Maria Bortoluzzi and Lieve Gies employ a theoretical framework of discourse analysis and metaphor studies to describe the social medium, ‘wikiforensics’, where the guilt or innocence of Meredith Kercher’s alleged murderer is speculated. -7-

In the final section, three papers highlight an area of forensic linguistics that will undoubtedly receive more attention as our field grows: ethics and morality. Larry Solan explores statutory interpretation and outlines occasions when the intent or purpose of the law is flouted. Following on from this, Ron Butters presents his ex-oficio plenary address where he brings into focus the need for a statement of ethics as part of the IAFL constitution. To commemorate Butters’ term as the President of the IAFL, Larry Solan kindly accepted an invitation to write a response to the plenary address, which we include in these pages as a way to stimulate discussion in an area close to Ron’s heart. The publication of this collection of conference papers is entirely down to the hard work and commitment of the authors who allowed us to include their research. We have enjoyed reading about their work in more detail and we hope that you will also find this collection to be as stimulating and varied as the conference. All of the papers included in these pages have been peer-reviewed. However, what makes this collection special is that neither the Editors, nor Aston University, assume any rights to these papers. The copyright remains with individual authors. To this end, we welcome, and in fact encourage the authors to submit their papers for publication elsewhere. We also owe a great deal of thanks to all of the delegates of IAFL10 who made the conference such a success, and whose comments, questions and feedback will undoubtedly have shaped the papers included in this collection. Finally, we the editors would like to convey our most sincere gratitude to the enthusiasm and dedication shown by our Editorial Assistant, Andrea Nini, who ensured that we were organised throughout this undertaking. Samuel Tomblin Nicci MacLeod Rui Sousa-Silva Malcolm Coulthard Aston University, Birmingham, March 2012 References French, P. & Coulthard, M. (1994) Forensic Linguistics: Editorial introduction, Forensic Linguistics: The International Journal of Speech, Language and the Law, 1 (1), vii-ix. Coulthard, M. & Johnson, A. (2007) An Introduction to Forensic Linguistics: Language in evidence, London, Routledge. Coulthard, M. & Johnson, A. (2010) The Routledge Handbook of Forensic Linguistics, Abingdon: Routledge Gibbons, J. (2003) Forensic Linguistics: An Introduction to Language in the Justice System, Oxford: Blackwell Publishing. Gibbons, J. & Turell, M.T. (2008) Dimensions of Forensic Linguistics, Amsterdam: John Benjamins Publishing Co. Solan, L. & Tiersma, P. (2005) Speaking of Crime: The Language of Criminal Justice, Chicago: Chicago University Press.

-8-

Tiersma, P. & Solan, L. (2012) The Oxford Handbook of Language and Law, Oxford: Oxford University Press.

-9-

Language struggle and power abuse through diglossia in a murder trial Giorgos Georgiou University of Cyprus, Cyprus [email protected] Abstract There has been an unexpected surge of research interest in the phenomena of bilingual speech, and in particular, code-switching. However, little or no attention has been given in the international literature to how code-switching adversely influences the trial procedure, especially when a Standard and a Dialect are involved, or more generally, when a recognized “high” and “low” code are used as in the situation of diglossia (Ferguson, 1959). In this paper this issue is explored using excerpts of examination and cross-examination sequences from the court transcript of a single criminal trial. The transcript involves code-switching from a dialect (Cypriot Greek) to a standard (Standard Modern Greek), which is a crucial indicator of lack of power in this domain, in accordance with the societal status and language habits of the people involved. Furthermore, a matched guise technique experiment is applied to explore language attitudes towards the Dialect. The main principles of Critical Discourse Analysis are applied to demonstrate in what way code-switching is related to language struggle in the courtroom, the discourse dimensions of power abuse and the injustice and inequality that result from it. The paper suggests that power imbalance in the courtroom is not only situational, but societal (Eades, 2008). Keywords:

CODESWITCHING; CRITICAL DISCOURSE ANALYSIS; POWER; CYPRIOT GREEK

Introduction Lawyers are acutely aware of the power of the words they use as a means of assertiveness during a trial procedure (Conley et al., 1978; Eades, 2008; O’Barr, 1982). From a sociolinguistic perspective, the way of expressing oneself in the court may be crucial for the verdict, to determine if, for example, a murder was committed on purpose (i.e. homicide) or by accident (i.e. manslaughter). Systematic work has been carried out in the field of Critical Discourse Analysis regarding the impact of lexical choices on the construction of reality (Jacquemet, 1996; Philips, 1998; Trinch, 2003), the same scheme of analysis adopted here to refer to the court procedure, when a dialect is involved. Danet’s (1980) analysis can be recalled in the wellknown abortion case about the difference between the words “baby” and “fetus” and the direct legal consequences of the one or the other lexical choice. What is said in the courtroom is the one thing (primary courtroom reality) and what the actual events are (secondary reality) is another (Hale and Gibbons, 1999). Inside the courtroom a lexical struggle is in action meaning a struggle over the choice and meaning of words to name experiences (Eades, 2006: 119). However, as evidence from some rape trials shows (e.g., Drew, 1992; Matoesian, 1993), the labelling is not limited to lawyers; witnesses sometimes resist lawyers’ constructions by providing “alternative descriptions, either with or without overt markers of correction”, as Drew (1992: 486—491) put it. And Matoesian points out: although the resources with which to direct the course and outcome of interaction are asymmetrically distributed between the [Defence Attorney] and [the Victim], both -10-

must struggle to negotiate meaning, to make their accounts count, and to reproduce their systematic courtroom relationship as a micro-mode of domination. (1993: 170) What about when the choices refer to code-switching between a dialect and a standard variety? Are there any consequences? Is there a “code struggle”, just as Eades identified a “lexical struggle”? Can the alternation of codes and a manipulation of the code choice between a standard and a dialect impact the credibility of a witness’s testimony? The aim of this paper is to answer these questions through a systematic analysis of the data (excerpts of examination and cross examination sequences from the court transcript of a single criminal trial where codeswitching is involved) in the Critical Discourse Analysis paradigm. The background The superposed variety in the Republic of Cyprus is Standard Modern Greek, as it is used in the Hellenic Republic of Greece, which no Greek Cypriot uses in everyday conversations, while the native variety is Cypriot Greek. The dialect is acquired in a natural way, whereas Standard Greek is usually taught and learned at school, i.e. after the critical period for acquisition, in an almost “artificial” way. Thus, Greek-speaking Cyprus is arguably a diglossic speech community, displaying all the characteristics of diglossia mentioned in the classic definition of Ferguson (1959), which constitute a distinction between superposed and naturally acquired variety and functional differentiation of the two depending on the communicative context. According to Ferguson, diglossia is a kind of bilingualism in a society in which one of the languages is (H), i.e. has high prestige, and another of the languages is (L), i.e. has low prestige. The official language of Greek-speaking Cyprus is Standard Modern Greek, which was defined by Moschonas (2002) as the coded, stereotyped, established variety of Greek. As far as Cypriot Greek is concerned, it is considered to be an oral dialect that has not undergone procedures of coding and stereotyping (strict orthography rules, writing of grammars and lexicons, use in education, etc.). When it comes to the use of the two varieties, the Cypriot dialect is used in informal communication, while Standard Greek is used in written and in formal communicative situations; for example, in political speeches or in university lectures, on television and, of course, as in our case, in the court. This mainstream view is endorsed by other researchers as well, such as Katsoyannou et al. (2006), who stress that the use of Standard Modern Greek is more or less imposed in sectors of social life such as school, administration, the army, the media and also in formal writing. Cypriot Greek is certainly distinguishable from Standard Greek. There are important differences between the two varieties as far as vocabulary and phonology, morphology and syntax are concerned; see, for example, Tsiplakou (2006a) and more recently Grohmann and Leivada (2011) for overviews. Usually, then, the diglossic model with Standard Modern Greek as the ‘H’ (High) code and Cypriot Greek as the ‘L’ (Low) code is taken for granted within the Greek-speaking community of Cyprus. However, research by Tsiplakou (2006b) has shown that in conversations between students and professors in seminars and lectures—where Standard Greek is not only expected but almost imposed—the Low variety appears unexpectedly. In this paper, it is argued that this phenomenon, the unexpected use of the Low variety in a formal context, namely, the courtroom, may be a resource for some witnesses or defendants to draw upon in order to assert their own version of the facts in the face of hostile -11-

questioning. I use the term “code struggle” for the purposes of this research. This term has this specific meaning—namely a power negotiation—only to the courtroom domain. Power deals with the fundamental issue of inequality (Fairclough, 1989) and this relationship exists, of course, in the legal context as well (Cotterill, 2003; Mayr, 2004). This sense of power (in the courtroom, but in general as well) is encompassed in the notion of hegemony, which means preponderant power in a political context; the ability of some groups to subordinate others. Power represents an ideological struggle and in the course of this research, Eades’ (2008) characterization, that power imbalance in the courtroom is not only situational, but societal as well, seems particularly appropriate. Everybody speaks the dialect in Cyprus, including judges and even the president of the country, but if you cannot “play by the rules” inside the court, if you cannot switch to the standard, at least to an acceptable version of the standard, then you have to sustain the consequences. In simple words, this means that the witnesses who use the dialect may be linguistically disadvantaged during the trial, which can affect the outcome and/or their credibility. Language attitudes It is known from numerous studies that listeners react subjectively to speakers according to the dialect that they use. Studies conducted in the United States have revealed the preference of both majority—and minority-group members for speakers of standard English over Black Vernacular English (Tucker and Lambert, 1969), and over Mexican-American accented English (Ryan and Carranza, 1977). In Canada, where studies of this type were first developed by Lambert and his colleagues, it has been demonstrated that French Canadian listeners view speakers of European French more favourably than speakers of Canadian French (d’Anglejan and Tucker, 1973; Lambert et al., 1960), and upper-class French Canadian speakers more favourably than lower-class French Canadian speakers (Brown, 1969). Similar studies conducted in Great Britain by Giles and his associates demonstrate that persons speaking with a prestigious accent—in the case of Britain “received pronunciation”—are evaluated more positively on semantic differential scales than are persons who speak with either regional, foreign, or lower-class accents (Bourhis et al., 1975; Giles, 1970; Giles and Bourhis, 1976). To speak about the impact of dialectal use we have to refer to language attitudes. A number of empirical studies have been conducted to examine Greek Cypriots’ attitudes towards Standard versus Cypriot Greek. Findings from the Greek Cypriot context (Papapavlou and Pavlou, 1998; 2007) seem to be in agreement with findings from other bidialectal or bilingual settings in that there seems to be a strong relation between competence/status-related traits such as intelligence, education and ambition with the standard variety. However, a similarly strong association between Standard Greek and solidarity and social attractiveness-related traits does not seem to be evident, as Greek Cypriot speakers of Standard Greek are not perceived by their peers to be friendlier, more sincere, kind and humorous than their dialect counterparts (Papapavlou, 1998). As in the case of other diglossic, bidialectal speakers, Greek Cypriots too downgrade their local variety when compared to the standard as a result of various social, political and cultural factors. So what is the situation in the court, where the judge is the actual evaluator of the linguistic repertoire of witnesses and defendants?

-12-

The experiment Duplicating the experimental design employed by O’Barr and his colleagues (1982) that involves the use of a subjective reaction type of test, a variant of what is known as the “matched guise technique”, I will show that various aspects of a witness’s speech play an important role in the formation of impressions of witnesses. Specifically, I will examine the role of convincingness, truthfulness and trustworthiness which are important as indicators of innocence or—in their absence—of guilt (Berk-Seligson, 1984; Hale, 2004). Subjects in verbal guise studies are typically asked to rate the speakers whom they have heard on a recording along various socio-psychological dimensions such as intelligence, strength, honesty, passivity, and so on. Subjects—here the judges and the lawyers acting like judges—heard the same dialectal and non-dialectal standard mode for use in the present study accordingly. The testimony (duplicated from the original where the Standard was used) explained why the defendant exceeded the speed limit in the highway and was five minutes long. The testimony was recorded on audio tape with two male actors playing the parts of witnesses speaking in the Standard and in the Dialect. In this recreation of the testimony the actors tried to replicate as closely as possible the speech characteristics of the Standard and the Dialect, accordingly. Each participant in the study was given a standardized questionnaire which contained ten lines representing one hundred-point scales. At one end of each line appeared the adjective (“trustworthy”, “convincing”). At the other end appeared the opposite of the evaluative term, as shown in Figure 1. Not Trustworthy 1: 2: […] 10 : 20 : 30 40: 50: 60: 70: 80: 90: 100%

Trustworthy

Figure 1: Intelligence Scale

The Sample 16 people participated in the experiment, 5 judges and 11 lawyers, divided into pairs of 8. Each pair heard one version. Pair one heard the version in the Dialect and pair two heard the version in the Standard. In the sample, 11 were men and 5 were women with a mean age of 57.6 years. Lawyers were asked to function as judges (because of the lack of more than 5 judges in our experiment since they are very reluctant in participating in this kind of experiment; nevertheless, are experienced lawyers that become judges in the Cypriot law system).

-13-

Table 1: Scores for each attribute

ATTRIBUTE

N

Mean

St. Dev.

Sig.

Use of the Dialect

16

70.33

2.2

p<.05

Use of the Standard

16

91.75

4.0

p<.05

Use of the Dialect

16

68.8

3.2

p<.05

Use of the Standard

16

88.25

2.1

p<.05

Use of the Dialect

16

81.1

1.9

p<.05

Use of the Standard

16

92.6

1.5

p<.05

Convincingness

Truthfulness

Trustworthiness

Table 1 presents a summary of the findings in this regard. Scores for each attribute range between 1, being the most positive evaluation, and 100, being the most negative. A t-test was applied to the subjects’ impressions regarding the degree to which the witness seemed convincing, truthful and trustworthy, comparing the mean answers of those who heard the “dialectal” version of the attestation with the means of those who heard the formal/standard version. Looking down the column labelled “mean” we see that in every case where the witness’s testimony was given in non-dialectal, Standard Modern Greek, the means reflected a more positive evaluation on each of the two socio-psychological attribute continua. This demonstrates that in the eyes of a judge, a speaker who uses the Dialect may be viewed negatively; though in questionnaires from a different study I conducted they say that the Cypriot Greek dialect has nothing to do with their evaluations, which rest on the law (Georgiou, forthcoming). Of course, linguists recognise that our metalinguistic awareness is something different from our actual attitudes. OK. They code-switch. Why? No Cypriot uses the Standard alone. It is code-switching or code-mixing that occurs in most cases. Only in formal speech domains is Standard the used alone, like television and not even always there. We have seen the outcome of this choice in the formation of judges’ impressions. But there are also complex networks binding code-switching with certain intentions. It is clear that code-switching is not without meaning. Similar to other actions, such as gestures, for example, code-switching functions as a meaningful signalling device helping speakers to convey meaning and listeners to understand the intended meaning (Shin and Milroy, 2000: 352). There are several theoretical models to explore code-switching like Gumperz’s (1982: 66) “we codes-they codes” identity model and the Markedness Model adapted by Myers-14-

Scotton (1998) which is based on the claim that “people usually do what they believe is likely to have the best overall outcome” by using a certain linguistic repertoire composed of the various languages, dialects and styles the speaker is able to use. A Conversation Analysis Approach that places emphasis on the interactional perspective, interpreting meanings in a turn-by-turn basis, is also a possibility. The Conversation Analysis Model consists of an analytical framework focusing on the ways in which speakers locally express and interpret meanings on a turn-by-turn basis; Auer (1984: 3) strongly advocates a sequential turn-by-turn analysis of code-switching as he believes that each individual code-switch is closely tied to the specific context in which it appears. This model is used because it gives emphasis in the turn-by-turn analysis with reference to basic Conversational Discourse Analysis principles. Below, some extracts of the murder trial that were used for the analysis are presented, where codeswitching appears very often. In the extracts a prosecution lawyer interacts with a defense witness. Questions (Q) are made by lawyers (L) and answers (A) are given by witnesses (W). Dialectal types are given in italics. Q (L). Υπήρχαν κλειδαριές στις πόρτες; Ipirxan kliðarjes stis portes? ‘Were there any locks in the doors?’ A (W). Μάλιστα, υπήρχαν. Malista, ipirxan. ‘Yes, there were’. Q (L). Η μια ήταν ρόλεξ, η άλλη με κλειδαριά μεγάλη; I mija itan rollex, i alli me kleidarja meγali ‘The one door was rollex, and the other with a big lock?’ A (W). Ναι. Nai ‘Yes.’ Q (L). Τσακροκλειδωνιάν εννοείτε; Tsakrokliδonjan ennoite? ‘A big lock you mean?’ A (W). Ναι. Nai ‘Yes.’ Q (L). Τι έγινε μετά; Ti eγine meta? ‘What happened next?’ A (W). Πήγαμε για φαΐ και γύρω στις 11.30 εστραφήκαμε σπίτι. Piγame ja fai ke giro stis 11.30 estrafikame spiti. ‘We went for food and around 11.30 we came back home.’

-15-

Q (L). Και τι έγινε όταν στραφήκατε σπίτι; Kai ti eγine otan strafikate spiti? ‘And what happened when you came back home?’ In this interlocution, the lawyer himself code-switches to the dialect (“Tsakrokliδonjan”—big lock) or sustains the use of the dialect as chosen from the witness (“strafikate”—came back) to accommodate him, to “come closer”, as the lawyers say in their own words. Lawyers use the dialect when they want to be more aggressive towards the witness, using the same variety, in order to show that they have revealed him, that they “uncovered” his hidden thoughts as expressed in his mother tongue. This is shown in the following extract: Q (L). Επήες στην Αστυνομία και έκρυψες τα εγκαύματά σου και το έκαμες για να μην καταλάβουν ότι επήγες να σιονώσεις πεζίνα και θίνερ τζιμέσα τζαι άναψές τα. Epies stin astinomia ke ekripses ta egavmata su ke to ekames γia na min katalavun oti epiγes na sionosis pezina ke thinner tzimesa tze anapses ta. ‘You went to the Police and you hid your burns, and you made that for the others not to understand that you have tried to spill gasoline and thinner inside and that you started a fire.’ In other cases, the lawyer may lead the witness to code-switch (interestingly from Standard Greek to the Cypriot Greek dialect!) in an attempt to diminish her/his credibility and/or her/his narrative style, as the next example shows. Defendant: Δεν έκαμα εγώ έτσι πράγμα, ούτε φοβήθηκα οτιδήποτε. Το μόνο που φοβήθηκα εγώ είναι ότι θα βρω τον μπελά μου. Den ekama eγo etsi praγma ute foviθika otiδipote. To mono pu foviθika eγo ine oti θa vro ton bella mu. ‘I neither make such a thing nor am I afraid of anything. I am only afraid of getting into trouble.’ Lawyer to the defendant: Q (L). Τι σημαίνει να βρεις τον μπελά σου; Ti simeni na vris ton bella su? ‘What do you mean by “getting into trouble”?’ A (W). Ηταν να μπερτέψω. Itan na bertepso. ‘I was going to be blamed.’ Q (L). Γιατί να μπερτέψεις; γiati na bertepsis? ‘Why be blamed?’ -16-

In Greek, the phrase “βρίσκω τον μπελά μου” [vrisko ton bela mou] ‘getting into trouble’ is clear enough. Despite that, the lawyer insists on putting the defendant in a position to use the dialect leading him through metalinguistic remarks about getting into trouble by using the dialectal type “μπερτεύκω” [berteuko], ‘to be blamed’ (-Ti simeni na vris ton bella su? -Itan na bertepso. γiati na bertepsis?) which has several negative connotations when being used for culpable behaviour A final remark Code-switching, especially in a high hierarchical speech domain such as the court, serves several purposes and has multiple implications for all parties. Language struggle through diglossia has certain characteristics that can be revealed on a turn by turn basis, where the microlinguistic features and discourse dimensions of an embedded injustice and inequality appear. In a broad picture this is not just a yielding for control in a certain situation, but an ideological and societal struggle for power. I suggest that one of the social resources on which power and dominance are based is the privileged access to discourse and communication, meaning that language users or communicators have more or less freedom in the use of special discourse genres or styles, or in the participation in specific communicative events and contexts. Thus, lawyers have access to the certain language use which makes them powerful in court in contrast with the defendants and other witnesses who do not. Similarly, lack of power is also measured by its lack of active or controlled access to discourse: lay people have passive access to the justice system (judges, police officers) and can be controlled participants (suspects in court). If powerful speakers or groups enact or otherwise exhibit their power in discourse, we need to know exactly how this is done, because this is a way of resistance that leads to better justice. References Auer, P. (1984) (ed.) Code-switching in Conversation Language, Interaction and Identity. London: Routledge, 29—48. Berk-Seligson, S. (1984) The Bilingual Courtroom. Court Interpreters in the Judicial Process. Chicago: The University of Chicago Press. Bourhis, R. Y, H. Giles, and Lambert, W. E. (1975) Some consequences of accommodating one’s style of speech: a cross-national investigation. International Journal of the Sociology of Language 6: 55—72. Brown, B. (1969) The Social Psychology of Variations in French Canadian Speech Styles. Unpublished doctoral dissertation, McGill University. Conley, J. M., O’Barr, W.M. and Allan, L. (1978) The power of language: presentational style in the courtroom. Duke Law Journal 78 (6): 1375—1399. Cotterill, J. (2003) Language and Power in Court: A Linguistic Analysis of the O. J. Simpson Trial. Houndmills: Palgrave Macmillan.

-17-

D’Anglejan, A., and G. R. Tucker (1973) Sociolinguistic correlates of speech style in Quebec. In R. Shuy and R. Fasold (eds) Language Attitudes: Trends and Prospects. Washington, D.C.: Georgetown University Press. Danet, B. (1980) “Baby” or “fetus”?: language and the construction of reality in a manslaughter trial, Semiotica 32: 187—219. Drew, P. (1992) Contested evidence in courtroom cross-examination: the case of a trial for rape. In P. Drew and J. Heritage (eds) Talk at Work: Interaction in Institutional Settings. Cambridge: Cambridge University Press, 470—520. Eades, D. (2006) Lexical struggle in court: Aboriginal Australians vs. the state. Journal of Sociolinguistics 10 (2): 153—181. Eades, D. (2008) Courtroom Talk and Neocolonial Control. Berlin: Mouton de Gruyter. Ferguson, C. (1959) Diglossia. Word 15: 325—340. Fairclough, N. (1989) Language and Power. London: Longman. Georgiou, G. (forthcoming) Judges' evaluations of the Cypriot dialect during trial (to be published in Greek in “Nomiko Vima”). Giles, H. (1970) Evaluative reactions to accents. Educational Review 22: 211—27. Giles, H., and Bourhis R. Y. (1975) Language assimilation among West Indian immigrants in a British city. Language Sciences 38: 9—12. Grohmann, K.K. and Leivada, E. (2011) Interface ingredients of dialect design: Bi-x, sociosyntax of development, and the grammar of Cypriot Greek. In A.M. Di Sciullo (ed.) Towards a Biolinguistic Understanding of Grammar: Essays on Interfaces. Amsterdam: John Benjamins. Gumperz, J.J. (1982) Discourse Strategies. Cambridge: Cambridge University Press. Hale, S. (2004) The Discourse of Court Interpreting: Discourse Practices of the Law, the Witness and the Interpreter. Amsterdam: John Benjamins. Hale, S. and Gibbons, J. (1999) Varying realities patterned changes in the interpreter’s representation of courtroom and external realities. Applied Linguistics 20(2): 203—20. Jacquemet, Marco (1996) Credibility in Court: Communicative Practices in the Camorra Trials. Cambridge: Cambridge University Press.

-18-

Katsoyannou, M., Papapavlou, A., Pavlou, P. and Tsiplakou, S. (2006) Διδιαλεκτικές κοινότητες και γλωσσικό συνεχές: η περίπτωση της κυπριακής. [Bidialectal speech communities and the notion of linguistic continuum. The case of Cypriot Greek]. In M. Janse, B. D. Joseph, and A. Ralli (eds.) Proceedings of the Second International Conference on Modern Greek Dialects and Linguistic Theory. University of Patras. Laboratory of Modern Greek Dialects. Patras. Lambert, W. E., Hodgson, R. C., Gardner Fillenbaum S. (1960) Evaluation reactions to spoken languages. Journal of Abnormal and Social Psychology 60: 44—51. Mayr, A. (2004) Prison Discourse: Language as a Means of Control and Resistance Houndmills: Palgrave Macmillan. Matoesian, G. (1993) Reproducing Rape: Domination through Talk in the Courtroom. Chicago, Illinois: University of Chicago Press. Moschonas, S. (2002) Κοινή γλώσσα και διάλεκτος: Το ζήτημα της «γλωσσικής διμορφίας» στην Κύπρο [Common language and dialect: the issue of diglossia in Cyprus]. Nea Hestia 151: 898—928. Myers-Scotton, C. (1998) Codes and Consequences: Choosing Linguistic Varieties. Oxford: Oxford University Press. O’Barr, W. M. (1982) Linguistic Evidence: Language, Power, and Strategy in the Courtroom. New York: Academic Press. Papapavlou, A. (1998) Attitudes toward the Greek Cypriot dialect: sociocultural implications. International Journal of the Sociology of Language 18: 15—28. Papapavlou, A. and Pavlou, P. (1998) A review of the sociolinguistic aspects of the Greek Cypriot Dialect. Journal of Multilingual and Multicultural Development 19 (3): 212— 200. Papapavlou, A. and Pavlou, P. (2007) The interplay of bidialectalism, literacy and educational policy. In A. Papapavlou and P. Pavlou (eds) Sociolinguistic and Pedagogical Dimensions of Dialects in Education. Cambridge Scholars Press. Philips, S. (1998) Ideology in the Language of Judges: How Judges Practice Law, Politics and Courtroom Control. New York: Oxford University Press. Ryan, E. B., and Carranza, M. (1977) Ingroup and outgroup reactions toward Mexican American language and varieties. In H. Giles, (ed.) Language, Ethnicity and lntergroup Relations. London: Academic Press. Shin, J.S., and Milroy, L. (2000) Conversational codeswitching among Korean—English bilingual children. The International Journal of Bilingualism 4: 352—383. Trinch, S. (2003) Latinas’ Narratives of Domestic Abuse: Discrepant Versions of Violence. Amsterdam: John Benjamins. -19-

Tsiplakou, S. (2006a) Cyprus: language situation. In K. Brown (ed.) Encyclopedia of Linguistics (2nd edition). Oxford: Elsevier, 337—339. Tsiplakou, S. (2006b) The emperor’s old clothes: linguistic diversity and the redefinition of literacy. International Journal of the Humanities 2: 2345—2352. Tucker, G. R., and Lambert, W. E. (1969) White and Negro listeners’ reactions to various American English dialects. Social Forces 47: 463—68.

-20-

The construction of admissions of fault through American rules of evidence: speech, silence, and significance in the legal creation of liability Janet Ainsworth Seattle University, USA [email protected] Abstract The rules of evidence in law both govern the admissibility of evidence in trials and determine the scope of meaning to be accorded to it. An examination of two common law evidentiary rules reveals that such rules incorporate normative and ideological assumptions about language usage. The adoptive admissions rule provides that, when a person is confronted with an accusation of wrong-doing without expressly denying it, the allegation is presumed to be true. The evidence rule construing apology as an admission of fault presumes that apologetic language means ‘I’m sorry I did something wrong’ rather than ‘I’m sorry that something bad has happened to you.’ Evidence rules such as these operate as ideological “gates” to channel and constrain the legal interpretation of language used by juridical actors, and additionally serve as one of the means through which the linguistic ideology in law is exercised to create and maintain social hierarchies. Keywords:

ADOPTIVE ADMISSION; RULES OF EVIDENCE; SILENCE; APOLOGY; LINGUISTIC IDEOLOGY

Law is an inherently normative enterprise (Schlag, 1991), and nowhere is that truer than in its normative approach to what may seem to be purely procedural matters such as the rules governing the admission of evidence in court. This paper examines the way in which unexamined norms of language usage and meaning implicature are embedded within common law legal system through an examination of two American rules of evidence: one in which the law expressed a normative preference for speaking and consequently penalizes those who instead remain silent, and one in which the law expresses a normative preference for silence and legally penalizes those who instead speak. In both instances, the law’s normative approach is empirically baseless and linguistically naïve. The United States, like other common law nations, uses rules of evidence derived from the English adversarial trial advocacy system as gatekeeping rules, determining the admissibility of testimonial evidence can be introduced in a trial. The overarching standard for admissible evidence requires that it be both reliable—that is, worthy of credibility—and probative—that is, tending to prove some material fact at issue in the case. The emphasis on reliability explains the preference in common law hearsay evidence rules for the testimony of first-hand direct observers over secondhand reports of someone else’s observations. The relevance rule, similarly, is grounded in the common law’s privileging of testimony that has some tendency, even if slight, to prove a fact that the laws deems to be material in resolving the dispute at hand. Evidence rules are seen in law as a gatekeeper ensuring reliability and probative power in admitted testimony. However, a critical examination of how the evidence rules operate exposes the ways in which common law rules of evidence are grounded in assumptions about human nature and behavior that are neither empirically -21-

grounded nor, candidly, even anecdotally persuasive as believable accounts of communicative performance. A few examples suffice to make this point. For example, while ordinarily out of court utterances are excluded under the hearsay rules because it is impossible to assess their credibility, statements made by those who believe they are on the point of death are admissible in court as reliable evidence. The theory justifying the admission of this evidence is that deathbed utterances must be reliable because no one who believed they were about to die would willingly do so with a lie on their lips. This justification may once have been persuasive, perhaps, but it is difficult to believe today. Might a person, knowing that death is near, be tempted to take a last opportunity to shade the truth in order to make himself look better or to get lastminute revenge on an enemy? Likewise, what a patient says to a doctor for purposes of getting medical treatment is legally considered to be reliable and thus admissible evidence because the law presumes that no one would intentionally lie to a doctor. Yet, the law’s assumption that such statements are reliable and true flies in the face of what we know about typical patient behavior. Certainly patients have reason to be less than candid with their doctors about aspects of their behavior—what they eat; whether they indulge in the consumption of alcohol, tobacco, and drugs; whether they engage in risky sexual practices; how often they exercise—because they wish to evade potential disapproval of their lifestyle choices. Again, the assumptions behind the rules of evidence appear outdated and naïve upon examination. The presumptions underlying the rules of evidence have never been subject to empirical testing by behavioral scientists. Even on non-scientific, common sense reflection, they seem implausible as a description of actual human motivation and behavior. Despite the strained plausibility of the justifications for these evidentiary rules, the situations described nevertheless lead to admissible testimony on the basis of unexamined beliefs encoded in law about how the ‘reasonable man’ behaves. In other words, the evidence rules acknowledge that the ‘reasonable man’ might lie at times in his life when it is in his interest to do so, but never when he is on the point of judgment day. Likewise, the ‘reasonable man’ presumably never evades the truth with his doctor, even when it would expose embarrassing or unflattering aspects of his lifestyle to medical disapproval. A closer examination of two other evidence rules based on similar unexamined assumptions about behavior and language exposes the implausibility of the behavioral assumptions upon which they are grounded. In the first example considered here, the rules of evidence construe a person’s silence in the face of accusation as admissible evidence of admission of wrong-doing, presuming that the ‘reasonable man’ would not remain silent under such circumstances. In the second example analyzed in this paper, the rules of evidence construe any apologetic language used by a person to constitute admissible evidence of an admission to legal responsibility for the situation referred to by the so-called apology. In each case, the law assumes that the linguistic behavior which it constructs as that of the ‘reasonable man’ is an adequate and appropriate basis for considering evidence to be both reliable and probative, and hence admissible in court. As in other areas of the common law, evidence rules are predicated on beliefs about the behavior—in this case, the linguistic behavior—of what the law traditionally called the ‘reasonable man.’ As a general matter, the construct of the ‘reasonable man’ sets the standard in law for legally privileged behavior and activity in the application of substantive legal doctrines, where acting as the law assumes the ‘reasonable man’ would do insulates the actor against both civil and criminal liability. -22-

That is, those who behave ‘reasonably’ under the law’s definition will be shielded from legal liability, whereas those who fail to live up to the ‘reasonable man’ standard will find the full force of the criminal and civil law imposed against them. The power of the ‘reasonable man’ construct in law extends beyond the substantive law of torts and criminal law, however, and extends its reach into the procedure realm of evidence law as well. The ‘reasonable man’ has long dominated the common law legal imagination (see, for example, Vaughn v. Menlove, 1837). As described within legal opinions, the ‘reasonable man’ is the embodiment of middle class values, sensibilities, and practices—in the words of one judge, ‘the man who takes the magazines at home and in the evening pushes the lawnmower in his shirt sleeves.’ (Hall v. Brooklands Auto Racing Club, 1933: 224). In recent years, the law’s ‘reasonable man’ has morphed into the ‘reasonable person’ in an attempt to render the standard more universal and less particularly masculine. Legal scholars have expressed skepticism as to whether this terminology change has effectuated any substantive changes in the behavior expected by the law of this newly gender-neutral ‘reasonable person.’ (Bender, 1988: 22; Cahn, 1992: 1405). Rather, the use of this gender-neutral standard may well serve only to impose a superficial mask of purported universality onto the unchanged behavioral norms and values incorporated in that original ‘reasonable man’ standard (Graddol & Swann, 1989: 110). For this reason, this paper will use the term ‘reasonable man’ and masculine pronouns, not in deference to the supposed ‘universality’ of the so-called ‘generic’ masculine pronoun, but instead because I believe and am arguing here that the purportedly universal ‘reasonable man’ in law actually incorporates gendered assumptions about legal actors. Note that, later in this paper, when I am discussing actual legal cases, I use the pronoun appropriate to the particular defendant in the case in question. The influence of the assumed characteristics of the ‘reasonable man’ in law can be seen in the workings of the rules of evidence at issue here. Specifically, the rules of evidence construe a person’s silence in some circumstances as though it were in effect a confession of wrong-doing. Federal Rule of Evidence 801 (d) (2) (B) provides that, if a person is confronted with an accusatory statement by someone else under circumstances in which a ‘reasonable man’ would rebut that statement, and instead the hearer remains silent, that silence is admitted as an adoptive admission of the truth of the accusation. Before silence can be taken as a confession, the accused person must have heard and understood the accusatory utterance in question, must have had the opportunity to object to it, and the context must be one in which it would be reasonable to expect that there be a response to the accusation (Wigmore, 1904: 102). Once these threshold requirements are met, silence in the face of someone’s accusation is treated by the rules of evidence as the legal equivalent to an actual confession to the charge. The presumption behind the rule is that, when faced with an accusation of wrong-doing that is untrue, the ‘reasonable man’ can have only one reasonable response—to explicitly and unequivocally deny it. This evidence rule, then, can be seen to be squarely premised on assumptions about how the ‘reasonable man’ will express himself in situated discourse. The legally constructed ‘reasonable man’ speaks directly, clearly, bluntly, and without qualification, hesitation, or mitigation (Ainsworth, 1993: 302—306, 315—317). He never shrinks from confrontation, regardless of the circumstances. Moreover, he must maintain constant vigilance in his discursive interactions in case someone might use insulting or accusatory language in his presence, vigilance needed so that he springs into action and immediately rebuts any implication of wrong-doing on his part.

-23-

When articulated in that fashion, the adoptive admission of evidentiary rule seems to be based on a caricature of discursive behavior that is unconvincing. Indeed, a handful of courts over the years have suggested that silence in the face of arguable accusation should not necessarily be treated as tantamount to confession. (See, for example, People v. Bigge, 1939; State v. Clark, 2008). Nevertheless, the vast majority of American appellate courts have endorsed these supposed adoptive admissions unreflectively. Worse, courts not infrequently apply the adoptive admission doctrine even in cases in which the foundational facts required by the rule for admissibility are highly questionable. The adoptive admissions rule makes no sense unless, at the very least, the party in question has actually heard the supposed accusation. In some cases, however, it is unclear from the evidence adduced at trial that the defendant even heard the statement in question because he may not have been in the room when the supposed accusation was uttered (Alvarado v. State, 1995) or he was in the back seat of a car when a person sitting in the front seat supposedly made an accusation in a conversation with another front seat passenger (U.S. v. Carter, 1985). In other cases, the person who was said to have used accusatory language in the defendant’s presence was unable in their courtroom testimony to recall whether or not the defendant remained silent upon hearing the purported accusatory statement, but the adoptive admission was used against the defendant nonetheless (Comm. v. Braley, 2007). The unfairness of the adoptive admissions doctrine is compounded by the fact that it is often applied in contexts in which the inferential link between the fact of silence and a presumed admission of guilt is extraordinarily unpersuasive as a matter of common sense. For example, in many cases, the supposedly accusatory language is not made in an interaction in which the accused person is a party, but instead merely apparently made within his earshot in a conversation to which he is not a party. Apparently, the law insists that the ‘reasonable man’ must monitor any conversation that happens to be within his hearing and police them for direct or indirect accusations of wrongdoing, whether he is an addressee of that conversation or not. The ‘reasonable man’ must moreover be particularly alert to ambiguous language that could potentially be interpreted as an accusation against him, especially if someone is using pronouns of ambiguous referential scope such as ‘we’ (People v. Sneed, 1995) or ‘they’ (People v. Riel, 2000) in describing someone’s actions. Statements that are ambiguous, if they have any potentially available interpretation that could be construed as possibly accusatory, are especially fraught with the peril of an inadvertent tacit admission of wrong-doing. According to the adoptive admissions rule, innocent sounding remarks with multiple potential meanings that might conceivably have an incriminating implication must be objected to or else risk an admission to the incriminating interpretation. For example, in a conversational exchange between two corporate officers, the statement by one of them that ‘I’ll see if we can get anything for this work,’ was construed to be an agreement that an illegal bid rigging be arranged. Despite the fact that this statement is entirely consistent with an attempt to procure business in a lawful manner, the defendant’s failure to anticipate and object to an interpretation of his statement that implied unlawful bid-rigging was considered by the court as an admission that he was aware of the illegal nature of the bid (U.S. v. Basic Construction Co., 1983). Even a failure to object to innocuous, non-accusatory statements can result in adoptive admissions when, after the conversation has concluded, those statements— innocent in the context in which they were made—turn out to be inculpatory. For example, when a stranger referred to the defendant as ‘John’ to another without -24-

objection by the defendant, the court held his lack of protest to be an admission that the defendant went by that name (State v. Wallingford, 2001). In another case, the defendant’s failure to object to another person’s claim that she usually carried a gun was construed as an admission that, not only did she usually carry a gun but also that she was in possession of a gun at the time of a crime that occurred long after the conversation in question. Apparently, according to this court’s analysis, she should have foreseen that failure to object to this characterization of her habitual behavior would constitute an ‘admission’ of a fact concerning a crime that had not even occurred at the time (State v. Browning, 1997). In other words, the law’s imagined ‘reasonable man’ is supposed to be alert to the possibility that inaccurate statements made in his presence—however inconsequential they might appear at the time—could turn out to incriminate him for crimes that might occur at some time in the future. According to this legal doctrine, a ‘reasonable man’ doesn’t simply ignore name-calling, but instead must object to it or be held to have agreed with the characterization. In one such case, when the defendant was called a ‘butcher,’ his failure to argue with the name caller was later held by the court to be an admission of that fact (State v. Gorrell, 1996). Similarly, ducking a rude, point-blank question about impropriety can result in the court construing a refusal to engage with the questioner as an adoptive admission of guilt, as when a corporate executive, asked by a reporter at a press conference if the corporation had been ‘cooking the books,’ responded by saying, ‘Next question, please.’ Although one plausible interpretation of this response is that the executive merely had no intention of engaging the reporter hounding him, the court instead found that his failure to deny the allegation constituted an admission on his part. In the court’s view, the ‘reasonable man’ in such a situation would have directly rebutted the reporter’s claim without attempting to avoid it (U.S. v. Henke, 2000). Even reacting angrily to an accusation may not be sufficient to avoid the adoptive admission trap, as when a frustrated and angry defendant told an accuser to ‘shut the f--- up.’ This angry outburst was insufficiently responsive to the accusation, in the opinion of the reviewing court, and actually constituted an admission that the accusation was correct (State v. Gilmore, 1999). At times, even an explicit denial of an accusation may not be adequate to avoid an imputed adoptive admission if the defendant fails to repeat the denial each time the accusation is made. For example, in one case the defendant was asked if he had committed the crime, and he explicitly denied it. The questioner responded to this denial by telling him that a third party thinks that he did it. Instead of repeating the denial anew, the defendant in this case reacted by turning his head and staring out the window. The court could easily have characterized this response as shock or dismay that someone could think that he had committed the crime, but instead construed it as an adoptive admission because he failed to repeat the denial that he had made just seconds before (State v. Gomez, 2004). Cases involving recorded telephone conversations pose special problems for appropriate implicated meaning because each conversant’s inability to see the other person’s facial expressions or body language can give rise to implicature that is not intended, including a false impression of acquiescence to an accusation. For instance, in one such case, a defendant’s silence during a recorded jailhouse telephone conversation while a friend read to him a newspaper account of the crime was admitted as an adoptive admission of the facts as set out in the news story, on the theory that the ‘reasonable man’ would have objected during the recitation of the newspaper’s version of the crime (U.S. v. Higgs, 2003). In another case involving a recorded jailhouse conversation, the defendant was told by the other party to the -25-

conversation that the police had asked that party ‘twelve times’ whether he had seen the defendant shoot the victims, to which the defendant replied, ‘Oh man, twelve times.’ From this response, the court concluded that the defendant had admitted to being the shooter because his response did not include a specific denial of that fact. Of course, there was no reason for him have objected in this conversation to the police theory that he was the shooter because the other party gave no indication that he shared that police belief. In fact, had the defendant done so, it would have been a nonresponsive non sequitur in the context of a conversation whose actual topic was the persistence of police questioning of the other party, not the accuracy of the police hypothesis of the defendant’s guilt. Apparently, the law’s imagined ‘reasonable man’ would express neither shock in the persistence of a false police theory of the case, nor empathy for the police grilling that his friend had undergone, but would instead respond only by explicitly informing the other party that the police were wrong in their suspicions (People v. Davis, 2005). Merely being quietly present while someone else plans a crime or boasts about past crimes can result in adoptive admissions that one is or was involved in those crimes. In one such case, the defendant nodded his head while someone at a party bragged about having committed crimes. Although the braggart’s version of the crimes at no point claimed that the defendant participated in those crimes, the reviewing court found that sitting silently and nodding his head constituted an adoptive admission by the defendant that he, too, had participated in some unspecified way in the braggart’s crime spree (U.S. v. Price, 2008). In another case, someone that the prosecution characterized as a racketeering kingpin ordered a third party to commit arson while the kingpin was present in the defendant’s home. The defendant did not speak up to object to the kingpin’s command to that third party. Remaining silent here was construed to be an admission by the defendant of his own culpability in the arson plot, said the court, since ‘an innocent man would not let others sit in his house and plan arson.’ (U.S. v. Manzella, 1986: 545). Apparently, the law’s imagined ‘reasonable man’ doesn’t fear the physical consequences of confronting powerful criminals who could potentially do him serious harm. In another similar case, a defendant who testified that he could not object because of fear to inaccurate inculpatory statements made by a Mexican Mafia boss in his presence nevertheless was saddled with those statements as adoptive admissions (Paredes v. State, 2004). Power imbalances making objection realistically unavailable are rendered invisible in cases such as these. Disparities in power between participants in a discursive interaction may inhibit the possibility of a denial of accusatory statements in other contexts as well. In one case, in the course of a court hearing addressing a guilty plea agreement, the judge in that case made statements on the record characterizing the facts of the crime to which the plea was to be entered. In a later matter, that defendant objected to those judicial statements as inaccurate and misleading. He was, however, held to have adopted the misleading details stated by the plea judge because he had not interrupted the plea judge during the plea hearing to object to the judge’s characterization of the crime. Apparently, the ‘reasonable man’ does not fear jeopardizing a plea bargain and risking a higher sentence by interrupting and contradicting the judge in his description of the offense for which the defendant hopes to get the benefit of his plea bargain (U.S. v. Miller, 2007). As cases like this show, the adoptive admission rule of evidence maintains that a ‘reasonable man’ not only confronts accusations made directly to him but must also be alert for any statements that he might overhear made to anyone else that might -26-

impugn him, whether explicitly or impliedly, with regard to supposed bad acts that have occurred in the past or that might occur in the future. Unless he defends himself with an categorical denial in any of those situations, the law of evidence will treat his silence as the legal equivalent of a confession. Because it is conceptualized as the equivalent of an express confession to the wrong-doing, the adoptive admission is deemed to be sufficient to sustain a conviction even where there is little or no other substantive evidence of guilt of the crime in question (see e.g. Commonwealth v. Braley, 2007; State v. Tolliver, 2001). The principle behind the adoptive admission rule of evidence can become a particularly insidious trap for anyone who is being investigated by the police on a potential criminal charge. Under the nearly fifty-year-old Miranda rule governing police interrogation in the United States, a person who is arrested must be informed by the police of his constitutional rights to remain silent and to have the assistance of counsel during questioning upon request. He will be reminded that if he chooses to speak, anything he does say can be used against him later in court (Miranda v. Arizona, 1966).The Supreme Court has acknowledged that, having just expressly informed a suspect that he need not respond during interrogation, it would be unfair to then turn around and use silence by the suspect upon being confronted with police accusations as evidence suggesting guilt.Taking a suspect’s exercise of the right to silence as an adoptive admission and proof of guilt would completely gut the Miranda rule, turning attempts to exercise the right to remain silent into the functional equivalent of confessions. Therefore, the Court has held that, once given Miranda warnings, an arrestee’s mere failure to respond to accusatory language by the police cannot be used as an adoptive admission (Wainwright v. Greenfield, 1986). Since the adoptive admissions rule by its own terms requires that the context of the interchange be one in which it would be natural for the accused to affirmatively deny the accusation, it might appear that the adoptive admissions rule could have no application at any point in a Miranda-governed interrogation system, with its guarantees of the right to remain silent. Yet, surprisingly, in a number of American jurisdictions, this is not the case, and a failure to respond to police accusations in certain circumstances runs the risk of inadvertently admitting those allegations by silence. The Miranda warnings, as the Supreme Court has recently reaffirmed, are part of the cultural fabric of the country (Dickerson v. U.S., 2000: 443). Reading the Miranda rights to the just-arrested culprit is a stock scene in police dramas on television and in the movies. As a result, there is hardly an American alive who could not recite the famous warning by heart. With that in mind, consider the common-place situation in which someone is confronted by the police but not read the Miranda rights because the police have not yet made the decision to make an arrest. (Police are free to interrogate suspects without providing Miranda warnings until the point of arrest.) In a situation like this, the suspect might well be aware from his stock of cultural knowledge that he has the right to remain silent, even though the Miranda warnings have yet to be formally given him. If this suspect has had previous police encounters in which he had the Miranda rights read on those earlier occasions, he might be particularly conscious of his right to refuse to answer police questions and to refuse to respond to their accusations. In a pre-arrest context like this, the suspect who stays mute in the face of police accusations could be motivated by his realization that anything he says could be used against him, so that the wisest course may well be to say nothing at all.

-27-

However reasonable this reasoning might be to a suspect, the Supreme Court has nevertheless permitted the prosecution to use evidence of pre-arrest, unwarned silence on the part of a suspect to be used to impeach his credibility if he later takes the stand at trial (Jenkins v. Anderson, 1980). Even post-arrest silence can be used for impeachment as long as the record does not show that Miranda warnings were actually given. Despite the fact that most people already understand that they have the right to remain silent in the face of police accusations, citizens are not justified in relying on the right to remain silent, says the Court, unless they can establish that the Miranda warnings had actually been read to them by the time of the police confrontation at issue (Fletcher v. Weir, 1982). Failing that, silence in the face of police accusations can be used as impeachment; that is, to attack the defendant’s credibility if he takes the stand to testify in his defense. The prosecutor will be permitted to argue that an innocent person would not have refused to answer police questioning and further, that the jury should now disbelieve his testimony in court because he failed to tell his story to the police during their investigation. This line of argument would be clearly unlawful if the defendant had been arrested and Mirandized prior to his refusal to answer police questions, and is only allowed in a pre-arrest interrogation because of the fiction that persons only become aware of their right to remain silent upon the reading of Miranda rights. One question that the Supreme Court has not yet decided is whether a person’s silence in the face of police questioning can be used not merely to impeach—to suggest that his trial testimony should not be believed—but as an adoptive admission that would constitute substantive evidence of guilt. Since adoptive admissions are treated in law as if they were explicit confessions, can silence in the face of police accusations be treated as though it were a confession to the truth of those allegations? If that were the case, then making the mistake of asserting one’s right to silence too soon—before the formal reading of Miranda rights—could be a disastrous choice for the unwary. Believing that remaining silent is the safest choice when confronted with police accusations of wrong-doing, the person opting not to cooperate would inadvertently be adopting the truth of those accusations and in effect unintentionally confessing to them. In the absence of controlling Supreme Court case law, lower federal appellate courts have been divided in answering the question of whether silence in the face of police accusations can be used as substantive evidence of guilt via the adoptive admission doctrine. The Fourth, Fifth, Eighth and Eleventh Circuits have all ruled that silence in the face of police accusations by arrestees who have not yet been read Miranda warnings can be used by prosecutors as substantive evidence of guilt; the Seventh, Ninth, and DC Circuits, on the other hand, have held that it is constitutionally impermissible to use that silence as affirmative proof of guilt. A similar split in the circuit courts exists on the related question with respect to police accusations made prior to any arrest, with the First, Sixth, Seventh, and Tenth Circuits determining that silence in the face of police accusations cannot be used as substantive evidence of guilt, whereas the Fifth, Ninth, and Eleventh Circuits have concluded that before arrest, a person who refuses to respond to police accusations does so at peril of adopting any allegations by silence (Skrapka, 2006: 374—387; Ryan, 2007: 908—913). Thus, despite the knowledge of virtually all Americans that they have the right to remain silent when confronted by the police, exercising that right in many American jurisdictions leads to the possibility that such silence will be construed as a confession—a possibility almost certainly not anticipated by the person

-28-

who naively believes that Americans have the constitutional right to remain silent when questioned by the police. As this analysis shows, the rules of evidence can have the effect of turning a person’s silence into ‘speech’ by presuming that the ‘reasonable man’ would speak to rebut a direct or even inferential accusation of wrongdoing if it were untrue, such that it is fair to construe silence—the failure to respond—as though it were actually a confession to the accusation. In essence, because the law’s imagined ‘reasonable man’ would speak to confront an accusation, all citizens have the legal duty to speak, enforced by the law’s construal of silence as the equivalent of an admission of criminal guilt or civil liability. In other situations, however, the ‘reasonable man’ is constructed under the law as someone who would not speak, and the person who instead chooses to speak is penalized under the evidence rules. One such case occurs when a person says something that sounds as though it might be an apology. Evidence law construes utterances such as ‘I’m sorry’ as admissions that the speaker is at fault for the negative circumstance being commented upon. As such, ‘I’m sorry’ becomes admissible against the speaker as a declaration against interest under Federal Rule of Evidence 804 (b) (3), and can be interpreted as a confession of fault. In other words, the law’s imagined ‘reasonable man’ never says ‘I’m sorry’ unless he intends to admit being at fault. Evidence law’s interpretation of apologetic language as an admission of wrongdoing betrays the law’s impoverished conceptualization of the pragmatics of such language. ‘I’m sorry’ is language that can bear a multitude of meanings, depending on context (Blum-Kulka, House & Kaspar, 1989; Fraser, 1981; Lakoff, 2003). In some circumstances, ‘I’m sorry’—or language like it—can of course be fairly construed as a confession of wrongdoing, coupled with an expression of remorse on the part of the speaker for the transgression. This confession-plus-remorse interpretation is the construction that the law’s ‘reasonable man’ invariably gives to apologetic language. However, confession-plus-remorse is by no means the only interpretation that can be given to apologetic language. Saying ‘I’m sorry’ can, in some contexts, be an expression of empathy completely devoid of any connotation of fault for the negative situation that provoked the expression. For example, when someone says ‘I’m so sorry’ upon hearing of another’s recent bereavement, there is no question that ‘I’m sorry’ is not meant to be an admission of responsibility for the death. Instead, it serves to build and reinforce interpersonal relational ties by an empathetic acknowledgement of the pain suffered by the addressee as a result of the bereavement. ‘I’m sorry’ can, in other contexts, be interpreted as an expression of nonempathetic regret, again without any implication of responsibility for wrongdoing on the part of the speaker, as when the speaker tells a former lover, ‘I’m sorry I ever met you.’ In yet another context, someone might say ‘I’m sorry’ even though the speaker does not truly acknowledge fault in an attempt to defuse a hostile or violent situation. In fact, in a dangerously violent confrontation, a weaker party might have little recourse other than to desperately attempt to placate the stronger party by the use of apologetic language, regardless of whether the apologizer was actually assuming responsibility for the negative situation. Even in contexts unmarked by marked power disparity, saying ‘I’m sorry’ can be an effective discursive strategy to short-circuit an argument when a more substantive response would only serve to prolong an unpleasant interchange. In that kind of situation, ‘I’m sorry’ serves as a useful

-29-

linguistic resource to smooth over interpersonal disputes without necessarily accepting full responsibility for being in the wrong. Despite the pragmatic richness of apologetic language and its myriad of interpretable meanings, the law of evidence always permits the construal of language such as ‘I’m sorry’ to be an admission that the speaker is at fault and thus potentially legally liable. The ‘reasonable man’ is careful to avoid apologetic language unless he intends to admit that he is at fault. He cannot contemplate the more nuanced social uses of apology because the law’s ‘reasonable man’ assumes that language is nothing more than a referential vehicle of propositional communication, and thus he ignores its use as a mechanism for maintaining and repairing social relationships (Ainsworth, 2008: 14—15). Apologies are avoided whenever possible by the legally imagined ‘reasonable man’ not only because they expose him to legal liability but also because apologies are seen as language of weakness. Apology puts the speaker in a vulnerable position with respect to the person apologized to. The admission of fault inherent in apology puts the apologizer in a one-down position through that very acknowledgement. Worse yet, from the point of view of the apologizer, is the possibility that the addressee might rebuff the apology and choose not to accept the offered expression of regret. Because the addressee of an apology has that unilateral power to reject an apology, those who might potentially apologize have a keen appreciation that their act of apology puts them at risk of serious face-threat (Goffman, 1967: 5—45). When the person apologized to fails to accept the apology unconditionally and instead delivers a reproach to the apologizer, the apologizing party may feel forced to adopt a defensive, self-justifying response in order to mitigate the face threat inherent in that reproach (Hodgins & Liebeskind, 2003). Not surprisingly, then, powerful persons may opt to avoid the face-threat inherent in apology in the first place. Those who are powerful will tend not to apologize even when in the wrong; those who lack power may end up having to apologize despite not being in the wrong. Here again, it is the presumptively ‘normal’ linguistic behavior of the ‘reasonable man’ that leads the rules of evidence to its default presumption that apologetic language be construed as admission of fault. The ‘reasonable man’—seen as self-regarding, autonomous, and individualistic—does not need to use language like ‘I’m sorry’ to maintain or repair his relationships or to express emotional solidarity with others for their misfortunes and tragedies. Nor is he concerned with placating those who may be more powerful than he is, or finding discursive methods to resolve disputes and reduce interpersonal tensions. The only use of apologetic language for the law’s ‘reasonable man’ is—when absolutely necessary—to admit wrong-doing. This cramped and impoverished conceptualization of the social meanings potentially available for apologetic language in the law results in insurance companies advising their policy holders never to say ‘I’m sorry’ when they are in an accident (Robbenolt, 2005: 1012). In fact, according to the language found in some insurance contracts, insurers actually have the right to void coverage and refuse to pay policy holders who use such apologetic language after an accident (Bartels, 2000: 153). That the law treats apologetic language as admission of fault giving rise to potential legal liability has other unfortunate consequences. For example, doctors have long understood the legal danger that apologetic language can present in cases of bad patient outcomes. Apologetic language is, of course, pragmatically ambiguous in such cases. It might be a true acknowledgement that the doctor has acted negligently and caused unnecessary harm to the patient. On the other hand, it also could quite -30-

plausibly be an act of emotional solidarity with a fellow human being—an empathetic expression recognizing the suffering of the patient, even if that suffering has come into being without any fault at all on the part of the doctor. Under the rules of evidence, however, apologetic expressions by medical personnel made to patients or their families—whether intended as merely sympathetic or as actually acknowledging some degree of responsibility for the harm—could be used as evidence against them in malpractice litigation. As a result, doctors and other medical workers have long been cautioned to avoid any kind of empathetic language in their communications with afflicted patients and their loved ones and, above all, to refuse to admit the possibility of mistake or misjudgment on their part (Taft, 2005a: 58). The traditional avoidance of apologetic language by doctors and other medical workers is not a costless discursive strategy, however. Losing the opportunity to connect on a human level with patients through expressions of empathetic solidarity makes the doctor-patient relationship more bureaucratic—less fully human—than it would otherwise be. Nor does it even turn out to fulfill its ostensible purpose of avoiding legal liability. Partly as a result of perceived physician stonewalling, those who suffer unexpectedly poor medical outcomes often file lawsuits out of frustration in an attempt to find answers to what happened and why. Those who suspect that they or their loved ones may be victims of medical mistakes are often motivated as much by a desire to know the truth behind their tragedies as by the pursuit of monetary compensation for the harms themselves (Robbenolt, 2005: 1015—1017; Taft, 2005a: 77—78). Ironically, doctors’ silence in the face of bad patient outcomes—a strategy designed to reduce the possibility of liability through avoiding admissions of fault— actually has the unintended consequence of promoting lawsuits. Because of this, there is currently a growing trend to create an exception to the rule that apologies are admissible as admissions of fault in the context of medical malpractice cases. More than half of American state legislatures have in the last few years passed special exceptions to the evidentiary rule construing apologetic language as admissions of fault when the apologetic language was used by medical personnel faced with bad patient outcomes (Todres, 2006: 686). In such states, doctors can freely use apologetic language with their patients without worrying that their words will come back to haunt them in any potential future trial. This special exception to the application of the evidence rule in the context of medical malpractice cases has indeed appeared to have its intended result of reducing overall doctor liability. In the words of one of the doctors who has been a major proponent of legislation insulating medical apologies from legal consequences, ‘Nothing is more effective in reducing liability.’ (Taft, 2005a: 63). Based on the success of the legislative move towards preventing doctor apologies from being used as evidence of fault in court, one might ask whether this represents a more general trend towards breaking the legal link between apology and liability. Might it even mean that the law’s ‘reasonable man’, with his abhorrence of apology, is losing his grip over the rules of evidence? This seems unlikely, upon further examination. The laws barring the use in malpractice cases of arguably apologetic language used by doctors has not led to a more general questioning of the proposition that apologies should be interpreted as admissions of fault. Rather, the legislative move to insulate doctors’ apologies from use at trial has been promoted as part of a package of laws designed to make it harder for patients to sue their doctors— including shortening the time limits for filing suits, restricting compensation for lawyers who represent injured patients, and capping compensable damages for pain and suffering regardless of the severity of the injury (Todres, 2006: 693—696). Thus, -31-

the doctors’ apologies being protected under these new laws are not being excluded from use at trial because of the law’s recognition that apologies are pragmatically variable in their implied meaning. Nor are they being enacted out of a desire to promote empathy, or maintain human relationships, or mitigate power imbalances. Instead, these laws are being promoted as an instrumental means to reduce the frequency of medical malpractice litigation, to limit patients’ ability to get compensation for their harms, and to free doctors from the consequences of admitting mistakes. Tellingly, in some states, not only are empathetic apologies excluded from evidence, but even specific admissions of acts of malpractice by medical personnel are also barred from use at trial (Taft, 2005b: 602). Apologetic language of this latter sort—articulating factual responsibility for harm caused, acknowledging fault, and expressing regret for the harm—is much closer to the kind of canonical apology that unambiguously confesses responsibility for the unjust suffering of another (Smith, 2008: 140—142). Those sorts of apologies are exactly the kind of apologetic expression that legitimately ought to give rise to legal consequences. Yet, in its enthusiasm for reducing medical malpractice liability, the law now refuses to give legal effect to precisely the kinds of apologies that linguistic analysis would confirm actually does support an implicature of admission of fault. In both the evidentiary rule on adoptive admissions and the rule construing apologies as admissions, legal liability is created through the use of normative presumptions about the linguistic practices of the ‘reasonable man’—when he would speak and when he ought to instead keep silent. Despite the lack of empirical support for these normative presumptions, evidence law persists in enforcing these ‘reasonable man’ based standards even in factual circumstances that raise serious doubt about their applicability—privileging discursive practices that accord with the ‘reasonable man’ construct and penalizing those that do not. In that way, evidence law both constructs and mediates its own discursive paradigms through its role as gatekeeper of admissible evidence in trials. As part of larger normative and ideological systems, evidence law, like law more generally, both reflects existing power relations and serves to maintain, reinforce, and amplify those hierarchies. References Ainsworth, J. E. (1993) In a different register: the pragmatics of powerlessness in police interrogation. Yale Law Journal 103, 259—322. Ainsworth, J. E. (2008) ‘You have the right to remain silent’…but only if you ask for it just so: the role of linguistic ideology in American police interrogation law. International Journal of Speech, Language and the Law 15: 1—21. Bartels, W. K. (2000) The stormy seas of apologies: California provides a safe harbor for apologies made after accidents. Western State University Law Review 28: 141—157. Bender, L. (1988) A lawyer’s primer on feminist theory and tort law. Journal of Legal Education 38: 3—37. Blum-Kulka, S., House, J., and Kaspar, G. (eds) (1989) Cross-cultural Pragmatics: Requests and Apologies. Norwood, NJ: Ablex. -32-

Bucholtz, M. (1999) Bad examples: transgression and progress in language and gender. In M. Bucholtz, A.C. Liang, and L. A. Sutton (eds) Reinventing Identities: The Gendered Self in Discourse. New York: Oxford University Press, 3—24. Cahn, N. R. (1992) The looseness of legal language: the reasonable woman standard in theory and in practice. Cornell Law Review 77: 1389—1446. Fraser, B. (1981) On apologizing. In F. Coulmas (ed.) Conversation Routine: Explorations in Standardized Communication Situations. Berlin: Mouton de Gruyter, 259—71. Goffman, E. (1967) On face-work: an analysis of ritual elements in social interaction. In Goffman, E. Interaction Ritual. New York: Anchor Books, 5—45. Graddol, D. and Swann, J. (1989) Gender Voices. Cambridge, MA: Blackwell. Hodgins, H. S. and Liebeskind, E. (2003) Apology versus defense: antecedents and consequences. Journal of Experimental Social Psychology 39: 297—316. Lakoff, R. T. (2003) Nine ways of looking at apology: the necessity for interdisciplinary theory and method in discourse analysis. In D. Schiffrin, D. Tannen, and H. Hamilton (eds) The Handbook of Discourse Analysis. Malden, MA: Blackwell, 199—214. Robbenolt, J. K. (2005) What we know and don’t know about the role of apologies in resolving health care disputes. Georgia State Law Review 21: 1009—1027. Ryan, M. E. (2007) Do you have a right to remain silent?: the substantive use of preMiranda silence. Alabama Law Review 58: 903—918. Schlag, P. (1991) Normativity and the politics of form. University of Pennsylvania Law Review 139: 801—932. Skrapka, M. (2006) Silence should be golden: a case against the use of a defendant’s post-arrest, pre-Miranda silence as evidence of guilt. Oklahoma Law Review 59: 357—402. Smith, N. (2008) I Was Wrong: the Meaning of Apologies. Cambridge: Cambridge University Press. Taft, L. (2005a) Apology and medical mistake: opportunity or foil? Annals of Health Law 14: 55—94. Taft, L. (2005b) On bended knee (with fingers crossed). DePaul Law Review 55: 601—615.

-33-

Todres, J. (2006) Toward healing and restoration for all: reframing medical malpractice reform. Connecticut Law Review 39: 667—737. Wigmore, J. H. (1904) Evidence in Trials at Common Law. Boston: Little Brown & Co. Cases Cited Alvarado v. State, 912 S.W.2d 199 (Tex. Ct. Crim. App. 1995). Commonwealth v. Braley, 867 N.E.2d 743 (Mass. S. Jud.Ct. 2007). Dickerson v. U.S., 530 U.S. 428 (2000). Fletcher v. Weir, 455 U.S. 603 (1982). Hall v. Brooklands Auto Racing Club, 1 K.B. 205, 224 (1933). Jenkins v. Anderson, 447 U.S. 231 (1980). Miranda v. Arizona, 384 U.S. 436 (1966). Paredes v. State, 129 S.W.3d 530 (Tex. Ct. Crim. App. 2004). People v. Bigge, 285 N.W. 5 (1939). People v. Davis, 115 P.3d 417 (Ca. S. Ct. 2005). People v. Riel, 998 P.2d 969 (Cal. S. Ct. 2000). People v. Sneed, 653 N.E.2d 1349 (Ill.App. 1995). State v. Browning, 485 S.E.2d 1 (W.Va. S. Ct. 1997). State v. Clark, 175 P.3d 1006 (2008). State v. Gilmore, 22 S.W.3d 712 (Mo. App. 1999). State v. Gomez, 848 A.2d. 221 (R.I. S.Ct. 2004). State v. Gorrell, 687 A.2d 1016 (N.J. App. 1996) State v. Henke, 222 f.3D 633 (9th cir. 2000). State v. Tolliver, 765 N.E.2d 894 (Ohio App. 2001). State v. Wallingford, 43 S.W.3d 852 (Mo. App. 2001). U.S. v. Basic Construction Co., 711 F.2d 570 (4th cir. 1983). U.S. v. Carter, 760 F.2d 1560 (11th Cir. 1985). U.S. v. Henke, 222 F.3d 633 (9th cir. 2000). U.S. v. Higgs, 353 F.3d 281 (4th cir. 2003). U.S. v. Manzella, 782 F.2d 533 (5th cir. 1986). U.S. v. Miller, 478 F.3d 48 (1st cir. 2007). U.S. v. Price. 516 F.3d 597 (7th cir. 2008). Vaughn v. Menlove, 132 Eng. Rep. 490 (1837). Wainwright v. Greenfield, 474 U.S. 284 (1986).

-34-

I object, or did I? Mel Greenlee California Appellate Project, USA [email protected] Abstract A common rule of criminal procedure in the United States is the ‘contemporaneous objection rule’: In order to raise a legal error as the basis for an appeal, an objection to the error must have been raised at the trial level. For example, a defendant may not complain on appeal that evidence (such as hearsay) was erroneously presented before the jury if defense counsel did not protest to the trial judge prior to, or at the time of the admission of that evidence. In some instances, even where counsel has objected, counsel’s words (‘This is unconstitutional’) may be deemed inadequate or too vague to preserve the error for further review and this basis for appeal is therefore denied. However, like the interpretation of other alleged waivers of rights (see Ainsworth (2008) on Miranda waivers), assessment of the adequacy of an objection requires reference not only to legal rules, but also to those of pragmatics and meaning in context. An overly literal interpretation may bar review of even very serious trial errors where counsel has failed to utter the proper incantation or ‘magic words.’ This paper examines the language of trial objections and the reviewing courts’ interpretation of their meaning in a small set of California cases, showing that even in capital appeals interpretation may be counter to normal understanding of language, resulting in overly broad forfeiture of grounds for appeal. Keywords:

OBJECTIONS; APPELLATE REVIEW; LITERALISM; TRIAL

Introduction Linguistic analysis of language in American trial courtrooms has shown intricate relationships between courtroom language, trial judges’ control of the flow of evidence and information before them, and complex rules of legal procedure. A large body of linguistic and ethnographic research has now been created examining courtroom language in relation to these rules, revealing not only the crucial restrictions on who can speak about what and when, but also traditions about the type of language that parties are expected to use in the courtroom. (See, e.g., Conley & O’Barr, 1990; Levi & Walker, 1990; Tiersma, 1999) In surveying courtroom language vis-a-vis rules of legal procedure, it is important to consider that in American trial proceedings there are three audiences for language in the courtroom—the trial judge, the jury, and at a later point, the appellate court. This paper examines one procedural rule in the trial court, concerning oral trial objections, especially as these utterances are evaluated by the third audience in the appeal of convictions in criminal cases.1 I argue that the appellate court, in applying legal rules designed for efficiency, may interpret language inconsistently and in a way that is not well-aligned with normal, pragmatic understanding of language—including courtroom language. Particularly when a ‘selectively 1

Trial objections have also been examined as ‘performative’ utterances by linguists, and studies of individual trials (such as Gaines’ (2002) analysis of the OJ Simpson trial) have shown just how frequent and far-ranging objections can be during a high-stakes criminal proceeding. Performative utterances create a change in circumstance by their mere expression, e.g., ‘I hereby sentence you… .’ (Austin 1962) ‘I hereby object…’ is a formal, yet possible, expression of a trial objection.

-35-

literal’ view of courtroom objections is combined with the reviewing court’s application of ‘harmless error’ review, valid legal claims may be unnecessarily barred from appellate consideration.2 The paper briefly discusses the three audiences’ view of trial objections, focusing particularly on the appellate court. It provides case examples of the adjudication of trial objections from California criminal appeals where, in my view, the appellate court has elevated ‘form over substance’ in its review of the objection and its pragmatic effect.3 Objections 101 How do objections work in criminal trials? U.S. courtroom television shows are now so pervasive that even readers in distant nations with different legal systems have heard instances of trial objections from both fictional and ‘reality’ sources. Television dramas such as Law and Order and genuine televised trials, such as O.J. Simpson’s of 1994-5, have been viewed around the world, providing numerous examples of the objection rule in action. One application of the contemporaneous objection rule pertains to the questioning of witnesses. When a witness is being examined or cross-examined on the witness stand, either party may object if the opponent’s question of the witness calls for inadmissible evidence. A classic fictional example is the oft-repeated objection of Hamilton Burger, the prosecutor in Perry Mason, a popular 1950s American television program. Example 1: Hamilton Burger Objects4 In this episode, the defense attorney (Perry Mason) poses a cross-examination question to the witness, an attractive young lady in gloves and a hat: Def.: Where were you at 12:30 on the afternoon of the murder? Pros: Objection, Your Honor. This is incompetent, irrelevant and immaterial. It’s also improper cross-examination. The prosecutor’s objection is a general one on relevance grounds, followed by an objection concerning the scope of cross-examination. The second part of his objection focuses on the restriction that unless the witness had previously addressed a topic (her whereabouts) on direct examination, questions on that topic are not allowed on cross-examination. The trial judge may either sustain the prosecutor’s objection (preventing the evidence from going before the jury), or overrule it (allowing the evidence to be presented). As here, often trial objections concern evidence, but they may also be made to other portions of the proceedings, such as the selection or excusal of jurors, the judge’s instructions to jurors, or opposing counsel’s arguments to the jury. Consider Example 2 below: Where a witness is questioned about a statement made by someone else who does not testify in the current proceedings, the witness’s report may be barred by evidentiary rules against hearsay, or the requirement that, under the Sixth Amendment to the United States Constitution, a defendant be allowed to ‘confront’ adverse witnesses. 2

Solan & Tiersma (2005: 211) have described the manner in which courts may ignore pragmatic context in making some legal determinations, particularly those concerning criminal procedure, while applying pragmatic context to interpretation of language in others; they dubbed this phenomenon ‘selective literalism.’ I am grateful to Larry Solan for pointing out this phenomenon at work in my case data on objections. 3 U.S. v. McCullah (10th Cir. 1996) 87 F.3d 1136, 1139. 4 In these case examples, ‘Def.’ is used for defense counsel, and ‘Pros.’ for the prosecutor.

-36-

Example 2: Hearsay Objection Overruled Pros: Witness: Pros: Def.: Judge: Witness:

Did you speak to Jones that evening? Yes. What did he tell you? Objection, Your Honor. The question calls for hearsay. Cite to Crawford, confrontation.5 Overruled. (to Witness:) You may answer. Jones told me Smith was the one who stabbed Williams.

Objections and Appellate Review If the trial court erroneously allowed the witness’s hearsay testimony (concerning the stabbing) by denying the objection and if the defendant was subsequently convicted in this trial, the defendant could later raise the hearsay error as part of an appeal to a higher court. The legal error raised on appeal would be that the defendant was denied the opportunity to test the veracity of Jones as a witness by confronting and cross-examining him before the jury because Jones was never in court. But in order to present such errors on appeal, the party objecting at trial generally must voice an objection in a timely fashion, reference the legal principle underlying the objection, and if necessary, take further steps (such as requesting an admonition to the jury or filing a motion for mistrial) to attempt to remove the taint from the error.6 Had the defendant’s attorney in Example 2 not objected to the question calling for hearsay evidence, on appeal, any legal error based on improper admission of that evidence would not be considered.7 In that situation, where defense counsel sat silent in the face of error, the legal claim is deemed forfeited on appeal and procedurally barred from appellate review. The rationale for this rule is judicial economy. In the ideal world (where few trial lawyers practice), the requirement of a contemporaneous objection serves judicial efficiency by correcting errors in the first instance, as they occur, without waiting for a second level review, where judges of the appellate court correct legal errors in the lower courts’ rulings. In the real world, both trial and appellate attorneys find unreasonable the degree of explicitness required in objections, and the readiness with which appellate courts deem legal errors forfeited (by lack of, or insufficient, objection). The contemporaneous objection requirement may set up a clash between good trial tactics—not alienating or boring the first audiences (the trial judge and jury)—and the explicitness of objection required by the appellate courts. Trial judges lose patience with counsel’s repeated objections that interrupt an opponent’s evidence or argument, and their annoyance may be conveyed to the jury, to counsel’s (and the defendant’s) detriment. In a capital case in which defense counsel objected that the prosecutor had misrepresented the evidence and misstated the law in argument, the 5

Counsel’s reference is to Crawford v. Washington (2004) 541 U.S. 36. For example, the general California rule concerning objections to prosecutor misconduct is: A defendant may not complain on appeal of prosecutorial misconduct unless in a timely fashion, and on the same ground, the defendant made an assignment of misconduct and requested that the jury be admonished to disregard the impropriety, but objection may be excused if it would have been futile or an admonition would not have cured the harm.People v. Hill (1998) 17 Cal. 4th 800. 7 There is a complex web of exceptions to the rule, discussion of which would exceed the scope of this paper. See, e.g., Bond, G. (2003) Procedural hurdles: Dealing with waiver, California Defender, Summer 2003: 60— 70. 6

-37-

judge discouraged defense counsel’s objections with a lengthy admonition in front of the jury which concluded, ‘Why don’t we just get on?’8 Further, judges may become annoyed if they believe that counsel is making objections without legal substance merely for dramatic effect. In order to prevent counsel from making lengthy ‘speaking objections’ in front of the jury, one capital judge fined the attorneys $100 for each instance of what he considered improper objections.9 In another case, after a witness had denied knowledge of a written list, the prosecutor went on to ask several questions presuming such knowledge, to which the defense promptly objected, citing a lack of foundation. But the judge discouraged the objections by remarking, ‘…I don’t expect objections after every question when the questions are legally sufficient and proper.’ 10 Like Woody Allen’s famous and lengthy mistrial motion in the 1971 movie Bananas (‘This trial is a travesty. It’s a travesty of a mockery of a sham…’) trial objections may bore or baffle jurors. Jurors will likely nod out or do something more interesting, such as text their friends, when subjected to lengthy and repeated objections on fine points of law. Nevertheless, for the crucial third audience, the appellate court, it is essential that trial attorneys not only make a timely objection, but also say enough, in the right way, to preserve the legal error for review.11 Let’s suppose that instead of the hearsay scenario in Example 2, the dialogue at trial was the one in Example 3: Example 3: ‘Secondary Evidence’ Objection Pros.: Call Ms. Cox to read Jones’ previously transcribed statement Def.: Your honor, I believe that Jones …should be produced to make his own statement before the jury… Let the jury see him and determine from his attitude…whether or not he is speaking the truth. This is certainly secondary evidence. Judge: Overruled. (To the Prosecutor): Proceed. Both Examples 2 and 3 deal with the same sort of hearsay evidence, and the same legal principle: Defendants have the right to confront the evidence against them, which includes the right to have witnesses cross-examined in front of the jury. In Example 2, the objection is compact, referencing in a shorthand way the same terms contained in the legal prohibitions—the ‘magic words’ hearsay (referencing the state evidentiary rule) and confrontation (referencing the constitutional principle).12 In Example 3, there are no ‘magic words’ but counsel’s lengthy objection uses the phrase ‘secondary evidence’ to refer to hearsay. In that 1946 California capital case, People v. Bob (1946) 29 Cal. 2d 32, at 324-6, the appellate court said the phrase ‘secondary evidence’ was close enough: 8

People v. Hill (1998) 17 Cal. 4th 800 (RT 2706). (‘RT’ refers to the Reporter’s Transcript on Appeal.) People v. Valencia (2008) 43 Cal. 4th 268 (RT 2434-2435). 10 People v. Seumanu (2000) SO93803 (RT 3332-3333). 11 One witty law professor ‘compare[s] trial lawyers to wizards who must recite precise magical incantations before they wave their magic wands and attempt to make evidence disappear.’ (G. Uelmen 2005) 12 The legal import of reference to key legal doctrines in trial attorneys’ objections (counsel’s uttering of the ‘magic words’) has been compared to the power of sorcerers’ incantations even in court opinions. (See, e.g., U.S. v. McCullah (10th Cir. 1996) 87 F.3d 1136, 1139 discussing the effect of ‘failure to utter the magic words… .’ [emphasis added].) A courtroom objection, in its legal ramifications, could be considered a classic illustration of ‘how to do things with words’ although objections need not be expressed in overtly performative language. (Austin 1962) 9

-38-

We do not feel inclined to deprive defendant of his right to demand that he be tried with competent evidence because of the oversight of his counsel in the midst of a difficult trial to remember that he should add the word, hearsay, to the statement of his objection. [Emphasis added.] The court went on to say something intriguing for linguists The mere fact that the objection could have been made in better form will not justify a refusal to consider it, where the intention of the defendant could not be misunderstood. [Emphasis added.]13 In the Bob case, the legal error was not forfeited but considered on appeal, and on the basis of harm from this evidentiary error, the defendant’s conviction for murder was reversed. However, how are we to interpret that key phrase—could not be misunderstood? As linguists know well, whether an utterance is understood (or capable of being misunderstood), depends crucially not only on the speaker’s words but also the hearer’s frame of reference and the speech situation.14 In analyzing whether an objection is understood, or should have been understood to reference a particular legal principle, the central issues are: 1) How clear is ‘clear enough’ in making an objection, to avoid forfeiture on appeal? 2) How explicit must the connection to the legal principle be for a reviewing court to consider the error? And 3) How consistently (given differences among reviewing judges) do appellate courts apply this procedural rule? Finally, why does anyone care? In capital litigation there are great differences in case outcomes, contributing literally to life or death decisions, depending on whether a defendant has properly followed the procedural rules.15 Had the Bob court been a stickler for ‘magic words’ and refused to consider the harm from admitting hearsay, Mr. Bob could well have been executed. That was 1946. What happens now? In modern times, as a capital defendant works his or her way up the appellate ladder from state courts to the United States Supreme Court, higher courts may be barred from considering even important lower-court errors (like the hearsay example, or other errors discussed post) if they were not raised in the right time, or in the right way, at trial. However, federal courts have determined that such bars may be imposed only if the defendant at trial has some way of predicting how they will be applied.16 13

Accord, People v. Frank (1985) 38 Cal. 3d 711, 729, fn. 3: In a capital case, ‘a technical insufficiency’ in the expression of an objection should be discounted; rather, the trial record is examined to see if a ‘miscarriage of justice has occurred.’ But see, People v. Jones (2003) 29 Cal. 4th 1229, 1255-6, questioning whether Frank is still good law and distinguishing the Frank rule from instances in which counsel failed to object, rather than having made an inartful objection. Compare US v. McCullah (10th Cir. 1996) 87 F.3d 1136, 1139 ‘U]nless we are prepared to elevate ‘form over substance,’ the defense adequately alerted the trial court to the problem based upon its allegation of ‘outrageous,’ i.e. improper, governmental conduct. This is a capital case—failure to say the ‘magic words’ should not result in the affirmance of a death sentence which might not otherwise have been imposed.’ 14 See, e.g., Solan & Tiersma (2005: ch. 10) concerning contextual analysis of alleged threats. 15 See Nancy J. King, Fred L. Cheesman II, & Brian J. Ostrom, Final Technical Report: Habeas Litigation in U.S. District Courts: An Empirical Study of Habeas Corpus Cases Filed by State Prisoners Under the Antiterrorism & Effective Death Penalty Act of 1996 45 (2007), available at http:// www.ncjrs.gov/pdffiles1/nij/grants/219559.pdf, finding that over a quarter of federal capital appeals were dismissed on procedural grounds. Cited in L. Kovarsky (2010) Death Ineligibility and Habeas Corpus. 95 Cornell L. Rev. 329, at note 2. 16 Steele v. Ayers 7/14/10 Order and Findings and Recommendations. (Ed. Cal. ICV W-03-0143 KJM) [contemporaneous objection rule inconsistently applied in state courts may not bar federal review].

-39-

If, for example, some courts refused to accept ‘secondary’ as sufficient to raise a hearsay objection, but others always did, this inconsistent application of the legal rule, and inconsistent interpretation of the objection requirement, should not constitute a bar to future relief. In what follows, more examples of the language of objections in a small sample of California cases are provided, showing that reviewing courts have become less generous and more obtuse, in construing the contextual meaning of objections than they were in 1946. Modern California Examples Table 1 presents representative objection examples from selected capital case transcripts and California court opinions. The Table provides schematic information on the legal issue in question in each case, the wording of counsel’s objection, and the California Supreme Court assessment of the objection (and the legal error it addressed). Table 1: Examples of Objections

Legal Issue 4. Improper Opening Argument by Prosecutor

Objection Def.: ‘It almost sounds like a closing argument.’

5. Excusal of Prospective Juror

Def. : ‘I am concerned about that...’

6. Prosecution Expert Recites Out-of- Court Witness’s Report ‘You’ve got a perfect example of...poisoning here....There’s no doubt in my mind what you have here.’

Def. : ‘I’m going to object. This just makes hearsay....’ ‘Your Honor, I’m going to renew the motion to strike that as hearsay.’

-40-

Ruling on Appeal Forfeited — too ambiguous. People v. Gurule (2002) 28 Cal. 4th 557, 610 Forfeited as not pursued People v. Gutierrez (2009) 45 Cal. 4th 789, 805 Constitutional error not sufficiently articulated; moreover, any error harmless.

People v. Catlin (2001) 26 Cal. 4th 81, 137-138

Legal Issue 7. Inadmissible Victim Impact Evidence

Witness 1: ‘We’re asking for the death penalty and it’s not out of revenge.... An eye for an eye, tooth for a tooth.’

Objection [Def: Prior objections to other portions of victim impact testimony] Def. : ‘I would like to take something up outside of the presence of the jury.’ ‘We’ll take it up after the lunch recess.’

J:

[Witness 2 testifies similarly]

Def. : ‘Your Honor…I make the same objection…It goes beyond the victim impact.’ [after lunch] Def. : ‘At this time, the defense would make a motion for mistrial,…in that the testimony…is way beyond that which is permitted by the Supreme Court and to victim impact testimony.’ [Emphasis added.]

8. Improper Closing Argument by Prosecutor ‘[Defendant] has had two competent attorneys who were representing him. They were entitled to present any witnesses they wanted over the course of the last month.’ [Emphasis added.]

Ruling on Appeal

Def.:

‘Objection, Your Honor. Improper argument.’

J:

‘Overruled.’ [Later]

Def.: ‘Your Honor, I’d be making a motion for mistrial, -41-

Not specific enough; related errors forfeited because not raised until mistrial motion. People v. Cowan (2010) 50 Cal. 4th 401, 486

Legal Issue

Objection based on the statement made by [Prosecutor]…in his…argument that the defense had the opportunity to put on every witness that they wanted to…¶This court excluded several witnesses and several areas of testimony…on evidentiary grounds. What Mr. [Prosecutor] told the jury, was wrong, was false.’

Ruling on Appeal

Appellate Court: [holding there was no error] ‘Reasonably implicit in the [prosecutor’s] statement was that the defense was entitled to present any witnesses it wanted consistent with the rules of evidence.’ [Emphasis added.] People v. Valencia (2008) 43 Cal. 4th 268, 284

In the first two examples in the table (Examples 4 and 5), counsel’s objections seem properly evaluated as too vague. The hedged comment on opposing counsel’s opening argument (‘It almost sounds like a closing argument’) and counsel’s expression of ‘concern’ in protesting the excusal of a prospective juror seem more like comments or offhand remarks than explicit legal objections. Example 4 appears to have been intended as an objection to the prosecutor’s opening argument (which referred to facts ruled inadmissible), but this is not clear. Example 5 expresses worry, but not the legal ground for it. These two vague remarks would fail as objections under the contemporaneous objection requirement. Using Bob’s language, they do seem capable of ‘being misunderstood.’ Holding counsel to a more explicit rendering does not seem unduly strict. In Example 6, defense counsel’s objections were much like those in Examples 2 and 3. In all three instances, counsel protested admission of hearsay. Counsel’s objections in each were based on the second-hand presentation of an out-of-court report, where the defendant had no opportunity to cross-examine or ‘confront’ the author of the original statement. -42-

In this case, the out-of-court statement concerned a medical opinion on the victim’s cause of death. A medical doctor, in his trial testimony, quoted another, absent doctor’s statement that expressed great certainty: ‘You've got a perfect example of …poisoning here. …There's no doubt in my mind what you have here.’ In a hearing before the testimony, defense counsel objected, ‘This just makes hearsay.’ He objected again after the witness quoted the out-of-court statement, ‘Your Honor, I'm going to renew the motion to strike that as hearsay.’ But the trial judge denied the motion, relying on the principle that experts may use hearsay as the basis for their opinion, and the court so instructed the jurors. However, the hearsay/confrontation principle remains strong: Neither an expert nor a lay witness may serve as a mere conduit for out-of-court hearsay, where the hearsay statement is considered for its truth. In this instance, had jurors considered the out-of-court expert’s opinion for its truth, they would have been invited to conclusively find that the cause of the victim’s death was poisoning, a poisoning the prosecution alleged was done intentionally by the defendant. According to both state evidentiary rules and federal constitutional principles, the witness, although qualified as an expert, should not have been allowed to quote and recite the out-of-court statement of a non-testifying declarant from the witness stand, where it would be considered for an improper purpose.17 In addition to the legal prohibition against hearsay, psychological research suggests that jurors are generally unable to separate out the supposed purposes of such evidence. Even when instructed to consider the statement only as a basis for the testifying doctor’s opinion, jurors would very likely have considered the statement as evidence of the truth of the charge against the defendant. ‘It is all but impossible for jurors to forget evidence for one purpose while remembering it for another.’ (Eichhorn 1989: 345) Particularly when the ‘secondary evidence’ of the absent doctor concerned an issue so central to the case, asking jurors to ignore an expression of such certainty would require highly improbable mental gymnastics. In its opinion on appeal, the California Supreme Court ruled that counsel’s objection served to preserve only the state law evidentiary error for review, because counsel did not utter the word confrontation in addition to his explicit hearsay objection. Thus, the constitutional error (confrontation) was deemed forfeited. Moreover, the court held that, even assuming arguendo that the constitutional error had been preserved by sufficient objection, any error in admitting the second-hand medical report would not have changed the jury’s verdict; any such error, the court held, was ‘harmless.’ As in many California capital cases, a finding of harmless error sustains the affirmance. The Court stated, ‘Defendant …contends that any such error constituted a violation of his Sixth Amendment right to confrontation… Defendant did not object on this ground below, and thus his constitutional claim has not been preserved for appeal. Further, even assuming for the purpose of discussion that the admission of [Dr. 1]’s declarations implicated defendant's rights under the confrontation clause, any error would be harmless beyond a reasonable doubt.’ (People v. Catlin (2001) 26 Cal. 4th 81, 138, fn. 14.) Clearly the court’s requirements of explicitness seem to have changed since the forgiving days of Bob. In Example 6, not only ‘hearsay’ but additional ‘magic words’ were required by the appellate court to preserve the legal issue. It is debatable whether at the time

17

‘[A] witness's on-the-record recitation of sources relied on for an expert opinion does not transform inadmissible matter into ‘independent proof’ of any fact.’ (People v. Gardeley (1996) 14 Cal.4th 605, 619, 59 Cal.Rptr.2d 356.) ‘Although experts may properly rely on hearsay in forming their opinions, they may not relate the out-of-court statements of another as independent proof of the fact.’ (Korsak v. Atlas Hotels, Inc. (1992) 2 Cal.App.4th 1516, 1524—1525.) Physicians can testify as to the basis of their opinion, but this is not intended to be a channel by which testifying physicians can place the opinion of out-of-court physicians before the trier of fact. (Whitfield v. Roth (1974) 10 Cal.3d 874, 895.)

-43-

of trial counsel could have predicted the reviewing court would require not only ‘hearsay’ but also ‘confrontation’ in order to avoid forfeiture of the constitutional issue. Example 7 presents another case in which the appellate court would appear to require clairvoyance on the part of trial counsel. Here, the objections are to penalty phase testimony. At the penalty phase of a California capital trial, the only question to be decided by the jury is whether the defendant will be sentenced to execution or to life in prison without the possibility of parole. The U.S. Supreme Court has held when the jury is faced with that decision, the prosecution may present testimony from the victim’s survivors, discussing the impact on them of the victim’s death, as a factor in aggravation of the defendant’s sentence. However, the Court has also held that under the U.S. Constitution’s Eighth Amendment such evidence may not be so inflammatory or emotionally wrenching that it influences the jury to vote for death on other than a rational basis.18 The prosecutor also may not rely on the survivors’ wishes or on religious doctrine to argue for a death sentence. 19 These principles and prohibitions inform defense counsel’s objections in Example 7. In this case, the prosecutor presented testimony from the victim’s survivors. Defense counsel had previously sought guidelines on the scope of permissible testimony and objected after the witnesses took the stand, stating that their testimony went ‘way beyond that …permitted by the Supreme Court and [sic] to victim impact testimony.’ His objections and a companion motion for mistrial, in which he noted the testimony was ‘beyond the permissible bounds of victim impact’ were both denied. One of the victim’s survivors explicitly told the jury, ‘We’re asking for the death penalty and it’s not out of revenge.’ She also referred to the Biblical doctrine of ‘An eye for an eye, tooth for a tooth.’ She made an argument for the death penalty from the witness stand based on religious doctrine and survivor preferences, which are barred from prosecutor arguments as improper bases for a death sentence—even though the potential for a death sentence on the basis of emotion is even greater when these improper themes emerge from the sympathetic and emotional testimony of family members rather than in closing arguments by the State’s representative.20 Defense counsel repeatedly objected (outside the jurors’ presence) to this victim impact testimony, and referred explicitly to the U.S. Supreme Court’s decisions. Unlike Example 6, the reviewing court found the Eighth Amendment basis had been sufficiently articulated by counsel’s reference to the victim impact jurisprudence of ‘the Supreme Court’ to preserve the issue. Nevertheless, to certain portions of the evidence, the appellate court found that counsel’s objections were not sufficient and that counsel had forfeited them for review. The rationale for their ruling was that in his objection, counsel did not explicitly mention the notion of the survivors’ ‘opinion[ ] about ... the appropriate punishment’ or the reliance on Biblical doctrine.21 Of particular interest to linguists, the court assessed the impact of the challenged victim impact testimony based on its proportionate length, rather than, as counsel argued, the 18

Payne v. Tennessee (1991) 501 U.S. 808; see, also People v. Edwards (1991) 54 Cal.3d 787, 835. Booth v. Maryland (1987) 482 U.S. 496, 503; People v. Pollock (2004) 32 Cal. 4th 1153, 1180 (survivors’ opinions ‘about the appropriate punishment’ are not within the scope of constitutionally permissible victim impact testimony). Nor are Biblical mandates. (People v. Hill (1998) 17 Cal. 4th 800, 836-7.) Arguments based on these topics may violate a defendant’s constitutional right to due process. (Payne v. Tennessee (1991) 501 U.S. 808, 831 (O’Connor, J., concurring).) 20 Other states have recognized that when the source of an exhortation for death is a victim’s survivor, the impact of the prohibited topics and comments is ‘magnified.’ For example, in Wimberly v. State (Ala.Crim.App. 1999) 759 So. 2d 568, the Alabama appellate court found survivor testimony containing ‘a strongly worded exhortation for the jury to return a death sentence’ to be clearly improper, warranting a new penalty trial. 21 Counsel’s objections were also rejected, in part on timeliness grounds. 19

-44-

nature of its content. In the majority’s view, the survivors’ remarks could have made only a minimal contribution to the verdict because ‘the challenged portion of the witness’s testimony constitutes only a few sentences out of more than 25 pages of reporter's transcript comprising the prosecution's penalty phase case-in-chief… .’ (People v. Cowan (2010) 50 Cal. 4th 401, 486-7) As in the previous example, the Court found any possible error harmless. Linguists might wonder if the court would take a similar view if the ‘few sentences’ had been, ‘I saw him do it, and he gave me the gun.’ It could reasonably be inferred that even in a lengthy trial, such a short phrase would have considerable impact. From a linguistic point of view, it seems ludicrous to base a determination of the potential impact of evidence merely on the length of the offensive passages. The substantial impact of even short references to prohibited topics is supported by empirical research in which capital jurors have been interviewed about their decision-making process. The Capital Jury Project has shown that for some jurors the penalty phase of a capital trial is viewed as a legal contest between the victim’s survivors and the defendant. 22 For those jurors a plea for death from a surviving family member would likely have a very strong effect, regardless of its brevity. As a final contrasting example, observe Example 8. The prosecutor stated in closing arguments before the jury that the defense had had ‘an opportunity to call any witnesses they wanted.’ However, as the prosecutor knew, the trial judge had ruled certain defense evidence inadmissible; that testimony had not been presented before the jury. Thus, when the prosecutor made the argument, defense counsel immediately objected that his statement (‘any witnesses they wanted’) was false. The trial judge overruled the objection. Although the appellate court held the error preserved, it rejected appellant’s contention that the prosecutor’s argument was false and thus misconduct, holding that in context there was a tacit understanding that ‘any witnesses they wanted’ meant ‘any witnesses they wanted consistent with the rules of evidence.’ (People v. Valencia (2008) 43 Cal. 4th 268, 284) Thus, to uphold the judgment of the trial court and affirm a capital conviction and sentence, in Example 8 the court read in an interpretation based on context, an interpretation that (as observed in Examples 6 and 7) it was unwilling to take when a challenge to the judgment was at issue. In Example 6, the Court was unwilling to find that hearsay was equivalent to ‘confrontation.’ Nor was a global reference to victim impact evidence specific enough to encompass the particular prohibited references (victim preferences and the Bible) which were at issue in Example 7. Yet in Example 8 ‘any witnesses they wanted’ was equivalent (and limited) to ‘any witnesses they wanted consistent with the rules of evidence.’23 That such interpretations tend to favor the prosecution was noted in Solan & Tiersma’s (2005: 211) treatment of selective literalism and in Ainsworth’s (2008) discussion of Miranda adjudications. The authors found that in construing language, reviewing courts may require much more explicit expression in instances where attention to context would favor the defendant (e.g., in determining whether Miranda rights have been invoked) than in circumstances where such an interpretation favors the prosecution (as in determining whether veiled and indirect expressions constituted criminal threats). Thus, explicit statements such as, ‘I don’t want to talk about it’ and ‘I don’t got nothing to say’ were deemed insufficiently clear to constitute invocations of the defendants’ Miranda rights (Ainsworth 2008: 9), while even a question, ‘Are all the windows insured?’ was found to be a threat (Solan & Tiersma 2005: 206). Their analyses concluded that reviewing courts tend to ‘ignore or minimize 22

Sundby (2005: 128) Some penalty jurors saw their task as ‘personally to act as the victim’s voice in the jury room and to ensure that the jury carrie[d] out its obligation to protect society’ by rendering a death verdict. 23 People v. Valencia (2008) 43 Cal. 4th at 284, concluding, ‘The jury was not misled in any significant way that prejudiced defendant.’

-45-

pragmatic information’ in construing language in instances where consideration of context would favor the defendant. The interpretation of objections in the current data shows a similar imbalance. Here courts required extremely explicit objections to challenge the judgment, while allowing the judgment to stand based on an interpretation of the prosecutor’s utterance in a pragmatic context not only considered, but even supplied, by the reviewing court. Conclusion In these examples, California reviewing courts appear to interpret the scope of objections too narrowly, requiring not only magic words but an exceedingly fine level of specificity in objection, while at the same time sweeping serious trial errors away with the broad broom of ‘harmless error.’24 While designed to favor judicial efficiency, the combined effect of these procedural rules may be to ‘demean[-] the constitutional interests at stake… .’ (Carter 1993: 115) In addition, the notion that prohibited and inflammatory testimony can be easily set aside is not concordant with what linguistic research has told us about language interpretation in context.25 When viewed in proper perspective, the impact of some of these errors may be akin to what one court described metaphorically as the effect of ‘the skunk in the jury box’ (Dunn v. United States (5th Cir. 1962) 307 F.3d 883, 886: ‘[I]f you throw a skunk into the jury box, you can’t instruct the jury not to smell it.’) That court wisely recognized that the harm from erroneously admitted testimony may be such that it can’t be eradicated from jurors’ minds and inevitably tends to infect their verdict. Thus, California objections and the appellate courts’ review of them have come a long way from the expansive approach of Bob. While defense counsel cannot claim to have clairvoyance to predict what narrow interpretations reviewing courts may impose on objections, or predict how consistently the courts may apply broad procedural bars, counsel are often advised, ‘When in doubt, object!’ 26 Trial attorneys are advised to make their objections as immediately and as explicitly as possible, on all statutory and constitutional grounds. Such actions, although not in accord with notions of economy of adjudication, are seen as necessary correctives in light of the extreme literalism of reviewing courts. If reviewing courts continue their selectively hyperliteral practices, the likely result will be more pretrial motions, more sidebars, and yet more complaints about the length and expense of capital trials, without any appreciable improvement in rational review. Overly literal and lopsided review of objections would continue to distort appellate decision-making, making the ideal world of objections—where efficiency rules and rationality prevails—and 24

Harmless error analysis, as in Example 7, may be particularly misapplied in review of penalty phase errors. (Carter 1993) 25 Fisher (2008: 18) cites Justice Cardozo in Shephard v. U.S. (1933) 290 U.S. 96, 103-4, recognizing the terrific power of ‘damning words.’ Even when jurors are instructed to consider such inflammatory material for a narrow purpose, it is unlikely that they could do so, for ‘T]he reverberating clang of accusatory words would drown out all weaker sounds.’ 26 See, e.g., Mickenberg, I. Preserving the record and making objections at trial: A win-win proposition for client and lawyer, North Carolina Defender Trial School (2005).

-46-

the real world, where much effort must be spent to achieve at best, pragmatically bizarre results—very far apart. References Ainsworth, J. (2008) ‘You have the right to remain silent...’ but only if you ask for it just so: the role of linguistic ideology in American police interrogation law. International Journal of Speech Language and the Law 15: 1—21. Austin, J. L. (1962) How to Do Things with Words. Cambridge, MA: Harvard U. Press. Bond, G. (2003) Procedural hurdles: dealing with waiver. California Defender, Summer 2003: 60—70. Carter, L. (1993) Harmless error in the penalty phase of a capital case: a doctrine misunderstood and misapplied. Georgia L. Rev. 28: 125—166. Conley, J. M. and O’Barr, W.M. (1990) Rules v. Relationships: The Ethnography of Legal Discourse. Chicago: U. Chicago Press. Eichhorn, L. (1989) Social science findings and the jury’s ability to disregard evidence under the Federal Rules of Evidence. Law & Contemporary Problems 42: 341—353. Fisher, J.L. (2008) The truth about the ‘not for truth’ exception to Crawford. The Champion Jan/Feb 18. Gaines, P. (2002) Negotiating power at the bench: informal talk in sidebar sessions. Forensic Linguistics 9(2): 213—234. King, N.J., Cheesman F.L. II, and Ostrom, B.J. Final Technical Report: Habeas Litigation in U.S. District Courts: An Empirical Study of Habeas Corpus Cases Filed by State Prisoners Under the Antiterrorism & Effective Death Penalty Act of 1996 (2007), available at http:// www.ncjrs.gov/pdffiles1/nij/grants/219559.pdf Kovarsky, L. (2010) Death ineligibility and habeas corpus. Cornell L. Rev. 95: 329—388. Levi, J. N. and Walker, A.G. (eds) (1990) Language in the Judicial Process. New York: Plenum Press. Mickenberg, I. (2005) Preserving the record and making objections at trial: a win-win proposition for client and lawyer, North Carolina Defender Trial School manuscript. Solan, L.M. and Tiersma, P. M. (2005) Speaking of Crime. Chicago: U. Chicago Press. Sundby, S. E. (2005) A Life and Death Decision: A Jury Weighs the Death Penalty. New York: Palgrave Macmillan. Tiersma, P.M. (1999) Legal Language. Chicago: U. Chicago Press.

-47-

Uelmen, G. F. (2005) Preserving federal claims: the Sevilla 'Mantra Motion.’ The Champion March, 34. Cases Booth v. Maryland (1987) 482 U.S. 496. Dunn v. U.S. (5th Cir. 1962) 307 F.2d 883. Korsak v. Atlas Hotels, Inc. (1992) 2 Cal.App.4th 1516. Payne v. Tennessee (1991) 501 U.S. 808. People v. Bob (1946) 29 Cal. 2d 321. People v. Catlin (2001) 26 Cal. 4th 81. People v. Cowan (2010) 50 Cal. 4th 401. People v. Edwards (1991) 54 Cal.3d 787 People v. Frank (1985) 38 Cal. 3d 711. People v. Gardeley (1996) 14 Cal.4th 605. People v. Gurule (2002) 28 Cal. 4th 557. People v. Hill (1998) 17 Cal. 4th 800. People v. Jones (2003) 29 Cal. 4th 1229. People v. Pollock (2004) 32 Cal. 4th 1153. People v. Seumanu (2000) SO93803. People v. Valencia (2008) 43 Cal. 4th 268. Steele v. Ayers 7/14/10 Order and Findings and Recommendations. (E.D. Cal. ICV W-030143 KJM) U.S. v. McCullah (10th Cir. 1996) 87 F.3d 1136. Whitfield v. Roth (1974) 10 Cal.3d 874. Wimberly v. State (Alabama Crim App. 1999) 759 So.2d 568.

-48-

Improving the comprehensibility of U.S. pattern jury instructions Bethany K. Dumas The University of Tennessee, Knoxville, USA [email protected] Abstract This article briefly outlines a process whereby linguists can assist organizations, particularly U.S. state bar associations, to improve the comprehensibility of pattern jury instructions. Bar associations represent the attorneys in specific U.S. jurisdictions and one of their responsibilities is the production of the pattern or uniform jury instructions that typically provide the framework for jury charges in jury trials. The process described below draws upon findings of past social science research (e.g., Charrow and Charrow, 1979) and experiences of linguists who have served on bar association committees (e.g., Tiersma, 1993; 2010), and recognizes that not all forensic linguists are members of the bar. It involves these steps: 1) discover, ideally from past legal cases, whether particular instructions under consideration or particular terms, legal or otherwise (burden of proof, captious) or syntactic patterns (use of the passive voice in subordinate clauses, multiple negatives within a sentence, sentence length) have presented problems in the instructions under consideration, 2) obtain a full copy of the instruction/s at issue; read the instruction/s, note the problem/s; identify instructions that appear to present the greatest comprehension difficulty, 3) suggest to the bar association or appropriate committee the proposed editorial changes or rewritten versions of instructions, 4) solicit feedback, and 5) on the basis of feedback, prepare a final proposal for revision. Steps are illustrated with examples from the civil pattern instructions of a U.S. state in which a forensic linguist is serving on a bar association civil instructions committee. Keywords:

COMPREHENSIBILITY; INSTRUCTIONS; JURORS; JURY; PATTERN

Most people are introduced to the notion of written genres early in life, often in literature classes where we learn to recognize the differences among such texts as short stories, essays, novels, sonnets, and free verse. Eventually we encounter and recognize many non-literary genres also—contracts, for instance—but we are not always encouraged to think of non-literary types of compositions as belonging to different genres. But many categories of institutional documents constitute special genres; they are not written in the ordinary discourse of everyday speech or even writing. And though they may make use of narrative technique occasionally, they are pretty clearly not literature. They are forms of institutional discourse. Nowhere is this clearer than in legal discourse, often described as idiosyncratic and known for its propensity to be obscure, ambiguous and complex. These characteristics derive from both the legal vocabulary and the usually quite complex sentence structure of court documents, statutes, contracts, and jury instructions. Because jury instructions must be used by persons not trained in law and therefore not familiar with legal terms and usually not accustomed to reading and understanding long, complex sentences, their language deserves careful attention. Jurors are the triers of fact in jury trials. Because they must sift through disputed accounts presented in evidence, it is crucial that they understand the nature of their tasks.

-49-

Jury instructions in all U.S. jurisdictions, mandatory in all jury trials there, exhibit a wide range of problems, many of which are fully documented in research literature. The results of the empirical research on the issue of whether jurors understood their instructions reported in Charrow and Charrow drew the attention of linguists not previously aware of the range of problems in jury instructions and opened the door to new research on the issue of comprehensibility (Dann and Logan, 1986; Schwarzer, 1990; Steele and Thornburg, 1988). This was partly because the research was reported not only in the Columbia Law Review but also in an article published in the Georgetown University Round Table Proceedings Journal (1979b) that year. Journal publication followed the Round Table conference earlier in 1979 at which Charrow and Charrow gave the presentation upon which the publications were based. Some linguists who heard the presentation and/or read the Proceedings article then read the much longer Columbia Law Review article—for some linguists probably their first law review article. Also, the plain language movement in the U.S. began in the 1970s. During the 1970s and the early 1980s, eyes were opened to a new venue in need of linguistic analysis. Prior to the pioneering empirical research into characteristics of the language of jury instructions by Charrow and Charrow, it was often assumed by lawyers and perhaps judges that the incomprehensibility of jury instructions was due primarily to the fact that they necessarily contain specialized legal terms, sometimes terms of art that must be used for legal precision—mens rea for instance—and also arcane and unusual vocabulary items—captious, for instance. But it was also generally recognized that certain linguistic constructions are intrinsically more difficult to comprehend than others. Charrow and Charrow tested three hypotheses: (1) standard jury instructions … are not well understood by the average juror, and …the same is true of instructions that are grammatically and semantically similar; (2) … certain linguistic constructions are largely responsible for this hypothesized incomprehensibility; and (3) … if the problematic linguistic constructions are appropriately altered, comprehension should dramatically improve, notwithstanding the ‘legal complexity’ of any given instruction. (1979a: 164; 1979b: 1309) A large body of subsequent research has supported their conclusion that hypothesis (1) is true: ‘standard jury instructions … are not well understood by the average juror, and …the same is true of instructions that are grammatically and semantically similar’ (see Steele and Thornburg, 1988: 87). Further, subsequent research has identified linguistic constructions that are most problematic (see Tiersma, 2006) and has offered evidence that informed rewriting dramatically improves comprehensibility (Diamond and Levi, 1996). Rewriting issues have been addressed for many years, notably by Tiersma, who has served on the California Judicial Council Advisory Committee on Criminal Jury Instructions since 2005. He has suggested that key principles include issues of context—keeping one’s audience in mind and adopting appropriate style and tone, general clarity—logical organization, concreteness, simple vocabulary and grammatical constructions, and sufficient detail—plus the use of pronouns and verbs rather than nouns (2006). Such a list can probably be found in many English composition courses at several levels. These specific principles address the need for clarity and are often missing in legal language, especially in jury instructions, which -50-

are after all addressed to jurors, and not lawyers. The principles include being careful to identify separate instructions clearly and consistently, using examples or illustrations, clarifying complex matters, and providing guidance rather than definitions and principles of law. All the principles are important because during deliberations most jurors are faced with a task they have never been asked to do before. U.S. jurors are triers of fact, as noted above, while judges are triers of law. Jurors must sort through evidence presented during trial, and then follow their instructions in order to arrive at a verdict. A typical general instruction might read, ‘If you believe A (a set of facts), you must find X (a verdict). If you believe B (a competing set of facts), you must find Y (a different verdict).’ Instructions for the reasoning process are much more complex. Pattern instructions were a step forward in some respects. They provided consistency and economy of time and effort. Their weakness is their form; they are written in dense, complex language and are often modeled upon appellate opinion language, language written by judges primarily for other judges to read. The judicial appeal of mirroring appellate language is that it is unlikely to give rise to a reversal for error. Put simply, jury instructions full of legalese are unlikely to be reversed for error by appellate courts if they use the same language used by those courts. It is true that pattern instructions have been improved in many jurisdictions over the past twenty-five years, but problems remain and they are being actively addressed in many jurisdictions. Two jurisdictions, Arizona and California, have made sweeping changes in their jury systems, including revision of instructions. Tennessee has implemented a number of changes based on the recommendations of a Jury Reform Commission (Report of the Tennessee Bar Association Commission on Jury Reform, 1999) and is continuing to consider suggestions (Cohen and Cohen, 2003). But the nature of the U.S. legal system is such that mere knowledge of the principles of composing comprehensible jury instructions does not mean that the principles will be implemented. Circumstances sometimes prevent that. To illustrate that fact, here are some comments from a U.S. Supreme Court justice recently made available to the legal community by Bryan A. Garner, who interviewed eight of our U.S. justices and transcribed the interviews for publication in volume 13 of The Scribes Journal of Legal Writing (2010). Garner is well known for his work in lexicography and law, and he is the editor in chief of all current editions of Black's Law Dictionary. In the interviews, Garner directed questions to the eight individual justices. Two of those were about legalese and writing skills: Let me ask you one last question. Legalese . . . and when I ask you about legalese, I don’t mean terms of art like habeas corpus and collateral estoppel. I mean pursuant to, in the instant case, and that sort of thing. Do you have a view on legalese? Why does it matter how well lawyers write? The answers are revealing, I think. In response to the first one, for instance, Justice Samuel A. Alito Jr. responded: Phrases like that are totally unnecessary, and they can be eliminated. What makes legal writing more difficult than other types of writing is that very often you have to use a particular form of words because it’s the legal term, because it’s the language of the statute, because it’s what was said in an opinion. And so in order to be precise, in order to avoid any impression that -51-

you’re changing the law in any way, you are stuck with reiterating these same phrases, which may be very cumbersome phrases. You can’t try to develop a synonym or some alternative language; if you do that, it’s going to introduce ambiguity into the opinion. And that makes it harder because in ordinary writing, of course, you wouldn’t do that. You would try to boil down those complex phrases or find various ways of saying the same thing. You’re limited in your ability to do that when you’re writing an opinion or a brief. (p. 181-82) Should Justice Alito tire of his present occupation, he would be very welcome in many composition classrooms, I think. In just a few sentences he both described sound principles of good writing and identified a major cause of problematic legal writing: the possibility of introducing ambiguity if one departs from the language of the statute at issue or a prior opinion. Linguists are occasionally asked to assist U.S. state bar associations—and other organizations—in improving the comprehensibility of pattern jury instructions. We can make use of the research conclusions of a number of scholars, Tiersma and others, as they have assisted with the drafting of more comprehensible jury instructions (Dumas, 2000; Dumas, 2002). Below I shall draw upon my experience in Tennessee serving on the Tennessee Bar Association Jury Reform Commission 19981999 (see Report of the Tennessee Bar Association Commission on Jury Reform) and, more recently, working with the Tennessee Judicial Conference Committee on Pattern Jury Instructions (Civil) (TPI Civil Committee). My goal in the rest of this article is to outline a process whereby linguists can assist U.S. state bar associations to improve the comprehensibility of pattern jury instructions, being careful to remain within the clear intent of relevant law (i.e., to avoid either broadening or narrowing legal content) in recognition that not all forensic linguists are members of the bar. U. S. bar associations represent the attorneys in specific U.S. jurisdictions, usually states. Typically they are responsible for administering state bar examinations, disciplining attorneys for practice violations, and related matters and also for producing the pattern jury instructions under discussion here. The process I will describe is a work in progress. I meet quarterly with members of the TPI Civil Committee. At each meeting I report on my work, hear commentary on it, and am given new homework—a particular issue to try to resolve, or a specific instruction to analyze and edit. My first TPI assignment was to examine a portion of TPI Chapter 14 (DAMAGES) and make recommendations for general revisions to improve comprehensibility. I examined a portion of Section A, Tort Damages, with particular attention to one section on Compensatory Damages, that devoted to Personal Injury. The original text is below: If, under the Court’s instructions, you find that the plaintiff is entitled to damages, then you must award plaintiff damages that will reasonably compensate the plaintiff for claimed loss or harm which has been proven by a preponderance of the evidence, provided you also find it was or will be suffered by the plaintiff and was legally caused by the act or omission [or condition] upon which you base your finding of liability. Each of these elements of damage is separate. You may not duplicate damages for any element by also including that same loss or harm in another element of damage. In determining the amount of damages, you should consider the -52-

following elements: Medical expenses. Medical expenses are the cost of medical care, services and supplies reasonably required and actually given in the treatment of the plaintiff as shown by the evidence [and the present cash value of similar services likely to be required in the future.] Loss of earning capacity. Loss of earning capacity is the value of earning capacity that has been lost in the past [and the present cash value of lost earning capacity that is likely to be lost in the future] as a result of the injury in question. It is not the loss of time or actual earnings that make up this item of damages, but the loss of the ability to earn. There may be a loss of earning capacity even though there has been no loss of earnings [and even though the plaintiff made more money after the accident than before.]The loss of the ability to earn money may include, but is not limited to, actual loss of income. In deciding what, if any, award should be made for loss of the ability to earn, you should consider any evidence of plaintiff’s earning capacity, including, among other things, the plaintiff’s health, age, character, occupation, past earnings, intelligence, skill, talents, experience and record of employment. Physical pain and mental suffering. Physical pain and suffering is reasonable compensation for any physical pain and suffering, physical and mental discomfort [p.706] suffered by the plaintiff, [and the present cash value for pain and suffering likely to be experienced in the future.] Mental suffering includes anguish, grief, shame, or worry. Permanent Injury. A permanent injury is an injury that the plaintiff must live with for the rest of the plaintiff’s life that may result in inconvenience or the loss of physical vigor. Damages for permanent injury may be awarded whether or not it causes any pain or inconvenience. Disfigurement. Disfigurement is a specific type of permanent injury that impairs a person’s beauty, symmetry or appearance. Loss of enjoyment of life: Loss of the enjoyment of life takes into account the loss of the normal enjoyments and pleasures in life in the future as well as limitations on the person’s lifestyle resulting from the injury. Pain and suffering, permanent injury, disfigurement, and loss of enjoyment of life are separate types of losses. A plaintiff is entitled to recover for these losses if the plaintiff proves by a preponderance [p. 707] of the evidence that each was caused by the defendant’s fault. No definite standard or method of calculation is prescribed by law by which to fix reasonable compensation for pain and suffering, permanent injury, disfigurement, and loss of enjoyment of life. Nor is the opinion of any witness required as to the amount of such reasonable compensation. In making an -53-

award for pain and suffering, loss of enjoyment of life, and/or permanent injury, you shall exercise your authority with calm and reasonable judgment and the damages you fix shall be just and reasonable in light of the evidence. My report included sections on identified weaknesses and specific suggestions for possible remedies, including roadmaps, sentence revision, paraphrasing, and examples. I identified these weaknesses: 1. lack of an introductory ‘roadmap,’ a list or diagram telling jurors where they are headed and by what route they can expect to arrive there; 2. use of hypothetical constructions rather than straightforward instructions (construction of ‘if ... then’ statements); 3. use of technical terms e.g., ‘present cash value’ (to improve comprehensibility, brief concrete examples might be helpful); 4. highly complex sentences (with respect to both the number and order of clauses); and 5. sentences that provide negative information before providing positive information (‘It is not the loss of time...’). With respect to the first two weaknesses, I proposed that an initial roadmap be added and that the initial ‘if ... then’ construction be rewritten: You will now be instructed about how to decide whether the plaintiff is entitled to damages and, if so, how to decide on the appropriate amount. To be entitled to damages, the plaintiff must have suffered loss or harm that has been proved by a preponderance of the evidence. Second, the loss or harm must either have been or will be suffered by the plaintiff as a result of the specific acts, omissions, or conditions that led you to decide that the defendant was liable for damages. If you decide that the plaintiff is entitled to damages, you will then need to decide on the amount. You must decide on the amount by considering the following types of loss and harm, known here as elements. These elements or types of damage must be considered and calculated separately. You must not duplicate damages for any element by also including that same loss or harm in another element of damage. When considering these elements, you must consider both actual expenses to date and also the present cash value of similar services likely to be required in the future. The third weakness involves definitions of technical terms. One way to explain the meaning of a technical term like ‘present cash value’ is to simplify the original text. Another is to offer a concrete example of its use. The present definition read thus: ‘Present cash value’ means the sum of money needed now, which, when added to what that sum may reasonably be expected to earn in the future when invested, would equal the amount of the damages, expenses, or earnings at the time in the future when the damages from the injury will be suffered, or the expenses must be paid, or the earnings would have been received. You should -54-

also consider the impact of inflation, its impact on wages, and its impact on purchasing power in determining the present cash value of future damages. (TPI 3—CIVIL 14.54) One possible corrective is paraphrasing: In other words, in order to make a reasonable adjustment for the present use, interest free, of money representing a lump sum payment of anticipated future loss, the law requires that the jury discount, or reduce to its present worth, the amount of the anticipated future loss, by taking (1) the interest rate or return which the plaintiff could reasonably be expected to receive on an investment of the lump sum payment, together with (2) the period of time over which the future loss is reasonably certain to be sustained; and then reduce, or in effect deduct from, the total amount of anticipated future loss whatever the amount would be reasonably certain to earn or return if invested at such rate of interest over such future period of time; and include in the verdict an award for only the present worth—the reduced amount—of the total anticipated future loss (suggested by The Honorable Dennis H. Inman; see Dumas,1998). An even more helpful corrective might involve the addition of a brief example: Hopefully, the simple example will be of some benefit: If you know that a person will need $1000 five years from now, you would normally not give him the $1000 now. But if you were required to give him money now, you would give him only the amount of money which, when invested, would equal $1000 in five years. How much that money should be now is for you to decide (suggested by The Honorable Dennis H. Inman; see Dumas, 1998) And that corrective might be improved by revision of the ‘if … then’ hypothetical to read thus: You may, under the Court’s instructions, find that a plaintiff is entitled to damages of $1000, but that the plaintiff will not need the $1000 for five years. You will not need to award $1000 now. Instead, you will award the amount of money that will be worth $1000 in five years. You will of course need to decide what amount of money should be awarded now. With respect to the fourth weakness, that of highly complex sentences, the most helpful process is to reduce both the number of clauses per sentence and also the depth of embedding of those clauses. Sentences in pattern instructions typically contain more than four clauses with three or four levels of embedding. In TPI 14.01 the initial sentence contains at least six clauses, depending upon the type of analysis employed: 1) If … you find that the plaintiff is entitled to damages, 2) then you must award plaintiff damages 3) that will reasonably compensate the plaintiff for claimed loss or harm 4) which has been proven by a preponderance of the evidence, 5) provided you also find 6) it was or will be suffered by the plaintiff and was legally caused by the act -55-

or omission [or condition] upon which you base your finding of liability. The final weakness identified involves providing negative information prior to providing positive information. The section on Loss of enjoyment of life specifies that ‘it is not the loss of time or actual earnings that make up this item of damages’ before stating that it is ‘the loss of the ability to earn.’ A minor revision would be helpful. On the basis of my work with the Tennessee committee, in June of 2011, I initially proposed three steps for linguists to follow in assisting bar associations to continue the important work of increasing the comprehensibility of pattern jury instructions: 1. Discover. Discover, ideally from past legal cases, whether particular instructions under consideration or particular terms, legal or otherwise (burden of proof, captious) or syntactic patterns (use of the passive voice in subordinate clauses, multiple negatives within a sentence, sentence length) have presented problems in the instructions under consideration. 2. Analyze. Obtain a full copy of the instruction/s at issue; read the instruction/s, note the problem/s; identify instructions that appear to present the greatest comprehension difficulty. 3. Report. Suggest to the bar association or appropriate committee the proposed editorial changes or rewritten versions of instructions. After receiving very helpful feedback from TPI Committee members, I added two additional steps: 4. Solicit feedback. 5. Revise. On the basis of feedback, prepare a final proposal for revision. I added these additional steps as the result of learning that I had inadvertently broadened the damages instruction in TPI 14.01 with respect to deciding that a defendant is responsible for the act or condition leading to liability. It is important for linguists to remember that the editing of jury instructions must be done with careful attention to the possibility that the original language may mislead the editor as well as a potential juror. As I continue working with the TPI Committee, I expect to be able to identify very narrowly the kinds of instructions (both civil and criminal) that most need attention. References Charrow, R.P., and Charrow, V.R. (1979a) Making legal language understandable: a psycholinguistic study of jury instructions. The Columbia Law Review 79(7): 1306—1374. Charrow, V.R., and Charrow, R.P. (1979b) Characteristics of the language of jury instructions. In J.E. Alatis and R.G. Tucker (eds) Language in Public Life [GURT 1979] Washington, D.C.: Georgetown University Press, 163—185. Cohen, N.P., and Cohen, D.R. (2003) Jury reform in Tennessee. The University of Memphis Law Review 34: 1—59. -56-

Dann, B.M, and Logon, G. (1986) Jury reform: the Arizona experience. Judicature 79: 280—286. Diamond, S.S., and Levi, J.N. (1996) Improving decisions on death by revising and testing jury instructions. Judicature 79(5): 224—232. Dumas, B.K. (1998) Jury Instructions: the Perspective from the Bench (with the Honorable Dennis H. Inman). Law & Society Association Annual Meeting, Aspen, June. Dumas, B.K. (2000) Jury trials: lay jurors, pattern jury instructions, and comprehension issues. The Tennessee Law Review 67: 701—742. Dumas, B.K. (2002) Reasonable doubt about reasonable doubt: assessing jury instruction adequacy in a capital case. In J. Cotterill (ed.) Language in the Legal Process. Basingstoke and NY: Palgrave Macmillan Press, 246—259. Garner, B.A. (2010) Interviews Conducted by Bryan A. Garner [with] United States Supreme Court Justices John G. Roberts Jr., John Paul Stevens, Antonin Scalia, Anthony M. Kennedy, Clarence Thomas, Ruth Bader Ginsburg, Stephen G. Breyer, and Samuel A. Alito Jr. The Scribes Journal of Legal Writing 13. Schwarzer, W.W. (1990) Reforming jury trials, The University of Chicago Legal Forum 119—146. Steele, W.W. & Thornburg, E.G. (1988) ‘Jury instructions: a persistent failure to communicate’ North Carolina Law Review 67: 77—119. Tennessee Bar Association Jury Reform Commission. (1999) Report of the Tennessee Bar Association Commission on Jury Reform. Nashville, Tennessee: Tennessee Bar Association. Tiersma, P.M. (1993) Reforming the language of jury instructions. Hofstra Law Review 22: 37—78. Tiersma, P.M. (2006) Communicating with Juries: How to Draft More Understandable Jury Instructions National Center for State Courts, Williamsburg, VA, 2006 [originally published in 10 Scribes Journal of Legal Writing 1 (2005—2006)]. Available at SSRN: http://ssrn.com/abstract=1507298. Tiersma, P. (2010) Redrafting California's jury instructions. In M. Coulthard and A. Johnson (eds) The Routledge Handbook of Forensic Linguistics. London and New York: Routledge, 251—264.

-57-

Garment, or upper-garment? A matter of interpretation? Eva Ng The University of Hong Kong, Hong Kong [email protected] Abstract In an adversarial common-law courtroom, where one party tries to defeat the other by using words as weapons, polysemous words more often than not pose a problem to the court interpreter. Unlike in dyadic communication, where ambiguity can be easily clarified with the speaker by the hearer, court interpreters’ freedom to clarify with speakers is to a large extent restricted by their code of ethics. Interpreters therefore can only rely on the context for disambiguating polysemous words. This study illustrates the problem of polysemy in an interpreter-mediated rape trial. It exemplifies how the interpreter’s goal to avoid contradictions by making her interpretation of a polysemous word consistent with the preceding context runs counter to that of the bilingual cross-examiner, whose primary goal is to identify inconsistencies in the hostile witness’s testimony in order to discredit him. This study also manifests a denial of the interpreter’s latitude in the interpretation of contextual clues and her loss of power in a courtroom with the presence of other bilinguals. Keywords:

POLYSEMY; AMBIGUITY; CONTEXT; INTERPRETING; BILINGUALS

1. Introduction 1.1. The bilingual Hong Kong courtroom

From 1842, when Britain began its colonial rule over Hong Kong, until 1974, when Chinese was given official status in Hong Kong, English was the only official language and trials in all courts were conducted in English. In a trial conducted in English, court interpreting is a sine qua non, as witnesses and defendants involved in trial proceedings are predominantly Cantonese-speaking. These lay-participants are either monolingual, or not sufficiently confident to testify in English, which for most of them is only a language they learned at school and may have used at work. The liberty to use Chinese in all courts after the changeover of Hong Kong’s sovereignty in 1997 has resulted in an increasing use of Chinese (i.e. Cantonese) for trial hearings over the past decade, especially in lower courts. In the High Court however English remains the dominant language (see E. Ng, 2009). Statistics from the Department of Justice show that in 2009 over two thirds of the criminal trials in the Court of First Instance of the High Court and almost half in the District Court were still conducted in English, though Magistrates’ Court cases were mostly heard in Cantonese1. Hong Kong is therefore, as K. H. Ng (2009: 120) puts it, ‘one of the most “interpreted” legal systems in the world’ and ironically, unlike in most other jurisdictions, court interpreters are hired to serve not the linguistic minorities, but the linguistic majority (E. Ng, 1997; K. H. Ng, 2009: 120).

1

Percentage of criminal cases conducted in Chinese in 2009: Court of First Instance—26.1%; District Court—55.5%; Magistrates' Courts—84.5% (Source: Department of Justice, HKSAR)

-58-

1.2. Power of the interpreter In an interpreted encounter, where the interpreter is the only bilingual, s/he is cast in a highly important role in facilitating a communicative act among the monolingual interlocutors, who otherwise would not be able to interact with one another. Anderson (2002) suggests that the interpreter’s position in the middle ‘has the advantage of power inherent in all positions which control scarce resources’. This advantage allows interpreters considerable latitude in defining their own behaviour and greater control over the interaction, and to translate selectively, while their monolingual clients are not able to ascertain the difference unless they overstep ‘rather wide bounds’ (2002: 212—213). In the bilingual Hong Kong courtroom however, interpreters do not enjoy this inherent power because the chance of working with bilingual legal personnel and even bilingual lay-participants is high. This inevitably subjects the interpreters to external pressure as their performance is scrutinised throughout the trial by the bilinguals, who can always comment on or even challenge their interpretation. In other words, the power of the interpreter in the Hong Kong courtroom is considerably reduced, as Anderson expresses it, ‘the power of the interpreter disappears, if a client happens to be bilingual’ (2002: 214). In case of a dispute over the interpretation of a single word or an expression, the bilingual legal professional, being a powerful participant in the courtroom, always has an upper hand over the interpreter as I will demonstrate with my data below. 1.3. Language in the adversarial courtroom The adversarial common-law courtroom relies heavily on the oral presentation of evidence. There is therefore a metaphor of the trial as a war of words, where opposing parties try to defeat each other by using language as their weapons (Maley & Fahey, 1991: 3), and ‘in which only one side will win’ (Danet, 1980: 190). Language is the medium through which competing versions of the same reality are constructed and presented to an impartial audience, i.e. a jury or a judge. While the prosecution tries to construct a story that will prove the guilt of the accused, the defence strives to present one that will at least create a reasonable doubt in the judge’s or the jury’s minds. The success or failure of the battle hinges to a large extent on how effectively language as a weapon is used by the parties. The strategic use of language thus becomes all-important, as Hale rightly points out, ‘[n]owhere does language play a more important role than in the adversarial courtroom’ (1997: 201). In her study of the O.J. Simpson trial, Cotterill (2003) illustrates how the attorneys employed strategic lexical choices in their attempts to orient the jury towards a particular perception of the trial events and personalities. The prosecution, through the use of words with negative prosodies and connotations such as ‘encounter’ and ‘control’, aimed to portray Simpson as an aggressive and violent wife-beater. The defence on the other hand strove to dismiss the alleged cycle of violence as mere ‘incidents’, ‘domestic disputes’, ‘verbal debates’ and even ‘conversations’. She suggests that ‘through the skilful exploitation of different layers of lexical meaning, it is possible for lawyers to communicate subtle and partisan information about victims and alleged perpetrators’ (2003: 67). Danet’s (1980) oft-cited study of a trial in a US court, where an obstetrician gynaecologist was charged with and later convicted of manslaughter in connection with a late abortion, best illustrates the power of language in the Anglo-American adversarial courtroom and the metaphor of the trial as a war of words. During the trial, the prosecution strategically referred to the result of pregnancy (RP) as ‘baby’, ‘child’ or ‘person’, which all carry the quality of aliveness and thus justify a manslaughter case. The defence on the other hand opted for terms such as ‘foetus’, ‘embryo’ or ‘product of conception’, which convey the notion that -59-

the RP was still inside the mother’s uterus and was not alive in the full sense, and therefore the abortion could not have constituted the crime of manslaughter. Through the strategic use of language both sides aimed to persuade the jury to return a verdict in their favour. The conviction of the defendant shows that the prosecution succeeded in this war of words. 1.4. Polysemy, ambiguity and context in court interpreting Polysemy refers to the association of one word with multiple meanings. The particular meaning of a polysemous word needs to be decided by the context in which it occurs. Polysemous words can be problematic in the courtroom as opposing parties attempt to arrive at meanings favourable to their case, especially when more than one meaning is possible in a particular context or due to a lack of context. In other words, the very nature of polysemy means that contextual ambiguity is often an inevitable result. Danet points out that ambiguity about the meaning of events is part of our everyday life, but ‘[w]hat makes the legal process an especially attractive setting in which to study this problem is that in the law, decisions must be made’ [italics hers] (1980: 190). In dyadic monolingual communication, contextual ambiguity can be easily clarified, although in the adversarial courtroom, counsel for the opposing side may sometimes want to leave intact a piece of ambiguous information in a witness’s testimony. Court interpreters however, bound by their professional ethics, cannot freely clarify with the speaker about the intended meaning of an ambiguous utterance as interlocutors in a dyadic monolingual communication usually do. Moreover, the use of ‘short consecutive’ as the norm in legal interpreting (De Jongh, 1992: 38), whereby the interpreter works with relatively short chunks of information at each turn, means that the contextual information available to the interpreter in court examination is, ‘limited to the local context unfolding at each turn’ (Lee, 2009: 94), or what Hatim and Mason (1997: 50) refer to as the ‘local cohesion’ covering no more than two or three sentences. Hatim and Mason suggest that to overcome this limitation, liaison interpreters, (a term which includes court interpreters), ‘resort to a more readily accessible strand of textuality’ or what they later refer to as ‘context’ (1997: 50). This means that court interpreters, like other liaison interpreters, have to rely heavily on the local linguistic context for their interpretation of an ambiguous utterance. The contextual information available to the court interpreters is limited, not only because of the mode of interpreting used in court, but also because interpreters are often denied access to the case file to prepare in advance for the trial for reasons of confidentiality. It follows that interpreters do not have the same access to the contextual information of a trial as do the other participants in the court proceedings: the witnesses and the defendant have direct experience of the case at trial and have had their own (and presumably different) versions ready to tell the court; counsel must have familiarized themselves with all the details of the case before the trial starts; by contrast the interpreter is kept in the dark. That is to say, court interpreters often lack the means–clarification questioning and context–required for disambiguating an obscure item, which may prove detrimental to the accuracy of the interpreted evidence as exemplified by Kredens and Coulthard (forthcoming). In their example, a Polish interpreter had in a police interview erroneously interpreted the alleged rape victim’s evidence about the defendant’s penetration, in the absence of a prepositional phrase, as a marked oral penetration instead of an unmarked vaginal one. This is ostensibly a result of the interpreter’s lack of contextual knowledge of the case and thus her need to rely on the local linguistic context, as well as her adherence to the ethical code by refraining from clarifying with the speaker. Had she been briefed on the details of the allegation before the interpreting took place, or had she taken the liberty to clarify the ambiguity with the speaker, the error would have been avoided.

-60-

The notion of context is central in any interaction. Drew and Heritage (1992: 18) define context in interaction as both ‘the immediately local configuration of preceding activity in which an utterance occurs’ and the ‘larger environment of activities within which that configuration is recognised to occur’. They suggest that utterances are doubly contextual in that they are both context-shaped and context-renewing because while the interpretation of an utterance is decided by the context in which it occurs, ‘the current utterance will itself form the immediate context for some next action in a sequence...’ (1992: 18). Following Drew and Heritage, Ehrlich suggests that the question-answer sequences of a trial are illustrative of the doubly contextual nature of utterances, as a question in a question-answer sequence depends on previous utterances for its production and interpretation, and at the same time it shapes and constrains the utterance (the answer) that will follow (2001: 31). This is echoed by Janney who notes that ‘interpretations of utterances depend on information provided by earlier utterances in the sequence and, at the same time, constitute information necessary for interpreting later utterances in it’ (2002: 458). This suggests that interlocutors, whether in ordinary conversations or courtroom interactions, do benefit from earlier utterances in their interpretation of later, in particular, ambiguous utterances. Interpreters working in the courtroom however do not enjoy the same degree of latitude as other participants in the proceedings do in their interpretation of contextual clues. The views that the interpreter should just translate, not interpret and that the role of the interpreter is to act as a mere conduit still prevail in the legal sphere (e.g. Hussein, 2011; Morris 1995), though empirical studies (e.g. Berk-Seligson, 1990; Hale, 2004; Roy, 2000; Wadensjö, 1998) have demonstrated to the contrary. All these studies reveal that interpreters play an active role in negotiating meaning with the primary participants and in making sense of what is said by them. When it comes to the interpretation of ambiguous words or expressions, the interpreter’s decision to opt for one meaning rather than another based on the contextual clues may attract criticism from parties who decide that it would favour their case if the word were interpreted otherwise. 2. Data and aims of the study This study represents a small portion of a larger research project on the reality of courtroom interpretation in Hong Kong. The data for the entire research project consists of recordings of 9 criminal trials from the three levels of courts in Hong Kong, namely, the Magistrates’ Court, the District Court and the High Court, totaling over 100 hours of recording time 2. The recordings were transcribed, following conventions for transcribing conversations (cf. Silverman, 2006), with information containing personal data changed or concealed in the transcript to ensure anonymity and confidentiality. The present paper focuses on a High Court rape case, and aims to illustrate the problem of polysemy in an interpreter-mediated trial conducted in English and the divergent strategies adopted by the cross-examiner and the interpreter in the pursuit of their respective goals. In the case under discussion, both the Prosecution and the Defence Counsel are bilingual in English and Cantonese, but the judge speaks only English. The point at issue is the polysemous Cantonese word saam13衫 uttered by the defendant in his evidence in chief. 2

I am indebted to the High Court Registrar of the Judiciary of Hong Kong for granting me access to the recordings of court proceedings, and the Leung Kau Kui Research and Teaching Endowment Fund of The University of Hong Kong for funding this project in part. 3 Romanisation of Cantonese characters in this study is based on Jutping, a Cantonese Romanisation system developed by the Linguistic Society of Hong Kong. This system distinguishes 6 tones in Cantonese and the number at the end of a syllable is a tone marker.

-61-

3. Analysis 3.1. Meanings of saam1 The word saam1 in Cantonese (or shān in Mandarin Chinese) is polysemous. The Chinese New Dictionary (Liu, 1993) gives two meanings to this Chinese character: 1. 單上衣 (upper garment); 2. 泛指衣服 (a generic reference for clothes). In other words, saam1 can be a generic term to refer to clothing/garment (usually used without a quantifier), or a more specific term, usually used with a quantifier gin6 件 (piece), to refer to a piece of clothing worn on the upper part of one’s body (a shirt, a blouse, a jumper and even a jacket)—upper garment, which is how it is usually translated in the Hong Kong courtroom if it is understood in the latter sense. However the use of a quantifier alone does not serve as a clear-cut distinction between the meanings of this problematic word. For example, the expression ngo5 dong1 si4 mou5 zeok6 saam1我當時冇着衫 (without the presence of a quantifier gin6) can be taken to mean BOTH ‘I was not wearing any clothing at that time, i.e. being naked’ AND ‘I was not wearing any upper garment at that time’. In other words, the absence of gin6 does not automatically give the word a generic sense, and vice versa. For example, the utterance ngo5 jat1 gin6 saam1 dou1 mou5 maai5 dou3 我一件衫都冇 買到 is usually understood to mean ‘I didn’t buy even a single piece of clothing’ despite the presence of the quantifier gin6. The argument in this case lies in the interpretation of the meaning of this particular word in Cantonese. 3.2. Prosecution case It is the prosecution’s case that the victim, Miss M, was raped by the defendant in his home. Miss M, a lesbian, had been a cohabitant of the defendant’s sister and had for over a year lived in the same premises where the defendant resided and where the alleged rape subsequently took place. On the day of the incident, Miss M’s homosexual relationship with the defendant’s sister had broken up and she had moved out of the flat, to which she returned to pack her belongings. At the material time no one was home except the defendant. According to Miss M, and the defendant’s own evidence, the defendant was wearing only a pair of shorts prior to the sexual intercourse, without any clothing on the upper part of his body. Miss M was raped while she was lying on the top bed of a bunk for a nap after packing her belongings, and she could not resist the defendant’s advance as she was weak and feeble at the material time. Miss M did not report to the police immediately after the incident because she was confused, but talked to her friends before a report was finally made to the police. 3.3. Defence case The defendant did not dispute having sex with Miss M, but alleged that the sexual intercourse was consensual. He elected to testify in the witness box and described in his evidence in chief how Miss M pulled his saam1 while he was standing on the ladder of the bunk bed, which he took as an invitation for him to go up to the bed. He acted accordingly and had sex with her on the bed. The following extract from the transcript of his testimony in chief and its rendition by the interpreter contains his description of Miss M’s pulling. (Words in italics are -62-

my transliterations/translations of the original Cantonese utterances or back-translations of the interpreter’s Cantonese interpretation. See the abbreviations and transcription symbols used in this study in the Appendix.) Extract 1: Examination-in-chief of Defendant

1.

1.

D

咁喺同一時間:,(.)佢就: si: eh用佢嘅:左手就(.)少少彎彎 地,就拉住我件衫,啫係腰度件衫。 Then at the same time, (.)she si: eh slightly bent her left arm to pull my piece of garment/upper garment (gin6 saam1), that is, my piece of garment/upper garment (gin6 saam1)at the waist area.

2.

2.

I

And then...你再講多一次。 And then...say that again.

3.

3.

D

Er 佢用佢嘅左手拉我 er 腰度,啫係腰部份嘅件衫。 She used her left hand to pull my waist area, that is, the garment...piece of garment/upper garment at my waist area.

4.

4.

I

And then, she used her left hand to pull the um (.) garment at my eh waist area.

3.4. The Interpreter’s strategy Note that the quantifier 件gin6 is used before the word saam1 by the defendant, which seems to suggest a more specific meaning of the word—upper garment. However, the interpreter’s choice of the word ‘garment’ in her interpretation seems to be a well-deliberated decision, bearing in mind that it had been established in both Miss M’s and the defendant’s previous own testimony that the defendant was not wearing any ‘upper garment’ throughout the whole incident. The use of the word saam1 with the quantifier gin6 by the defendant seems to take the interpreter by surprise. Her request in turn 2 for the answer to be repeated without first asking for the court’s leave to do so may be regarded as a delaying tactic to buy her time as she searches for an English equivalent for the Cantonese word saam1. Note also the filler ‘um’ and the brief pause before she utters the word ‘garment’ in turn 4. Apparently, the interpreter, having gained the extra time from the defendant’s repetition of his answer and a brief hesitation of her own, decides to opt for an interpretation consistent with the alleged victim’s and the defendant’s earlier evidence. In other words, the witness’s and the defendant’s earlier utterances have shaped her interpretation of the word in question. Her goal to strive for unity and consistency nonetheless goes against that of the cross-examiner. 3.5. The cross-examiner’s strategy In cross-examination, it is the goal of the examining counsel to discredit the hostile witness and hence the opposing side’s version of events. The usual strategy is to identify inconsistencies or contradictions in the witness’s testimony so as to create a reasonable doubt in the mind of the jury or the judge. It follows that when faced with a polysemous or an ambiguous word or expression in a witness’s utterance, what the cross-examiner is after is an -63-

interpretation which will create contradictions or inconsistencies with the witness’s earlier utterances. The interpreter’s rendition of saam1 as ‘garment’, which has in effect removed a possible contradiction with the defendant’s earlier utterances, was nonetheless not challenged by the prosecutor at this stage. It is possible that the prosecutor wanted to leave a reasonable doubt in the mind of the jurors, who were all English/Cantonese bilinguals, with Cantonese as their native language4 and would most likely rely on witnesses’ testimony in Cantonese, rather than the interpreted version for their verdict. Challenging the interpreter’s rendition in the defendant’s evidence in chief would give the Defence Counsel an opportunity to clarify with the defendant over what he meant by his utterance of the word saam1, which would most likely result in the contextual ambiguity and hence a reasonable doubt being removed. Instead, the prosecutor chooses to question the credibility of the defendant’s testimony in his cross-examination of him by pointing out to him the contradiction in his evidence. In the following extract of the transcript, the prosecutor first gets the defendant to confirm that he was not wearing any upper garment during the whole event. Extract 2: Cross-examination of Defendant

1.

1.

PC Alright. From the moment you opened the door for Miss M (.) until the sexual intercourse completed, nothing at the top.

2.

2.

I



3.

3.

D

=係。 =yes.

4.

4.

I

Yes.

Having got the defendant to confirm that he was not wearing any upper garment, the prosecutor proceeds to suggest that there was no such pulling by Miss M as alleged by the defendant and accuses the defendant of fabricating a story favourable to him as illustrated in Extract 3. Extract 3: Cross-examination of Defendant

1.

PC

=Alright. Now, you have made up (.) a very good story about how (.) that day developed into your going up onto (.) her bed.

2.

I



3.

D

我冇作到古仔,[我可以發誓。 I didn’t make up a story.[I can swear.

4.

I

[I didn’t. I didn’t eh make up any story. I can swear(.) by God.

4

I witnessed the empanelling of the jury on the first day of this trial as I took my students to the High Court for a visit.

-64-

5.

PC

There was only one hitch though. You were not wearing any upper garment that day.

6.

I

咁但係呢,有一點就係話當時呢,你係冇着到任何嘅上衫 個噃,係咪呀? But there is one thing, that is, you were not wearing any upper garment. Is that right?

7.

D

係。 Right.

8.

I

Ya.

9.

PC

So, how could she have pulled (.) you by the upper garment (.) at the waist position?

10.

I

咁佢當時又點樣係扯你嗰件上衫......係eh腰部嘅衫呀? So how could she have pulled your upper garment…that is...eh...garment at the waist position?

Note that the Prosecution Counsel uses ‘upper garment’, in both turn 5 and turn 9, despite the interpreter’s rendition of the word saam1 as ‘garment’. In the Cantonese interpretation in turns 6 and 10, the interpreter has to add the word soeng6上(upper) before saam1— soeng6saam1—to differentiate it from the more ambiguous word saam1 and of course her rendition of it as ‘garment’. However, since the actual word uttered by the defendant is saam1, not soeng6saam1, the interpreter must have realised the need to reconcile the discrepancy in her Cantonese interpretation in turn 10 when she hesitates after rendering ‘upper garment’ as soeng6saam1 and finally settles on saam1, in an obvious attempt to make her interpretation consistent with the actual word used by the defendant, and presumably her interpretation of it as ‘garment’ as well. It must also be noted that the term soeng6saam1 is marked and not an ordinary or idiomatic expression. The prosecutor’s challenge of the defendant’s evidence has now necessitated the expatriate judge’s review of his notes. The judge, being monolingual in this context, can only follow the testimony based on the English interpretation. As the interpreter has rendered saam1 as ‘garment’, not ‘upper garment’, his notes show no inconsistency in the defendant’s testimony. The following extract demonstrates the discussion between counsel and the judge in this regard.

-65-

Extract 4: Interaction between Judge and Counsel

1.

J

Now, the question is, did he say she pulled him by the shorts, or at the waist area. (19) Well, my note reads ‘she used her left hand to pull my garmenT(.) [at the waist area’.

2.

DC

[Yea.

3.

PC

[Yes, I have got the Chinese as well, (.) My Lord.

4.

DC

‘garment’, he said.

5.

J

‘Garment’, (.) not ‘upper garment’.

The prosecutor however points out in Extract 5 below that the word ‘garment’ is the translation of the interpreter, whereas the word from the mouth of the defendant is saam1, which he has understood to be ‘upper garment’. This subsequently leads to the playback of the recordings. Extract 5: Interaction between Judge and Prosecution Counsel

1.

PC

It was...it was...the Chinese was saam1. One can (.) play back the video...the...the digital (.) recording. Although ‘garment’ was the word er used by the lady interpreter, the Chinese out of the mouth of the witness was saam1, which means ‘upper garment’.

2.

J

(12) Er well, better listen to this.

In his later argument, the prosecutor also mentions the use of ‘upper garment’ by the Defence Counsel in his follow-up question to the defendant regarding the alleged pulling by Miss M, which the prosecutor suggests is indicia that the word saam1 uttered by the defendant is also understood by the Defence Counsel, as a Cantonese speaker himself, to mean ‘upper garment’. Here is the follow-up question by the Defence Counsel in the examination-in-chief earlier on. Extract 6: Examination-in-chief of Defendant

1.

DC

Now, when...when she used the hands to touch or drag...pull your...your... your garment, upper garment, at the waist-le… area, did you treat as invitation from her to (.)to up to the bed?

2.

I

嗱,當佢呢,係用隻手呀去...去扯你er哩一個腰嗰件......er嗰個 衫...嗰啲嘅衣物嘅時候呢,你::係咪er視佢哩個動作呢,作為係 一個邀請呀? Now, when she used her hand to pull…pull your…uh…piece…uh that one garment…those clothing at the waist area, did you:: take that as an invitation? -66-

A closer look at the transcript, however shows that the Defence Counsel starts off by using the word ‘garment’ before switching to ‘upper garment’, which in a way demonstrates his own indecisiveness over the meaning of this word. As in Extract 3, the interpreter seems to have a problem with the Defence Counsel’s use of the term ‘upper garment’, which after all is not her rendition of the defendant’s utterance of saam1. She is first observed to use the quantifier ‘piece’ (gin6, which as mentioned above, would tend to give saam1 a meaning of ‘upper garment’). Following a brief hesitation, she changes to use another quantifier ‘one’ (go3), which is however not an appropriate quantifier to collocate with saam1, before settling on ‘those’ (go2di1ge3) followed by a generic term ‘clothing’ (ji1mat6) she has finally come up to replace the ambiguous word saam1. Her move again can be regarded as an effort to strive for consistency in her interpretation of the defendant’s utterance of ‘saam1’ as ‘garment’. Following the playback of the recordings, the judge summarises the evidence in Extract 7. Extract 7: Interaction between Judge and Defence Counsel

My interpreter interpreted it as ‘garment’ in English, [not ‘upper (.) garment’.

1.

J

2.

DC [Yes, yes.

3.

J

4.

DC Yes.

5.

J

6.

DC

But, he did use the word saam1.

Em (.) when you asked him (.) the follow-up question, (.) you (.) said ‘garment’, [‘upper garment’= [first.

In the following extract, the prosecutor goes on to suggest that the translation of saam1 as ‘garment’ by the interpreter is not accurate, citing the defence counsel’s reference to it as ‘upper garment’ in his follow-up question to reinforce his argument. Extract 8: Interaction between Judge and Prosecution Counsel

1.

PC

Saam1, the Chinese used by the accused himself was saam1, and it was translated (.) as ‘garment’=

2.

J

=Yes, you say that the translation is incorrect. It should be ‘upper garment’?

3.

PC

To be er...to be exact, it should be ‘upper garment’. But, Mr. W somehow got the point and asked the question by using the word ‘upper garment’

3.6. The interpreter’s dilemma Now the time has come for the interpreter to be confronted with a misinterpretation allegation. This inevitably places the interpreter in a dilemma: adopting the suggestion of the -67-

prosecutor is tantamount to the confession of an interpretation error, whereas insisting on her earlier interpretation would for sure spark further heated discussion and would most likely attract criticism or even hostility from the prosecutor. The former would entail a loss of face on the part of the interpreter while the latter would obviously be seen as a confrontation to the authority, neither of which seems to be an easy way out for the interpreter. In any case, a decision has to be made. In this case, since the judge does not speak Cantonese, he cannot act as a referee and therefore has to leave the matter entirely in the hands of the interpreter as indicated in turn 1 of Extract 9. The interpreter’s response in turn 2 is ambiguous and has led the judge to believe that she would leave the interpretation as just ‘garment’ (turn 3). The judge seems to be taken aback by the interpreter’s decision to adopt the prosecutor’s suggestion and there is also a tone of resignation in his utterance in turn 5. Extract 9: Interaction between Judge and Court Interpreter

1.

J

(7) Well, I suppose insofar as the first one is concerned, the question is whether my interpreter is happy with the interpretation she’s uh...she’s given, or whether she wants to er qualify that in any way.

2.

I

< in a low voice >Yeah, I am happy with that=

3.

J

=You are happy with interpretation just ‘garment’?

4.

I

Er with er ‘upper’.

5.

J

‘upper garment’, okay. Right

Now having successfully established that saam1 equates with ‘upper garment’, the prosecutor goes on to suggest to the defendant that since he was not wearing any upper garment at the material time, there was no such pulling by the victim and that the alleged invitation from her is a story invented by the defendant. 3.7. Re-examination by Defence Counsel Seeing that the defendant’s evidence has been attacked by the Prosecution Counsel for want of consistency, which could be potentially damaging to the defence case, the Defence Counsel makes use of the re-examination process to give the defendant an opportunity to account for the contradiction. In Extract 10 below, the Defence Counsel tries to clarify with the defendant what he means by his utterance of the very word saam1. Extract 10: Re-examination of Defendant

1.

DC Now, but, when you said in the evidence, which is recorded, ‘腰部’...’腰部嘅衫’< jiu1 bou6 ge3 saam1—garment/upper garment at the waist area>, (.) what...what do you mean by that?

2.

I



-68-

3.

D

4.

I

啫係講緊(.)橡筋嗰個位,[er: Talking about the elastic band area, er: [Er what I meant was that the elastic er area

5.

DC Area of what?

6.

I



7.

D

Er褲嗰橡筋嗰個位。 Elastic band of the shorts/trousers.

8.

I

That is the elastic of (.) the shorts area.

9.

DC You mean the band of the...of the shorts?

10.

I



11.

D

係。 Yes.

Notice the omission of the quantifier gin6 by the Defence Counsel in his quotation of the defendant’s earlier testimony. In turn 1, Extract 1, the defendant says ‘我件衫,啫係腰度件衫’ (ngo5 gin6 saam1, zek1 hai6 jiu1 dou6 gin6 saam1—my piece of garment/upper garment, that is, the piece of garment/upper garment at the waist area), with the placing of the quantifier gin6 before saam1 in two places in his initial utterance. When asked by the court interpreter to repeat his answer, the defendant says in turn 3 of the same extract ‘ 腰部份嘅件衫’ (jiu1 bou6 fan6 ge3 gin6 saam1—the garment…piece of garment/upper garment at the waist area) as an elaboration of his answer in turn 1, which again includes gin6 before the word saam1. However, this utterance is not grammatical, with the insertion of the possessive particle ge3 before gin6 as ge3 and gin6 are syntactically redundant and semantically mutually exclusive. As has been pointed out earlier, the use of gin6 before saam1 in this context tends to give saam1 a more specific sense. By contrast, the use of ge3 before saam1 is likely to make it a generic reference. It seems that the utterance of ge3 by the defendant in the first place is a slip of the tongue, which he has then corrected to gin6. Now the Defence Counsel must have realised the damaging effect that the use of gin6 might have on the credibility of the defendant when he quotes the defendant’s earlier testimony with this word omitted. This has the effect of rationalising the defendant’s subsequent explanation about his intended meaning of saam1. 4. Conclusion This study has sought to illustrate the metaphor of the trial in the adversarial courtroom as a battle of words and to demonstrate the problem of polysemy in an interpreter-mediated trial. It exemplifies how the interpreter’s aim to avoid contradictions by making her interpretation of a polysemous word consistent with the preceding context runs counter to that of the crossexaminer, whose primary goal is to identify inconsistencies in the hostile witness’s testimony -69-

in order to discredit him. Meanwhile, this study manifests the external pressure the interpreter is subjected to and a loss of power on the part of the interpreter in the encounter when the interlocutor(s) happen(s) to be bilingual as suggested by Anderson (2002). In the case under study, the bilingual skill of the examining counsel has empowered and enabled him to challenge and ‘correct’ the interpreter’s interpretation of the ambiguous utterance to suit his purpose. The interpreter, on the other hand, has seen her power as a communicator in the encounter diminished by the bilingual counsel and her latitude to interpret an ambiguous utterance based on earlier utterances is obviously restrained; she is, as Fenton (1997: 30) puts it, characterised as someone who is not supposed to ‘take an intelligent, thinking interest in the proceedings’. The study also exemplifies the dilemma the court interpreter is faced with when the accuracy of her interpretation is called in question. Her concession may also be regarded as a submission to the power in court, where the power asymmetry between lay-participants and legal professionals is palpable. Appendix: Abbreviations and transcription symbols Abbreviations J=Judge I=Interpreter PC=Prosecution Counsel

D=Defendant W=Witness

DC=Defence Counsel

Transcription symbols = : <> (2) (.) [ boldface CAPS Italics

latched utterances, with no pause between the end of one utterance and the start of the next (i.e. no pause between turns) a colon indicates prolongation of the immediately prior sound. The length of the row of colons indicates the length of the prolongation. angle brackets contain transcriber’s descriptions rather than transcriptions. a number in parentheses indicates the length of a pause in seconds a dot in parentheses indicates a brief pause of less than a second. left square brackets indicate the start of an interruption and the utterance which is interrupted words in boldface represent elements under discussion in this paper words in capital letters indicate a louder voice relative to the adjacent talk words in italics are the author’s transliterations/translations of Chinese utterances/back-translation of the interpreter’s Chinese rendition

References Anderson, R. B. (1976/2002) Perspectives on the role of interpreter. In F. Pöchhacker and M. Shlesinger (eds) The Interpreting Studies Reader. London/New York: Routledge. Atkinson, J. M. and Drew, P. (1979) Order in Court: The Organisation of Verbal Interaction in Judicial Settings. Atlantic Highlands, N.J.: Humanities Press.

-70-

Berk-Seligson, S. (1990) The Bilingual Courtroom: Court Interpreters in the Judicial Process. Chicago: University of Chicago Press. Cotterill, J. (2003) Language and Power in Court. New York: Palgrave Macmillan. Danet, B. (1980) ‘Baby’ or ‘fetus’?: language and the construction of reality in a manslaughter trial. Semiotica 32(3-4): 187—220. De Jongh, E. M. (1992). An Introduction to Court Interpreting: Theory & Practice. Lanham, Md.: University Press of America. Drew, P. and Heritage, J. (1992) Analyzing talk at work: an introduction. In P. Drew and J. Heritage (eds) Talk at Work: Interaction in Institutional Settings. Cambridge: Cambridge University Press: 3—65. Ehrlich, S. (2001) Representing Rape: Language and Sexual Consent. London/New York: Routledge. Fenton, S. (1997) The role of the interpreter in the adversarial courtroom. In S. E. Carr, R.P. Roberts, A. Dufour, and D. Steyn (eds) The Critical Link: Interpreters in the Community. Amsterdam: John Benjamins Pub. Co, 29—34. Hale, S. (1997) The interpreter on trial: pragmatics in court interpreting. In S. E. Carr, R.P. Roberts, A. Dufour, and D. Steyn (eds) The Critical Link: Interpreters in the Community. Amsterdam: John Benjamins Pub. Co, 201—210. Hale, S. (2004) The Discourse of Court Interpreting: Discourse Practices of the Law, the Witness, and the Interpreter. Amsterdam/Philadelphia: John Benjamins Pub. Co. Hatim, B. and Mason, I. (1997) The Translator as Communicator. London/New York: Routledge. Hussein, N. M. A. (2011) Legal Interpreting in the Criminal System: An Exploratory Study. Unpublished PhD thesis. De Montfort University. Janney, R. W. (2002) Context as context: vague answers in court. Language and Communication 22: 437—456. Kredens, K. and Coulthard, M. (forthcoming) Corpus linguistics in authorship identification. In P. Tiersma and L. Solan (eds) The Oxford Handbook of Language and Law. Lee, J. (2009) Interpreting inexplicit language during courtroom examination. Applied Linguistics 30(1): 93—114. Liu, B. (Ed.) (1993) New Chinese Dictionary (Zhonghua Xin Cidian 中華新詞典). Hong Kong: Chung Hwa Book Co. Maley, Y. and Fahey, R. (1991) Presenting the evidence: constructions of reality in court. International Journal for the Semiotics of Law 4(10): 3—17. -71-

Ng, E. (1997) The Role of the Court Interpreter. Unpublished M.A., University of Birmingham, Birmingham. Ng, E. (2009) The tension between adequacy and acceptability in legal interpreting and translation. In S. Hale, U. Ozolins & L. Stern (eds) The Critical Link 5: Quality Interpreting—A Shared Responsibility. Amsterdam: John Benjamins Pub. Co., 37— 54. Ng, K. H. (2009) The Common Law in Two Voices: Language, Law, and the Postcolonial Dilemma in Hong Kong. Stanford, Calif.: Stanford University Press. Roy, C. B. (2000) Interpreting as a Discourse Process. New York: Oxford University Press. Silverman, D. (2006) Interpreting Qualitative Data: Methods for Analyzing Talk, Text, and Interaction (3rd ed.). London; Thousand Oaks, Calif.: SAGE Publications. Wadensjö, C. (1998) Interpreting as Interaction. London; New York: Longman.

-72-

Appreciate others’ beauty as well as one’s own: a contrastive multimodal discourse analysis of two courtroom trials1 Yuan Chuanyou Guangdong University of Foreign Studies, Guangzhou, China [email protected] Abstract This paper makes a multimodal and contrastive analysis of a Chinese courtroom extract and an American counterpart. The author studies the courtroom discourses from multiple visual and verbal modes, focusing on participants’ gestures and movements as well as their speech. The analysis is based on systemic-functional linguistics, including the metafunctions and Appraisal theory. Ideationally, substantive and procedural goings-on in the courts are constructed through the transitivity choices in participants’ speech, supplemented with other modes like gaze, pitch and bodily movements. Interpersonally, different relations between or among participants are construed via the choice of Mood type, Modality and Appraisal resources, and are reinforced by other modes like proximity, movement and angles. Textually, the courtroom trial is realized either as a battlefield by using oral weapons (debating) or as a lecture hall by using prepared written speech (lecturing), by means of Thematic structure and information flow, which is further reflected via settings and groundings. The study reveals that Chinese courtroom discourse is quite different from American courtroom discourse in making ideational, interpersonal, and textual meanings by means of various modes and modal density. Chinese courtroom tends to employ fewer modes than the American courtroom and discourse and meanings are not constructed as completely and thoroughly as in American courts. The differences are attributable to the different trial systems prevailing in the two countries. Keywords:

MULTIMODAL; COURTROOM DISCOURSE; METAFUNCTIONS

Introduction Even a fleeting glance into a Chinese courtroom and an American courtroom reveals a stark contrast in terms of multimodality. Chinese courtroom participants—including judges, procuratorates (the equivalent of common-law prosecutors but who are not lawyers), defendants and defense lawyers—rely heavily on language, either through written documents submitted or those documents read out loud, to convey information and construct meaning. In contrast, American courtroom participants, public prosecutors and defense lawyers in particular, employ many modalities synchronized with speech to achieve the same effects as their Chinese counterparts. American attorneys point and gaze at witnesses and/or defendants 1

This article is partial fulfillment of the project ‘Multi-dimensional Study on Language and the Law’, project

code 08GK-03, sponsored by the Guangdong Social Science Project (GDSSP). I would like to thank Professor Janet Ainsworth for her patient proof-reading and expertise in law as well as invaluable help in the final process of shaping this paper.

-73-

when addressing them, beat out the rhythm of their talk with distinct gestures, and move freely on the courtroom stage displaying various facial expressions. American attorneys do not just speak, they perform. They imbue their performances with multimodality in order to create interpersonal involvement that enhances the ideational meaning and persuasive force of their words. This paper adopts a Hallidayan approach—including Martin’s Appraisal theory—in its multimodal analysis of American and Chinese courtroom discourse. Matoesian (2008) points out ‘several limitations in current research in the field of legal discourse…(one being) a concern with verbal resources instead of an integration of verbal and visual or multimodal communicative practices (e.g., gaze, facial expression, body alignment, and realignment)’ (p.195). Matoesian (2010) further argues ‘that legal discourse involves much more than just words and that language and embodied conduct work together as co-expressive semiotic partners—as multimodal resources—in utterance construction and the production of meaning in courtroom talk.’ He rightly asserts, ‘[f]ocusing on just words neglects the role of multimodal activities in legal proceedings … and leaves the study of forensic linguistics with an incomplete understanding of legal discourse’ (2010: 541). Matoesian (2010) also points out that multimodality encompasses written texts, material artefacts, technical devices, gesture and other semiotic forms. In this study, I will follow Matoesian in focusing on gesture, gaze and postural orientation employed by American and Chinese prosecutors and defense attorneys in two high-profile cases, and will attempt to reveal how multimodality synchronizes with speech in making meanings in culture-specific court trials. This paper is a pilot study with some preliminary findings for a more ambitious research project on Multimodality in Forensic Discourse. Multimodality in (Courtroom) Discourse Analysis Multimodality has been extensively studied in the fields of semiotics and discourse analysis for at least two decades (see Goodwin, 2007; 2006; Kenden, 2004; Kress and van Leeuwen, 2001; O’Halloran, 2004; etc.). Linguistic literature on this research topic is growing rapidly, including some studies addressing forensic or courtroom discourse. Fairclough and Wodak (1997: 164) touch upon multimodality by saying, ‘social semiotics draws attention to the multi-semiotic character of most texts in contemporary society, and explores ways of analysing visual images (from press photographs and television images to Renaissance art) and the relationship between language and visual images.’ Jaworski and Coupland (1999: 7) also underline that ‘discourse reaches out further than language itself. When we think of discourse in the wider context of communication, we can extend its analysis to include nonlinguistic semiotic systems… (including) painting, sculpture, photography, design, music and film.’ In her recent work on perspectives on discourse analysis, Alba-Juez points out:

-74-

The current aim now in DA is to describe language where it was originally found, i.e. in the context of human interaction. In this respect, it is important to point out that this interaction often involves other media besides language. Examples of these other semiotic systems may be gesture, dance, song, photography or clothing, and it is also the discourse analyst’s job to explain the connection between these systems and language. (2009: 12, my emphasis) Jannedy & Mendoza-Denton (2005) argue that ‘gestures [not only] facilitate the information transfer from the speaker to the listener/viewer…[but also] play an important role for the naturalness of speech and for cuing speaker stance.’ (p. 233—234, my emphasis). They further explain that gesturing is not just a facilitation device for the listener/viewer but a mode of self-expression for the speaker. In the study of law and language, multimodality, in particular the use of gesture, has aroused researchers’ interest. Hibbitts (1995a: 3) suggests that legally significant gesture—as a phenomenon—serves at least eight broad functions: indicative, ordinative, evidentiary, demonstrative, communal, mnemonic, regulatory, and psychological. He finds that the gesture simultaneously supports and confirms the accompanying words, and that legal gesture can also express emotions which accompany or are occasioned by a legal change or relation. However, he confines his research to those conventional legal gestures such as raising one's right hand close to the body to signify the taking of an oath. He ignores the many ‘ordinary’ gestures used by lawyers and other legal actors to express emotion or emphasis, because, for him, they lack an explicitly legal import (Hibbitts, 1995b: 54). In the realm of forensic linguistics, it is Matoesian who first systematically explored so-called ordinary gestures in legal discourse, the speech synchronized gestures that he called ‘speech-gesture ensembles’. Matoesian (2008; 2010) conducted multimodal analyses over a witness’s and victim’s discourse in a high-profile trial, the William Kennedy Smith rape trial (see Matoesian, 1993 for a case brief). In the 2008 article, Matoesian analyses and discusses how multimodality, i.e., ‘a dense constellation of verbal and visual conduct including gaze, body movement, facial display, head nods, and paralinguistic features’ (2008: 200) realigns participation roles and shapes institutional identity, and describes the extralinguistic strategies that the witness employs to convey her epistemic and affect stances. Matoesian (2010) focuses on the role of gesture, gaze and posture of the victim in her courtroom talk and demonstrates ‘how a dense constellation of multimodal resources—gaze, gesture and talk’ (2010: 556) is used by the victim for the production of accusations, identity and stance. Citing Kendon (2004), Matoesian says,

-75-

Because they perform meaning visually, gestures add another dimension to speech: complementing or clarifying verbal messages, performing distinct speech acts, intensifying commitment to an assertion (functioning as stance markers), foregrounding information, coordinating the rhythm of speech—parsing it into significant segments—and pointing out objects of attention in the extralinguistic world via spatial–temporal proximity. (2010: 544) To conclude the literature review, I quote again from Matoesian (2010) to stress the significant role that multimodality plays in forensic discourse analysis, as follows: …a more adequate conceptualization of legal language requires a detailed appreciation and understanding of the role of multimodal channels of interaction and their meaning making potential in context. If courtroom discourse is multimodal and if multimodal conduct conveys meanings not necessarily transmitted by verbal means alone, then it follows that omitting such information may obscure a more comprehensive and robust description of what people are actually doing in legal proceedings—and thus ‘lose’ the phenomenon. (2010: 545) …incorporating the integration of these semiotic resources into the analysis of legal discourse offers a new direction for future studies in forensic linguistics … (2010: 546) Case briefs and data This paper presents a contrastive study of two similar criminal trials, one an American murder case, referred to as the Harvard graduate murder case, and the other a Chinese intentional homicide case, referred to as the Ma Jiajue homicide case. Both cases involve the killing of university students. In the American case, Alexander Pring Wilson, a graduate student from Harvard University, was tried for killing a stranger with a knife in a street fight. He was accused of first-degree murder, but claimed self defense. He was convicted of voluntary manslaughter and sentenced to 6—8 years in prison. In the Chinese case, Ma Jiajue, a senior undergraduate student from Yunnan University, was tried for killing four classmates over a quarrel arising out of a card game. He was charged with intentional homicide, and was convicted and sentenced to the death penalty. Both cases were broadcast live on television: Court TV in the American case, and China Court in the Chinese case. The existence of live broadcasts facilitates the present study by making it possible to split screen video clips and transcribe the excerpts for observation, analysis and comparison of the participants’ speech, gestures, gazes, and postures.

-76-

Data Analysis The first stage of analysis examines the closing arguments made by the prosecutor, named Lynch, and the defense attorney, named Kaufman, in the Harvard murder trial. Both sides presented powerful closing arguments to Middlesex Superior Court. The analysis will focus on their speech and on the speech-synchronized gestures, gazes, and postures employed in presenting their cases. As a contrast, I then describe and analyse the Chinese prosecutor’s and defense attorney’s speech and multimodal resources employed in Ma’s court trial. Through these cases, I will identify some regular patterns in the employment of multimodality in making meaning: in Hallidayan terms, the ideational meaning, interpersonal meaning, and textual meaning conveyed and constructed both by language-as-speech and by the use of other modal resources. Lynch’s closing argument of first-degree murder Excerpt 1

Ladies and gentlemen, and the jury, Michael Colono made fun of the defendant and it cost him his life, because of the defendant’s ego, because of his intoxication, coupled with his Spiderco military knife. A senseless verbal exchange escalated to a murder of an eighteen-year old man. And there is one person and one person only responsible for this senseless act of violence. And that’s Alexander Pring Wilson. Lynch, the prosecutor, in the opening of her closing argument restates and reiterates the happenings of the fight and murder, reprimanding and blaming the defendant for his ‘senseless act of violence’. Her speech specifies the moral grounds of accusation (i.e., an interpersonal meaning) by using an explicitly negative appraisal resource senseless. Notably, she deictically uses there and that to refer to the defendant in order to distance herself from him and to mark him as the object of the condemnation. Non-verbally or multimodally, the prosecutor frequently employs pointing gestures, gaze shifts, and bodily movements. The pointing with index finger in Figure 1 (a and b) go together with the words one person and one person only to accentuate the number of aggressors, and the pointing in Figure 1(c) in concert with rhythmic beats grounds her affective and epistemic stance by hammering out points of significance, such as her condemnation of the senseless act of violence. The pointing in Figure 1 (d, e, f) performs a deictic or ‘indicative’ function (Hibbits, 1995) in that it picks out the defendant’s vector in the deictic field (Matoesian, 2010). She points at the defendant and identifies him as the only person to blame. Notice the synchronization of the words,And that’s Alexander Pring Wilson with her pointing gestures, her shifts in gaze and bodily movements (facing the jury ^ towards the defendant ^ back to the jury), drawing multiple participants into her account and shifting personal and institutional alignments in play simultaneously.

-77-

(a)

(b)

(c)

one person, and one person only responsible for … (d)

senseless act of violence

(e)

And that’s

(f)

Alexander

Pring Wilson

Figure 1: Lynch’s pointing gestures synchronized with speech to convey condemnation Excerpt 2

In the final analysis, the one and only person who is responsible for the murder of Michael Colono, and he was murdered, ladies and gentleman, is Alexander Pring Wilson. (a)

(b)

the one and only person

(c)

is Alexander

Pring Wilson

Figure 2: Lynch’s pointing gestures in combination with gaze shifts and bodily movements to reinforce her stance

In this brief excerpt, Lynch employs pointing and beating gestures with her index finger (Figure 2(a)) and gaze shifting synchronized with her pronouncement of the name (Figure 2(b)) while readjusting her body position towards the jury (Figure 2(c)) to reinforce her affective and epistemic stance similarly conveyed in Excerpt 1. -78-

Excerpt 3

He used this dangerous weapon on a young man, who used only his two bare fists to defend himself. (a)

(b)

(c)

He used

this

dangerous weapon

(d)

(e)

(f)

on a young man,

who used only his two

bare fists to defend himself

Figure 3: Lynch's combined use of speech, gesture and tool to accentuate condemnation

In Excerpt 3 it is not only her words, but the prosecutor’s gestures, gazes, tones and pitch which make the presentation powerful. Verbally, the speaker chooses a material process He used this dangerous weapon to describe the event. In spite of the negative Appreciation inscribed by the appraisal term of dangerous, the verbal clause alone only dimly conveys the seriousness or gravity of the event. Gesturally and visually, however, we see she holds and opens a knife to concretely evoke the material tool on the one hand, and as well to show her instrumental and epistemic stances on the other, while saying this dangerous weapon. As for her tone of voice, she places stress on this to call for attention from the audience. Her pitch increases on the phrase on a young man. When she comes to the attributive clause who used only his two bare fists to defend himself, she clenches her fists and pauses at, and stresses, the words bare fists. Depending on context, a clenched fist could indicate anger or solidarity. Her clenched fists here act in concert with her words, vividly performing the fight itself, emphasizing her story that the fight was a knife versus two bare fists. Her gestures and tones, together with her verbal speech, dramatically reenact the event and fully encompass her strong attitude of condemnation for the defendant’s actions and her sympathetic stance towards the victim. -79-

Kaufman’s closing argument on self-defense Excerpt 4

You know your reputations are based on what you do everyday of your life. And you know that the people that know you best know everything about that because of what you do, because what you do and how you act in the world. And you know that’s being discussed in the stairways and you know that’s being discussed in the home coming, because he is remarkable in that regard.

he is remarkable in that regard. Figure 4: Kaufman employing a loosely closed fist to convey solidarity and support

In this excerpt of her closing argument, defense lawyer Kaufman focuses on the defendant’s character and reputation for peacefulness, as she later says he ‘never had murder on his mind’. She grounds her evaluation—Judgement in terms of Appraisal Theory—of the defendant’s character by drawing on the common knowledge of the audience, that is, that people’s reputations are consistent with their behaviour. The defendant in this case, Pring Wilson, is no exception, she suggests. He is remarkable in his behaviour and, his reputation, according to her. Notice that in referring to the defendant she uses the personal pronoun he to humanize her client, and assigns him the Attribute of remarkable, whereas the prosecutor chooses the impersonal that in her words and that’s Alexander Pring Wilson to dehumanize the defendant. Also, notice the defense lawyer’s accompanying gesture. When she says he is remarkable, she doesn’t use her index finger to point at the defendant. Instead, she employs a loosely closed fist to identify with her client and his character. While a pointing gesture with an index finger can have a sharper connotation of criticizing the person pointed at, a loose fist conveys instead an interpersonal meaning of solidarity and support, if not appreciation. Excerpt 5

And people don’t act out of their character in situations like this. People don’t just become another person. They suddenly argue with or respond to a Tom on the street with murder on their minds. People don’t do that. You all know that. People don’t if they’ve never been -80-

violent or aggressive when they’re drunk, and they’re not have drunk enough to make them drunk, and they’re the way they regularly are five minutes before the incident. They don’t suddenly switch and change. That’s not your collective life experience. That’s not the way it works. (a)

(b)

People don’t just become another person. They suddenly argue with or respond to … (c)

They don’t suddenly switch (e)

That’s not

(d)

and change. (f)

your collective life experience.

-81-

(g)

That’s not the way it works. Figure 5: Kaufman's hand gestures and postural orientation to align the jury with her reasoning

In this excerpt, the defense lawyer continues to invoke the jury’s common understanding of the consistency of people’s behaviour and their character. By saying you all know that, she aims to evoke a common sense assessment from the jury, by her language invoking the collective life experience of the general public. Although she doesn’t expressly mention that the victim provoked the perpetrator and caused the fight, her implication of the cause is obvious, explicated and reinforced with her hand gestures and her postural orientation or bodily movement. As Hibbits pointed out, gesture can perform functions which may broadly be termed ‘communal’, or productive of a heightened sense of individual connection to a community (1995: 65). In Figure 5 (a, b c and d), the defense lawyer expands her arms widely with open palms upward to underline her meaning. According to Kendon (2004), expanding the arms widely may be an attempt to seem larger or more commanding. With the expansion of arms, the defense lawyer is inviting, perhaps even demanding, mutual understanding and assent of the audience with her reasoning and account. In Figure 5 (e and f), she begins to withdraw and close her hands while drawing up her conclusion, that is, That’s not your collective life experience. That’s not the way it works. While saying these words, ‘collective life experience’ in particular, she keeps her arms close to her body and folds her hands against each other (Figure 5 (f)), and this gesture brings the juror audience closer to her and establishes common ground with them. In this way, she aligns the jury with her reasoning and builds solidarity with her audience. However, this gesture may also indicate her prediction of a potential clash between her reasoning and the understanding of the jury. This interpretation is reinforced by her gesture in Figure 5 (g), where she closes her hands and rubs her thumbs against each other, a gesture indicating uncertainty or unsureness. Under this interpretation, this final gesture might reveal a different meaning for her words, That’s not the way it works. She might be less confident in arguing for this proposition, attempting to convince the jurors of the truth of this proposition. Notice her facial expression, her gaze in particular, in Figure 5 (g). She looks insecure and disoriented, with a hollow gaze at nowhere specific. Also, if we look at the change in her postural orientation in Figure 5 (f) and Figure 5 (g), we see that she repositions her body and turns away from the jury, perhaps trying to avoid a challenging gaze from the audience. Her embodied gesture, gaze and posture, together with her verbal speech, create, as Matoesian -82-

claims, ‘a dynamic fusion of disparate yet complementary modes of expression to yield a more vivid, coherent and integrated performance of meaning in social interaction’ (2010: 544). Defense arguments on mental disease in Ma’s case I will now turn to the data from the Chinese trial to examine how multimodality is employed by the Chinese procuratorates and defense lawyers in Ma’s homicide case. In this case, both the prosecution and the defense agree with the fact that Ma killed four people with an iron hammer. During the trial, the two sides argued over two legal issues, one being whether Ma was suffering from a certain mental disease that would raise a doubt concerning his criminal intent or mens rea; the other being whether he surrendered himself voluntarily to the police or was arrested by them. Excerpts 6 to 8 are taken from the court debate between the defense attorney and the prosecution, focusing on the legal dispute concerned Ma’s mental condition. Excerpt 6 is a part of the defense lawyer’s arguments raising doubt about Ma’s mental condition. Excerpt 6

那么在这个案件当中,我们感觉到,让人无法置信的是,马加爵作为基本不做家务,也 没有(连)可能(连)鸡都没有杀过的在校学习的大学生,为打牌这样一个区区的小事, 这样一个区区的争执,就居然惨无人道地连续三天杀死 4 名同窗同学,这样的一个动机 发生在一个正常的学生身上,无论如何也有悖于常理,也不符合逻辑。除非马加爵在实 施上述行为当中,有非正常的因素。 Then in the present case, we feel what is incredible is Ma Jiajue, being a college student who rarely does housework and might have never (even) killed a chicken, could have atrociously killed 4 classmates in 3 successive days just for such a trivial matter of playing cards and ensuing quarrel. Such an intent, (happening) by a normal student, is anyhow beyond convention, and is illogical unless Ma Jiajue is (abnormal)… (author’s translation)

-83-

(a)

(b)

Such an intent

(happening) by a normal student

(c)

is anyhow beyond convention (illogical)

(d)

unless Ma Jiajue (is abnormal).

Figure 6 : Chinese defense lawyer sitting motionless with crossed arms and shifting gazes

In this excerpt, the defense lawyer first opines that Ma’s behaviour of killing four classmates over such trivial matters is incredible. He then argues that such an intent (happening) by a normal student is beyond convention, giving rise to his inference that Ma must be abnormal and may be suffering from some mental disease that would require a more scientific mental examination. In his argument, he mainly relies on mental process (eg., feel), material process (eg., killed) and attributive relational process (eg., is beyond convention and is illogical, and is abnormal) to construct his experiential or ideational meaning. He employs certain appraisal resources to convey the interpersonal meaning, for instance, his judgement of his client’s behaviour and character and his assessment of the incident using the appraisal words incredible, atrociously, normal, anyhow, illogical, and abnormal. Textually, he chooses a marked Theme, what is incredible to emphasize his argument. Nonverbally, in stark contrast to the American counterparts, the Chinese defense lawyer, like the Chinese prosecutor, makes very little use of gestures, except for the shaking of the fist in Figure 6 (a). Most of the time, Chinese trial lawyers sit motionlessly with their arms crossed before their chest and on the edge of the table. Crossed arms might indicate that the actor is defensive, self-protective or closed-off. Apart from crossing his arms, the defense lawyer rarely uses other gestures to accompany his verbal language in expressing his meanings. However, notice his gaze and gaze shifts. The defense lawyer first gazes at the desktop while saying such an intent, seeming to find the intent in his computer files, then shifts his gaze away from the computer, lowering his head and looking at nobody (Figure 6 -84-

(b)) while saying happening by a normal student, seeming to be lost in thought and addressing himself. He then raises his head and looks up towards the prosecution and gazes at the judges (Figure 6 (c)), saying is anyhow beyond convention and is illogical, providing his opinion. Then in the last stage, he suddenly trains his gaze at the defendant (Figure 6 (d)), underlining his inference that Ma Jiajue is probably abnormal and seeking confirmation from his client. Prosecution’s arguments on killing intent Excerpt 7

作为辩护人谈到的杀人动机这个问题,我觉得杀人的动机很简单。在司法实践当中,任 何一个鸡毛蒜皮的事情,就是一个眼神、一个口角都可能引起大打出手,… As for the question of killing intent raised by the DA, I think it rather simple. In the judicial practices, any trivial matter such as a gaze, a quarrel may lead to a fight, even … (a)

(b)

As for the question of killing intent (c)

I think it rather simple. (d)

In the judicial practices, any trivial matter such as a gaze, a quarrel may lead to a fight, Figure 7: Chinese prosecutor’s unchanging gesture and constant eye gaze

-85-

In this excerpt, the prosecutor argues against the defense case concerning the defendant’s ‘abnormal’ mental state and whether he had an intent to kill. She chooses a marked Theme (As for the killing intent) to make the topic textually salient, and puts forward her opinion (rather simple) assertively, though modified by I think, in the Rheme. She makes very little use of gesture, apart from her unchanging crossed hands or palms (Figure 7 (a-d)), which may indicate her certainty in her argument. Her eye gaze during this excerpt keeps consistent and constant, which may indicate her belief that it is unnecessary to argue about this topic. Her posture, sitting with her body hunched forward, also conveys information about how she feels, perhaps expressing being bored with or indifferent to the defense lawyer’s argument. Note also the male prosecutor’s posture and gaze in figure 7 (d). When the female prosecutor mentions a gaze, he raises his head and gazes at the defense, as if he is modelling what his partner says, that a gaze may lead to a fight. Excerpt 8

那作为犯罪的动机,只有被告人马加爵他自己是最清楚的,而且在自始至终的供述当中, 他也承认了这一点。难道我们非要无中生有地猜测出其他的犯罪动机来吗? As for the criminal intent, only Defendant Ma Jiajue knows it best, and he admits it consistently in his confessions. Do we have to, out of nothing, guess some other criminal intent? (a)

Do we have to, out of nothing,

(b)

guess some other criminal intent?

Figure 8: Chinese prosecutor's eye gaze contributing to interpersonal meaning construction

This excerpt is a further response to the defense lawyer’s argument about the defendant’s ‘abnormal’ state of mind. The prosecutor rebuts this defense argument with a forceful rhetorical question, Do we have to, out of nothing, guess some other criminal intent? By this rhetorical question, in particular, the use of the word guess to describe the mental process in question, the prosecutor constructs the ideational meaning that the defense argument is simply guesswork, entirely groundless, out of nothing. Using this rhetorical question in the -86-

interrogative mood, the prosecutor conveys the interpersonal meaning that she, as a representative of the state, has the power to criticize the defense lawyer and his argument. Textually, the prosecutor puts the circumstantial out of nothing in the Theme, treating it as known information, thus strongly implying that the defense argument is imaginary and unreasonable. This speech, utilizing the rhetorical question in this case, does explicitly express the prosecutor’s meanings. However, other modalities, especially her gaze, also contribute significantly to meaning construction here, particularly interpersonal meaning. Her gaze in Figure 8 (a), synchronized with the words out of nothing, is targeted nowhere and oriented towards nobody, conveying the interpersonal meaning of disdain, scorning to engage with the defense argument. She then moves the body and fixes her gaze at the defense lawyer (Figure 8 (b)), which, together with her use of the word guess, constructs the interpersonal meaning of disalignment, or more exactly, the moral stance of defiance and an affective stance of complaint and even contempt. Also her slightly downturned mouth (Figure 8 (b)) can be an indicator of disapproval, sarcasm or cynicism with regard to the defense argument. Discussion In the foregoing section I analysed excerpts of prosecutor and defense attorney discourse taken from an American and a Chinese courtroom trial, describing and interpreting how the legal actors make use of gesture, gaze and posture, together with their speech, to make meaning—ideational, interpersonal and textual—and to take stances—instrumental, epistemic, cooperative, moral, affective. My preliminary data analysis reveals that American and Chinese legal actors employ different multimodal resources to make meaning and take stances. Ideationally, in both American and Chinese trials, speech itself delivers the core information, or material content, or experiential meaning by different transitivity processes, such as material processes (use and kill), mental processes (think and guess), relational processes (is), behavioural processes, existential processes (There is one …). Other modalities differ significantly in ideational meaning construction. While American legal actors employ much gesturing, especially pointing with index finger (as described in Excerpts 1—3), palms and fist (as in Excerpts 4 and 5) to complement or supple the concrete facts and opinions, Chinese prosecutors and lawyers make more limited use of gestures and rely heavily on the referential content of their speech. In other words, gesture is less utilized as an expression of meaning in the Chinese court than in the American court. Interpersonally, American legal actors rhetorically emphasize in their speech qualities of tonic salience, pitch and length variation to convey and invoke various feelings in both the speakers themselves and in their audience, as shown, for example, in the example of two bare fists in Excerpt 3. They are flexible in choosing Mood types, such as statement, commands, and questioning, to establish speaking roles and interpersonal relations, that is, their respective power or solidarity relations. They make good use of appraisal resources like Judgement and Appreciation to assess the personality and behaviour of their clients, for instance, He is REMARKABLE in this regard, as analysed in Excerpt 4. As for multimodality, American legal actors use gestures like opening palms to engage (align) or disengage -87-

(disalign) the audience; they use gaze to express anger, condemnation, and contempt or constastingly, to express understanding, sorrow, or pity. In contrast, Chinese legal actors scarcely change the tone and pitch in their speech and they rarely use gestures to express their attitude. They do, however, use gaze and movement in gaze to convey certain stances, as illustrated in Excerpt 8. Textually, both American and Chinese legal actors follow regular thematic structure (Theme-Rheme) to transition between Given and New Information in their speech, but accomplish this differently. The difference lies in that the Americans use gesturing while the Chinese use gazing to accompany the speech information. Just as speech contains peaks in the Rheme to carry new information, gesturing or gesticulation in McNeill’s (1992) term consists of gesture strokes to synchronize the speech peaks, and gazing can also be staged into gaze dispersion and gaze fixation to synchronize the verbal Theme and Rheme. Nobe (1996) has documented a robust synchrony between gesture strokes and the peaks of acoustic aspects, suggesting that the information the gesture stroke carries has an intrinsic relationship with the accompanying speech information prominently pronounced with these peaks (1996: 35). Following Nobe, I demonstrate with the following two instances that there seems to be similar patterns between thematic structure and gesture strokes and gaze fixation. Gesture stroke He used this dangerous weapon on a young man [T R-h-e-m-e (GI G-e-s-t-u-r-e S-t-r-o-k-e who used only his two bare fists to defend himself [T R-h-e-m-e (GI G-e-s-t-u-r-e S-t-r-o-k-e T=Theme, GI=Gesture initiation Gaze fixation Do we have to, out of nothing, guess some other criminal intent? [T-h-e-m-e R-h-e-m-e (Gaze dispersion Gaze fixation The two illustrations suggest that the information carried by the gesture stroke and gaze fixation has an intrinsic relationship with the accompanying speech information pronounced by the Rheme. How do we account for the differences between the American and the Chinese legal actors in making meaning by different multimodal resources? The answer may lie in cultural differences, both in national culture and in legal culture. Although it is widely assumed that the use of gesture—and I would argue use of gaze as well—as an accompanying part of spoken utterances is a universal feature of human expression, as Matoesian (2010) claims, it seems that cultures may differ in the kinds of use and the degree of importance that this or -88-

that form of modality may have in that communication. With respect to national culture, I would suggest that Chinese people are more reserved or restrained, more withdrawn or withheld than Americans might be in similar contexts. They therefore tend to avoid blaming other people and refrain from open confrontation. This trait may well explain why the Chinese legal actors in this case prefer the mode of gaze, as it is a more subtle means of conveying emotions. In contrast, Americans are renowned as more open and ostentatious in their interpersonal communication, even stagey in their manner of expression. They are explicitly expressive in terms of both verbal speech and nonverbal communication such as gesturing. Therefore, they are inclined to choose gestures to convey meanings, as gestures are more explicit than gaze and other modalities. Legal culture is also significant in determining which modality is chosen by legal actors. As is well-known, the American legal system, particularly in its trial processes, is adversarial, meaning that the prosecution and the defense share equal power and have at least in theory an equal chance to persuade the jury and the judge of their cases. Therefore, they have an incentive to utilize every available communicative means and mode to impress the judge and jury. Gestures are a vivid and impressive communicative resource in this regard. In contrast, since the Chinese courtroom trial system is inquisitorial in nature, there exists a power asymmetry between the prosecution and the defense. Chinese prosecutors, or procuratorates, as officials representing the state, are endowed with much more power in the courtroom than are defense lawyers. Given their superior identity, prosecutors do not have to perform with gestures to persuade the judges, who are structurally natural allies with each other. Prosecutors usually affect a stern face and sharp eyes, gazing fixedly at the defense. The defense lawyers, in contrast, are structurally weaker, not daring to challenge and confront the prosecutors verbally or nonverbally. Verbally, they must be cautious in their choice of words in making their desired experiential meaning. Nonverbally, they tend to evade the gaze of the prosecutors and avoid gazing directly at the prosecutors in making interpersonal meaning. Consequently, their defense performances have been criticized for being less genuine and ineffective than are the prosecutors. Conclusion This paper undertakes a contrastive multimodal discourse analysis of a Chinese courtroom trial and an American one. The author analyzes courtroom discourses considering various visual and verbal communicative modes, including speech, gestures, gaze movement, and postures. This study partly supports Matoesian’s research findings as follows: Some multimodal actions display how law is not only spoken into being, but embodied into being as well. Because they perform meaning visually, gestures add another dimension to speech: complementing or clarifying verbal messages, performing distinct speech acts, intensifying commitment to an assertion …(2010: 544) The study also reveals that Chinese courtroom discourse is quite different from American courtroom discourse in making ideational, interpersonal, and textual meaning by means of -89-

various modes and modalities. Chinese legal actors tend to employ fewer modes than their American counterparts, and meanings are constructed and conveyed less completely and thoroughly than in American courts. More specifically, Americans are more effective in making meanings by gestures, while Chinese are effective at using gaze to construct meanings. These differences are attributable in part to the cultural differences in Chinese and American national cultures, and in part to differences in the trial systems prevailing in the two countries, the inquisitorial system of China versus the adversarial system of the United States. As famous Chinese sociologist Fei Xiaotong’s wisdom goes, ‘[a]ppreciate the culture and values of others as do to one's own, and the world will become a harmonious whole.’ Chinese law practitioners may learn from their American counterparts in employing multimodal resources, gestures in particular, to make their arguments more convincing and effective. As a concluding remark, I must point out two caveats of this case study: firstly, it is undeniably impressionistic and needs more systematic theoretical support; secondly, it is possible that the legal actors observed and their discourse analysed here may be idiosyncratic due to the small sample size of the two cases chosen. Certainly more data from both China and America would be helpful in assessing the validity of the conclusions drawn from this study in making broader generalizations about the nature of courtroom discourse in these two countries. References Fairclough, N. and R. Wodak (1997). Critical discourse analysis. In T. van Dijk, Teun, (ed.), Discourse as Social Interaction. London: Sage. Hibbitts, B.J. (1995a) Re-membering law: legal gesture in the past, present and future. Paper presented to the Conference on Human Factors in Computing Systems (CHI-95), Denver, CO, May 1995. Hibbitts, B.J. (1995b) Making motions: the embodiment of law in gesture. Journal of Contemporary Legal Issues 6: 51—81. Hibbitts, B.J. (1996) The re-vision of law: the pictorial turn in American legal culture. Paper presented to the Annual Meeting of the College Art Association, Boston, MA, February 1996. Jannedy & Mendoza-Denton (2005) Structuring information through gesture & intonation. Interdisciplinary Studies on Information Structure 3: 199—244. Jaworski, A. and Coupland, N. (1999) Perspectives on discourse analysis. In A. Jaworski, and N. Coupland (eds) The Discourse Reader. London/New York: Routledge.

-90-

Kendon, A., (2004) Gesture: Visible Action as Utterance. Cambridge: Cambridge University Press. Kress, G. and T. Van Leeuwen (2001) Multimodal Discourse: The Modes and Media of Contemporary Communication. London: Arnold. Kress G. and T. Van Leeuwen (2006) Reading Images: The Grammar of Visual Design (2nd ed.). London: Routledge. Alba-Juez, L. (2009) Perspectives on Discourse Analysis: Theory and Practice. Cambridge Scholars Publishing. Matoesian, G. M. (2008) You might win the battle but lose the war: multimodal, interactive, and extralinguistic aspects of witness resistance. Journal of English Linguistics 36: 195—218. Matoesian, G. M. (2010) Multimodality and forensic linguistics: multimodal aspects of victim’s narrative in direct examination. In M. Coulthard and A. Johnson (eds) The Routledge Handbook of Forensic Linguistics. London: Routledge. Matoesian, G. M. (1993) Reproducing Rape: Domination Through Talk in the Courtroom. University of Chicago Press. O’ Halloran, K. L. (2004) Multimodal Discourse Analysis: Systemic Functional Perspectives. London: Continuum. Case sources http://cnettv.cnet.com/harvard-murder-trial/9742-1_53-50023881.html# http://www.necn.com/Boston/New-England/PringWilson-pleads-guilty-to-stabbing-teen/1200 107725.html http://www.cbsnews.com/video/watch/?id=649509n http://caselaw.findlaw.com/ma-supreme-judicial-court/1043558.html

-91-

Language and the construction of a positive identity among inmates in Kenyan jails Emmanuel Satia Moi University, Kenya [email protected] Abstract This paper describes the ‘linguistic resources’ used by a group of inmates in Kenyan jails to construct their identities in a positive way. They did so in 33 letters which they wrote to a religious leader who was their benefactor. Analysed within Fairclough’s (2001) critical discourse analysis and Shuy’s (2003) theory on the speech event of image repair, the paper finds that the inmates constructed their identities positively through the use of euphemistic vocabulary, passive and active constructions that avoided presenting the inmates as wrongdoers, denial, mitigation of guilt, a description of their past lives as well as through the use of a religious register. Apparently, they resorted to these resources in an attempt to distance themselves from the crimes they had committed. Keywords:

IDENTITY; INMATES; IMAGE REPAIR; RELIGIOUS REGISTER; EUPHEMISM

1. Introduction The view that identity is constructed through language is widely held. For example, Joseph (2004: 8) argues that ‘individual identity is socially and linguistically constructed’ (my emphasis) while Bogoch (1999: 330) maintains that ‘the language chosen’ by interlocutors, ‘conveys not only the content of the message, but also information about the interlocutors’. These positions underscore the place of language in identity construction. But TabouretKeller’s (1997:315) assertion that ‘the language spoken by somebody and his/her identity are inseparable...’ is stronger in its recognition of the close connection between language and identity. Predictably, some scholars, for example, Bucholtz and Hall (2004; 2010) have postulated linguistic theories of identity construction. These and a majority of other studies on identity have, however, focused on the construction of identity in non-legal contexts. A number of studies have investigated the construction of identity within the legal context. Eades (2008: 151), for example, examines the strategies employed by defence lawyers in constructing the identity of four young boys appearing as witnesses in the Pinkenba Inquiry as being that of ‘lying criminals’ and how the boys themselves resist the imposed identities. Eades (2008: 171) also catalogues similar cases by citing Harris (1984), Matoesian (1993; 2001), and Erhlich (2001; 2005) all of who examine resistance to imposed identities. Cotterill (2003: 66) examines the OJ Simpson Trial and points out that OJ Simpson’s identity is negatively constructed by the prosecuting lawyers ‘as that of a violent man capable of murdering his wife’ while the defence lawyers make attempts to ‘minimise and neutralize the negative prosodies evoked by the prosecution’. Other studies in the legal context include Bogoch’s (1999) on the gendered construction of identity within the courtroom, Matoesian’s (1999; 2001) on the construction of expert identity and age identity and Drew’s (1990; 1992) on the defensive strategies used by a woman to fend off an incriminating version of events in a rape case. These studies seem to support Pavlenko and Blackeledge’s (2004: 27) argument that ‘... individuals are agentive beings who are constantly in search of new social and linguistic resources which allow them to resist identities that position them in undesirable ways...’ -92-

However, these cases deal with the courtroom context only. This paper seeks to make two contributions in this area of study by analyzing ‘linguistic resources’ that a sample of inmates in Kenyan jails used to construct their identities positively and by focusing on the prison context where ‘little’ has been ‘found’ about its language (Gibbons 2003: 84). 2. The source of the language data under analysis The data analysed in this paper was drawn from a corpus of about ten thousand words composed of letters from thirty three inmates and ex-inmates (all considered as inmates in the paper) in Kenyan jails. The thirty three letters were written, one each1, by thirteen female inmates and twenty male inmates and were addressed to Rev. Francis2 (not his real name). The letters tend to suggest that the inmates knew him because he was involved in pastoral activities and also assisted needy inmates in the jails that he visited. The letters cover the period between January 2002 and December 2009. Twenty one of letters were written as requests for assistance, four as expressions of gratitude while the remaining eight expressed both gratitude for assistance they had already received while at the same time made further requests of assistance. One of the letters was written to a government minister through Rev. Francis. The letters reveal a number of pertinent details about the inmates. These include reasons for their incarceration, their age, the nature of the crimes that they had committed, the duration of their incarceration and the inmates’ level of education. All but three letters reflect writers whose command of the English language may be regarded as being below the ‘fourteen years’ benchmark that Buregeya (2006: 204) suggests3. Inevitably, the letters contain many instances of expressions indicative of non-standard English. These are reflected in most of the examples which are cited in this paper all of which have been faithfully transcribed. The crimes mentioned in the letters include child negligence, theft, drug trafficking and murder. Although the letters were written in Kenyan jails, the authors are from four different countries: Kenya, Uganda, the Democratic Republic of Congo, and the United States of America. From a forensic linguistic analysis point of view, those letters, whose length ranges from between 196 and 596 words, would be considered to be of adequate length for analysis. According to Coulthard (2005: 6; 2006: 2) such texts ‘are very short indeed—most suicide notes, threatening notes and threatening letters, for example, are under 200 words long and many contain fewer than 100 words’. The data were analysed within Fairclough’s (2001) critical discourse analysis and Shuy’s (2003) theory on the speech event of image repair. Two critical aspects to the study in Fairclough’s (2001) theory were vocabulary and grammar in both their relational and experiential values while denial and mitigation of guilt, ‘cataloguing all the good things’ one has done and distancing self from the problem were core in Shuy’s (2003: 1—2) theory. In the analysis that follows, the letters from which illustrative material has been drawn will be referred to as L1, L2, L3, etc (for Letter No. 1, Letter No. 2, Letter No. 3, etc.).

1

Four of letters were written jointly but I have treated each of the letters as having been written by one inmate. The author is grateful to Rev. Francis for allowing the author to access the letters and their use for this analysis. 3 Buregeya (2006: 204) argues this benchmark was inspired by Skandera’s (1991: 47—9) discussion. He argues that ‘after so many years of having been taught the language and having been widely exposed to the English as a medium of instruction, whatever English as a medium of instruction, whatever English grammar one has acquired is definitive’. This argument would apply to these letters because all but one were written by inmates who would have learnt English as a second language. 2

-93-

3. The inmates’ use of language to construct a positive identity of themselves This section shows how inmates, in their letters, chose certain lexical items and syntactic structures that tended to distance them from the crimes for which they had been convicted and therefore helped them to construct their identities positively. 3.1. Distancing themselves from crime by using euphemistic words In 21 out of the 33 inmates’ letters, the inmates neither stated nor gave details of the crimes they had committed. Instead, they used lexical choices that were euphemistic references to the crimes. The discussion below focuses on this aspect by discussing the following words, misunderstanding, colliding, accident, and landed, to show how these have been used to distance the inmates from their crimes. The inmate in letter L16 describes his offence as a ‘misunderstanding’ when he writes: ‘I was put into prison due to misunderstanding with my landlord who owes me 10,000 and a further 18,000 from unregistered women groups’ (my emphasis), while in letter L20, the inmate explains the circumstances that led to his detention. He claims to have had a misunderstanding with police officers at a police station where he had gone to report a case of ill-treatment by a matatu4 tout. As a result of the misunderstanding, he got detained. In both cases the inmates use the word misunderstanding to refer to the offences which they have committed and which subsequently led them to jail. The Oxford Advanced Learner’s Dictionary (7th edition) defines misunderstanding as ‘a slight disagreement or argument’. The fact that the ‘misunderstanding’ in the case of the two inmates caused both of them to be jailed appears to have camouflaged the seriousness of the crimes that they had committed. Similarly, in L28, the inmate describes his arrival in prison as a ‘landing’. He explains that he had been arrested following a disagreement with his business associate and concludes his account by stating: ‘That is how I landed here.’ The word ‘landed’ appears to have some positive connotations as the following concordance listings taken from the British National Corpus appear to show: Mark Lammar has landed the job what life would be like if you landed the jackpot Actor David Threfall landed a plum role Yesterday she landed at Heathrow after tycoon Gerald Ratner’s famous ‘crap’ remark, landed in court TV star Terry Wogan landed up in a ditch the South African rider landed his seventh win SACKED chef who went on the rampage landed with a £700 bill yesterday Although there are negative connotations in some of the concordance listings given below, the use of ‘landed’ with positive connotations appears to be common. after tycoon Gerald Ratner’s famous ‘crap’ remark, landed in court TV star Terry Wogan landed up in a ditch SACKED chef who went on the rampage landed with a £700 bill yesterday It could therefore be argued that the inmates’ choice of the verb land appears to project them in a more positive light than to be jailed would. The inmates’ use of ‘misunderstanding’ and 4

A matatu is a kind of public service vehicle in Kenya, typically a minibus.

-94-

‘landed’ are therefore euphemistic and seem to be aimed at distancing the inmates from the crimes that they committed. In L20, the inmate attributes his arrest to a collision. He writes: ‘I came here [to prison] after colliding with the police at ... station’. Again, although the word colliding connotes a serious disagreement, it does not mention the actual crime or offence committed and is therefore euphemistic. By using this term, the inmate also brings in another dimension: the fact that there was more than one party involved in the ‘misunderstanding’. Therefore, should blame be apportioned, it should not go to the inmate alone but also to the police officer(s) involved. But it may also be argued that since a collision evokes the image of an accident and, in normal situations, accident victims are to be sympathised with rather than being blamed, the inmate appears to seek pity from Rev. Francis (the addressee) and wants the latter to view him more as a victim rather than an offender in the matter. In L22, on the other hand, the inmate describes his offence as an ‘accident’. He writes: ‘I came here [Kenya] as a musician whereby this accident got me...’ By referring to his offence euphemistically as an ‘accident’, the inmate is also distancing himself from the crime that he has committed and may be appealing to the benefactor’s sense of pity. His detachment is further emphasised through the use of the verb got in the segment whereby this accident got me (my emphasis). The accident ‘getting’ him suggests his non-involvement. The inmate therefore appears to suggest that he should be viewed as a victim rather than the perpetrator of the crime for which he has been convicted. However, the 8-year jail term that the inmate was serving suggests that he must have committed a serious offence. The inmate’s use of the accident is probably aimed at constructing a positive image. These examples illustrate how the inmates’ choice of euphemistic vocabulary avoids ‘embarrassing ... words’ (Mulholland (1994: 99)) and ‘negative values’ (Fairclough (2001: 99)), which may portray them negatively. The following section analyses aspects of grammar. 3.2. Using passive and active voice constructions that avoid links to the police All inmates who described circumstances of their arrest and detention used agentless passives. For example, in L13 the inmate describes the circumstances of her arrest as follows: ‘I was a house girl before I was brought to prison’ (my emphasis). In this passive construction, the inmate avoids the agent-phrase, which most likely would have been the prepositional phrase by the police. By not mentioning the police, she appears to distances herself as a participant in police matters and therefore avoids the attendant negative associations of wrongdoing which the phrase would have elicited. It also creates ambiguity, as it raises the awkward 'Who by?' question (Simpson 1993: 87). As a result, it is not clear whether she was taken to prison for a visit or whether she was taken in to be jailed. The following examples, all taken from different letters, are constructed in the same way. (i) ‘I was arrested on January 3rd 2008’ (L5); (ii) ‘We were arrested in 1997 and charged with an offence of murder’ (L21); (iii) ‘I was arrested back in the year 1997 with a case of murder at home district Marigat’ (L25). In spite of the argument that the use of agentless passives may have been used by the inmates to distance themselves from the crimes that they had committed as well as the negative associations arising thereof, it is also possible to interpret their use as a means of avoiding

-95-

‘redundancy’ (Fairclough 2001: 104) given that the notion of being arrested would imply that it was by the police. Interestingly, though, the inmate in L20, states: ‘I came here after been arrested...’ The main clause, ‘I came here,’ is in the active voice and suggests that the inmate decided to be imprisoned, as if being imprisoned is an act of one’s own volition. This type of decision would imply that for the inmate to be incarcerated was a good thing. However, given that prisons limit freedoms; the decision to go to prison which is encoded in the active voice, may be viewed as an attempt by the inmate to project herself as a strong, tough and resilient woman capable of withstanding prison life. By juxtaposing ‘I came here’ with ‘after [being] arrested’, avoids the construction ‘I was brought here after being arrested’ which may have sounded more humiliating, and in turn, may have contributed to the construction of a negative identity. 3.3. Distancing oneself from crime through image repair strategies Shuy (2003: 1—2) argues that when people do bad things and get caught most of them try to repair the damage done to their image through: denial or mitigation of [their] guilt, explaining that [their] intentions were actually quite good, shifting the real blame to someone else, placing [themselves] in the best light by cataloguing all the good things we've done, or trying [their] best to minimize or distance [them]selves from the problem with ambiguity by ‘redefining and camouflaging it in the process. The inmates’ letters reveal that the inmates who referred directly to the crimes that they had committed used three of these strategies i.e. denial, mitigation of guilt and placing oneself in the best light ‘by cataloguing all the good things’ one has done. The use of these strategies is discussed in the following sections. 3.3.1. Distancing through denial Three examples of this strategy are discussed in this section. The first example is taken from L12 in which the inmate states: ‘I am currently incarcerated at the above mentioned prison undergoing trial for murder [mention of crime] which I know deep down in my heart that I never committed [denial]’. In this example the inmate mentions his crime and immediately after denies any involvement. The inmate further emphasises his innocence by indicating the extent of his denial through the relative clause i.e. ‘which I know deep down in my heart’. Similarly, the inmate in L14 mentions the crime for which she has been incarcerated and immediately after denies her involvement. She writes: I got arrested on 18/5/05 and charged with a murder case [mention of the crime] on 25/11/05 and talking the truth before the living god I don’t know where the child went [denial]. Am the biological mother of the said child and we quarrelled with the father of the child and I left the child with him. ... The case now is NEGLIGENCY OF A CHILD but the child was never found meaning nobody knows where the child is (emphasis in the original). This inmate’s extent of denial is emphasised in his vow ‘and talking the truth before the living god’. In addition, the inmate goes a step further by downgrading the offence from that of ‘murder’ to that of child negligence. The emphasis, created through capitalisation, may be -96-

indicative of her conscious effort to minimize the charge of murder which would project him negatively. In both of these cases, denial appears to be a means that is used to distance the inmate from the crime and the negative associations that may result thereof while portraying the inmate positively before Rev. Francis, the benefactor. A further example of this strategy is found in L5 where the inmate implies her innocence by denying any knowledge of the robbery of a car in which she was arrested. The inmate’s denial is made in a fairly consistent narration of the events leading to her arrest. Her denial appears to be aimed at inviting her benefactor to evaluate her request without being prejudiced about her involvement in the robbery, she states: On 2/08/2008, I planned to come to Kisumu city to visit my aunt. I had no enough money so I accompanied another man who was also going to Kisumu, although he was a total stranger to me. To my surprise, he accepted since he had a private car. On reaching Kericho, I saw a policeman who stopped the car and thereafter arrested us. Later on while in police custody, I came to find out that the car we came with was stolen [identification of crime]. I tried to explain that I didn’t know of any robbery [denial and identification of crime] and its only that the man was a good Samaritan, but it was all in vain. That is how I landed in Kazi Nyingi Prison with capital offence and charged with 2 files. 3.3.2. Mitigation and admission of guilt Examples from the letters show that both mitigation of guilt and admission of guilt are used by the inmates. They are, however, realised in two ways. First, the inmate admits to having committed a crime and thereafter provides a justification for it either directly or indirectly. In justifying the crime, the inmates in question tended to suggest that they were involved in criminal acts because of factors such as poverty or family pressure. For example, in L34, the inmate admits to having been convicted on a charge of drug trafficking but immediately after making this admission she directly justifies her involvement in the crime by attributing it to the need to meet the needs of her large family and orphaned children under her care. She states: ‘I was arrested [charged] with drug trafficking [admission of crime] while I tried to meet my children’s needs [mitigation]. I have six children and four orphans under my care, yet all were at school’ [justification]. A similar phenomenon is found in L16 where the inmate admits to having had a ‘misunderstanding’ with his landlord but indirectly justifies it by stating that his landlord owed him 28,000/=5 shillings. The misunderstanding is therefore to be viewed as a natural consequence. The inmate states: ‘I was put into prison due to misunderstanding with my landlord [indirect admission of guilt] who owes me 10,000 and a further 18,000 from unregistered women groups’ [justification/mitigation]. In the second strategy the inmate admits to having committed a crime and then immediately after displays his/her remorsefulness for his/her crime. In L13, for example, the inmate states: ‘I did steal [admission of guilt] and I am remorseful of what I did [remorse]. I have promised God not to do the same in my life [further evidence of remorse].’ The reasons provided to justify the inmates’ involvement in the crime seem to suggest that the inmate in question had no alternative other than to commit the offence. And illegal as it may be, some of the inmates believed that they had good reasons to commit their offences and should therefore not be viewed negatively. Through the admission of guilt, statements of mitigation and expressions of remorsefulness, the inmates in question appear construct their identities positively. The last 5

28,000/= figure indicating a sum of money in the Kenyan currency. This is equivalent to about $280 or £190.

-97-

part of this section examines how the inmates place themselves in ‘the best light’ by cataloguing the good things that they have done. 3.3.3. Description of positive aspects of own lives In describing their own lives some of the inmates were keen to show that they were useful members of their societies before they were jailed. Some of them portrayed themselves as creative, skilful, hardworking, financially stable, and generous persons. The following examples reveal the inmates’ use of this strategy. In L2, in seeking to have her songs produced for sale so that she may use the proceeds not only to hire a lawyer but also to support her mother and children, the inmate writes: Reverend, I am a singer, I sings gospel songs with my own composition. I think you’ve hears it singing in the church, and it has been a blessing to the madam’s6 and also to capital7 and ordinary8. It has also changed many, because God has blessed me with this talent. Reverend, I now plead to you, to help me to produce these songs, and sell them, so that I can afford to hire a lawyer, and also help my mother and my children. The underlying argument here is that the inmate is talented and capable of utilising her musical skills to earn a living and to cater for her own needs but is forced by circumstances to seek assistance. A similar argument is advanced by the inmate in L13 who describes herself as being skilful in ‘making sweaters’ and is therefore seeking funds which would help her to put her skills into good use, raise money and fend for herself and her family. She writes: ‘I have got good skills in making sweaters with the machine, and I am sure that if I am given that machine with little funds to buy the wool and rent a room, I will be able to support myself with my children’. The inmate in L16,portrays himself more positively as an entrepreneurial person when he writes: I hereby would like to express my personal problem. Regard, I was very focused and hardworking, capable of taking care of my siblings affectionately and financially. I was in the vegetable business and able to educate my children. I built a latrine pit for my village community of which I volunteered to do with own funds. I was generous and gave out debts which I have not been paid back to date. This inmate constructs a positive identity of himself by focusing on his magnanimous contribution to his community: obviously, a man who sinks a pit latrine out of his own resources for the benefit of his community, ‘gives out debts’ [gives loans to members of his community], supports his siblings financially and is hardworking is certainly worthy of respect. So, such a man’s present predicament should be viewed as a ‘little failing’, rather than as a flaw to his otherwise positive contribution to society. This section has discussed a range of image repair strategies that were used by inmates to distance themselves from their criminal acts and thereby position themselves positively before their benefactor. In the section that follows, focus will be on the religious register indicative of a strong religious theme that runs through most of the letters. 6

Madams refers to female prison warders. Capital refers to inmates on a death sentence. 8 Ordinary, part of prison jargon referring to inmates serving custodial sentences. 7

-98-

4. Religious register and positive identity A close analysis of the letters revealed a strong religious register. This was evident in the vocabulary used by the majority of the inmates and appeared to support Crystal and Davy’s (1969: 159) claim that ‘the linguistic features which uniquely identify texts as belonging to a single variety of religious English are concentrated in vocabulary...’ The following examples illustrate three different aspects of this register. 4.1. Collocational idiosyncrasies and lexical choices The inmates’ letters reveal a selection of lexical items that are indexical of Christians. Specific illustrations of these lexical items are indicated in bold type in the examples given below. (i) (ii) (iii) (iv) (v)

‘I greet you in the name of Jesus Christ our Saviour; I hope you are doing well through God’s mercies’. (L2) ‘... through God’s will he was released last year by court of appeal’. (L24) ‘Servant of God, I have no word more, but only to pray that God continue to give you long ... life to serve the poor and the sick’. (L20) ‘… will guide and touch your kind hearts to do his will. (L32) ‘Connection to that I want to remind you once again that am still saved...’ (L26)

In these examples, the expressions God’s mercies, God’s will, Servant of God and the poor and the sick,[H]is will and saved identify the inmates as sharing similar Christian beliefs with their benefactor. A lexical choice, like saved in example 5 may be regarded as an affirmation of the inmate’s deep Christian devotion. Used this way, the collocational idiosyncrasies also serve to mark their religious identity. This appeal to sameness further underscores a shared religious identity. 4.2. Formulaic expressions A further illustration of the religious identity of the inmates is found in the inmate’s use of formulaic expressions (indicated in bold type) in their letters. (i) (ii) (iii) (iv) (v)

‘I salute you in the name of Jesus Christ. I am happy to write to you this letter and I know that God will bless you’ (L4). Much greeting in the name of Jesus Christ for staying with us during Easter...’ (L6) ‘Kindly accept my humble Christian greetings in the almighty Precious Name of our Lord and Saviour Jesus Christ’. (L7) ‘May God bless the work of your hands’. (L28) ‘And God bless you’. (L31)

These formulaic expressions, marked in salutations and conclusions, are apostrophic and lend a sense of immediacy to the letters. They are also, presumably, used to create a bond and a sense of brotherhood between the inmates and Rev. Francis. As the examples cited above show, formulaic expressions, lexical choices and collocational idiosyncrasies were used by a cross section of the inmates. One inmate, however, stood out by paraphrasing and equating the Biblical verses to Swahili proverbs as illustrated in the section below. -99-

4.3. Quotations, paraphrases and own creations of biblical verses In a bid to exploit the Christian bond that he shares with his benefactor, one inmate, who admits to liking St. Augustine whom he describes as ‘the former criminal like I’, (L25), quoted some verses in the Bible, paraphrased some and equated other verses to Swahili proverbs as the following examples show. In Example 1, the inmate quotes Romans 8:1. He writes: (i) ‘The bible says that, there’s no condemnation to those who are in Christ Jesus, who do not walk according to the flesh, but according to the spirit Rom 8:1’ However, the actual verse reads: ‘If you belong to Christ Jesus, you won’t be punished’ (Holy Bible: 977). Another paraphrase is found in his attempt to quote from Jeremiah 29:11 and Corinthians 10:13. The inmate writes: (ii)‘God has good plans to me, plans of good but not plans to destroy me jeremiah 29:11. Corth 10:13’. This is a paraphrase of the two verses which read as follows: Jeremiah 29:11: ‘I will bless you with a future filled with hope—a future of success, not of suffering’ (Holy Bible: 673). The verse in Corinthians is not specific as there are two books in Corinthians, Corinthians I and Corinthians II. However, an examination of the two chapters would indicate that the inmate was referring to 1 Corinthians: 13, which reads: ‘But God treated me with undeserved grace! He made me what I am, and his grace wasn’t wasted. I worked much harder than any of the apostles, although it was really God’s grace at work and not me’, (Holy Bible: 966). 2 Corinthians 10:13 reads as follows: ‘But I was worried when I didn’t find my friend Titus there. So I left the other followers and went to Macedonia’ (Holy Bible: 1000). In the final example of Biblical paraphrases, the inmate purports to quote from Mathew 25:39—40 and Hebrews 13:3. This is what he writes: (iii) ‘Why? Coz - have frontiled [volunteered] to help me personally without expecting fame from people or nobody else. Mathew 25:39—40 Hebrews 13:3’. The actual verses read as follows: Matthew 25:39—40: ‘When did we welcome you as a stranger or give you clothes to wear or visit you while you were sick in jail?’ The king will answer, ‘Whenever you did it for any of my people, no matter how unimportant they seemed, you did it for me’ (Holy Bible: 853). Hebrews 13:3, on the other hand reads as follows: ‘Remember the Lord’s people who are in jail and be concerned for them. Don’t forget those who are suffering, but imagine that you are there with them’ (Holy Bible: 1050). In both of these paraphrases, the inmate equates the preacher’s visit and assistance to what the scriptures say. He seems to suggest that Rev. Francis’s visit is similar to that which is alluded to in the Bible. The inmate’s ingenuity is to be found in the following examples which illustrate how the inmate equates Biblical verses to Swahili proverbs, apparently, to demonstrate his knowledge of the scriptures. In Example 4 the inmate writes: (iv) ‘Wise man says in Swahili, Ada ya mja hunena muungwana ni kitendo-[Actions speak louder than words] au akufaaye kwa dhiki ndiye rafiki.[A friend in need is a friend in deed], Prov 17:17’. -100-

In this quotation, the inmate equates the two Swahili proverbs to the verse in Proverbs 17:7 which reads: ‘A friend loves at all times, and a brother is born for adversity’. Although the two Swahili proverbs may not substituted for the relevant Biblical verses, the fact both the proverbs and the verses share a thematic semblance may be regarded as a reflection of the inmate’s knowledge of the Bible. Viewed against Riley’s (2007: 155, 123) arguments that the ‘use of domain specific discourse, including its lexis, will be perceived in itself as a claim to membership of the epistemic community in question,’ and that ‘knowledge is largely constitutive of identity’, these collocational idiosyncrasies, paraphrases, and associations of Biblical verses to Swahili proverbs may be regarded as the inmate’s attempt to portray himself as one who is conversant with the scriptures and therefore deeply religious. This identity carries positive connotations and contrasts with his earlier criminal identity. The strong use of the religious register illustrated in the preceding paragraphs, seems to go beyond just reflecting strong Christian convictions on the part of the inmates. As Johnstone (2002: 126) argues, the inmates may be ‘adopting their behaviour to the behaviour’ of their benefactor in order ‘to reduce social distance’ (Gibbons 2003: 117). But more significantly, going by Giles (1979: 48) argument that where choices are involved, ‘we tend to choose the alternative which maximizes the chances of a positive outcome...’, the inmates’ choice of the religious register may be a deliberate attempt to align their identities with that of the benefactor, a Christian minister, in order to achieve their goals, i.e. gain his assistance. This view is confirmed by a number of the inmates who wrote to Rev. Francis to express their gratitude for assistance that he had accorded them as well as those who expressed their gratitude and placed further requests for assistance. 5. Conclusion This paper has shown how a group of inmates in some Kenyan jails constructed their identities positively through the use of ‘linguistic resources’, namely, euphemisms, grammatical choices, denial, mitigation of guilt, descriptions of the inmates’ past lives and the use of the religious register, all of which distanced them from the crimes that they had committed. These ‘linguistic resources’ portrayed them not as criminals (convicted of crimes) or suspects in serious crimes, but as ordinary citizens one may expect them to be. The same language also portrayed them as God-fearing and God-praising individuals (like the Reverend to whom they were addressing the letters) who knew the word of God very well and could demonstrate their knowledge by quoting directly from the Bible and by paraphrasing Biblical verses. Others still portrayed themselves as victims of some ‘misunderstanding’, or ‘collision’. Regarding their arrests, some inmates ‘landed’ in prison as if by ‘accident’ while others ‘came’ (and were not ‘brought’ there) there, as if of their own free will. But for those who were ‘brought’ there, it was by some unknown negative force. Ultimately, the inmates construct a positive identity for themselves. References American Bible Society (2004) Holy Bible—Youth Bible Global Edition. Blackledge, A. and Pavlenko, A. (2001) Negotiation of identities in multilingual contexts. International Journal of Bilingualism 5(3): 243—257. Bogoch, B. (1999) Courtroom discourse and the gendered construction. Law and Social Inquiry 24: 329—375. -101-

Buregeya, A. (2006) Grammatical features of Kenyan English and the extent of their acceptability. English World-Wide 27(2): 199—216. Cotterill, J. (2003) Language and Power in Court: A Linguistic Analysis of the O.J. Simpson Trial. Basingstoke: Palgrave. Coulthard, M. (2005) ... and then...language attribution and author attribution. http://www1.aston.ac.uk/lss/staff/coulthardm/ . Coulthard, M. (2006) Some forensic applications of descriptive http://www.businessenglish.ch/downloads/Malcolm%20Coulthard/Forensic.applications.pdf.

linguistics.

Crystal, D. and Davy, D. (1969).Investigating English Style. Harlow: Longman. Drew, P. (1992) Contested evidence in cross-examination: the case of a trial for rape. In P. Drew and J. Heritage (eds) Talk at Work: Interaction in Institutional Settings. Cambridge: Cambridge University Press, 470—520. Drew, P. (1990) Strategies in the contest between lawyer and witness in cross-examination. In J.N. Levi and A.G. Walker (eds) Language in the Judicial Process. New York: Plenum Press, 39—64. Eades, D. (2008) Courtroom Talk and Neo-colonial Control. Berlin: Mouton de Gruyter. Fairclough, N. (2001) Language and Power. 2nd edn. Harlow: Longman. Gibbons, J. (2003) Forensic Linguistics: An Introduction to Language in the Justice System. Oxford: Blackwell. Giles, H. and Clair, R. (1979) Language and Social Psychology. Oxford: Blackwell.

Johnstone, B. (2002) Discourse Analysis. Oxford: Blackwell Publishing. Joseph, E. (2004) Language and Identity. London: Palgrave. Joseph E. J. (2010) Identity. In C. Llamas and Dominic Watt (eds) Language and Identities. Edinburgh: Edinburgh University Press, 9—17. Mulholland, J. (1994) Handbook of Persuasive Tactics. London: Routledge. Riley, Phillip. (2007). Language, Culture and Identity. London: Continuum.

Shuy, R. W. (2003) Tobaccospeak: image repair as a variety of American English. Proceedings of the American Dialect Society Atlanta, Georgia, January 2—5 2003 http://www.uga.edu/tobaccodocs/papers/tobaccospeak.doc

-102-

Simpson, P. (1993) Language, Ideology and Point of View. London: Routledge. Tabouret-Keller, A. (1997) Language and identity. In F. Coulmas (ed.) The Handbook of Sociolinguistics. Oxford: Blackwell, 315—326.

-103-

Norm-enacting activity as on object of study in forensic linguistics: propositions and first impressions João Pedro Pádua PUC-RIO, Brazil [email protected] Abstract Forensic linguistics, in its more than three decades of research and academic endeavour has so far paid little attention to the process by which legal norms are created and enacted. More general discourse analytic endeavours, on the other hand, when electing parliamentary discourses and activities as its object of study tend to focus mostly on more political activities and implications of these activities—thus leaving aside the more legal-oriented types of activity. This paper aims at giving examples of how to address the specific activity type of norm creating discursive interaction inside parliamentary contexts from a genre analysis-based method (or, it will be argued, activity-type analysis-based method), that is already common in forensic linguistics. It will be argued, then, that this kind of analysis can bring new insights to the studies in Law and Linguistics. To conclude, some interesting subtopics and future research trends will be proposed. Keywords:

FORENSIC LINGUISTICS; PARLIAMENTARY CONTEXTS; ACTIVITY TYPES; NORMENACTING ACTIVITY; LEGAL NORMS

1. Introduction One of the most important tenets of modern legal theory and of the institutional designs of modern states is the Separation of Powers. The first systematization of that idea as it is generally accepted today is generally credited to Montesquieu, a French Baron from the 18 th century. In describing what he saw as the general institutional design of the modern states, reflected specifically by the English Parliamentary Monarchy, Montesquieu ([1748] 1973: 156—157) stated: There is, in each state, three species of power: the legislative power, the executive power of the things that depend on the international law and the executive power of those that depend on the internal law1. By the first, the prince, or magistrate, makes laws for a certain period or forever and corrects or abrogates those that are in force. By the second, he makes peace or war, sends and receives embassies, establishes security, prevents invasions. By the third, he punishes the crimes and judges the disputes between the individuals. We will call this last power the power to judge and the other, simply the executive power of the state. 1

The original terms from the Portuguese translation that I used are ‘direito das gentes’, that I translated here to international law and ‘direito civil’, that I translated here to internal law. The Portuguese phrases seem, themselves, to be a direct translation of the original French phrases ‘droit des gens’ and ‘droit civil’, that are the first phrases used to distinguish, in the Roman-Germanic Law Family, international and internal (or national) Law, before the actual branch of International Law was consolidated. I chose those terms for the English translation because, to my knowledge, a phrase like ‘People’s Law’ or ‘Law of the People’ would bring unwanted ambiguities, due to the difference in the development of the Law from Continental Europe to England. In any case, Montesquieu, about two paragraphs later from the one quoted uses the more familiar phrases ‘legislative power’, ‘executive power’ and ‘judicial power’, instead of these somewhat misleading ‘types’ of executive power mentioned here.

-104-

Some hundred years later, the Framers of the American Constitution have put that same separating of powers idea into force, but gave it a far more radical institutional design, when creating the first written constitution in History, further adding to Montesquieu’s description the concept of ‘checks and balances’ (Kingdon, 1999). Other Western European nations, through Liberal Revolutions, in the same political and societal context of the Modern Era, followed, with pointed differences, the same path in political institutional design (see, generally, Bobbio, [1985] 2000; Habermas, [1998] 2000). In contemporary constitutional states, which resulted from this Modern movement, the basic assumption is that it is up to the Legislative Branch of government to create the general, impersonal and abstract norms, to the Executive Branch to enforce those norms and regulate them to concrete details, and to the Judiciary to apply them to particular cases. Therefore it seems obvious that when linguists step in to gather data and analyse the legal phenomena from their point of view, the last two branches will receive their first attention—especially the Judiciary, which turns general norms into concrete provisions out of societal disputes. And yet, it all starts with the Legislative Branch (or legislative activity, when in Parliamentarism). Even in countries in the Law Family of Common Law (U.K. and former colonies, mostly—cf. David 1998), the legislatures still have a very pronounced power in terms of creating Law (i.e.: legal norms). In the U.S. there seems to be even a crescent role for legislature-created statutes, in matters previously left to the common Law developed by the courts (cf. Tiersma, 2005). In the Law Family of Romanic-Germanic Legal Systems (also called ‘Civil Law’), the general norms enacted by parliaments (or whichever bodies holds the Legislative Power) are the primary source of Law (cf. David, 1998; Tiersma, 2005). It then seems natural that norm-enacting activity inside parliamentary contexts be a point of concern for the linguistic study of Law and legal contexts. In what follows it will be briefly shown that the scant interest that discourse analysis have shown for parliamentary contexts so far is both not directed at norm-enacting activities and (for that reason) not very useful to forensic linguistics. Afterwards it will be shown that genre analysis already very developed within forensic linguistics can also be profitably used for norm enacting activities, and, to demonstrate that, it will be presented some data examples, along with some first analytical impressions. The paper will be conclude with some remarks about practical issues of Law and Politics that can be better addressed from the vantage point of this sort of discourse analysis. 2. Discourse analytic interest in parliamentary contexts There are by now some interesting, although few, studies that focus on interaction inside parliamentary contexts (e.g. Gales, 2009; Van Dijk, 2004). Summarizing those kinds of studies, Ilie (2010: 880) asserts: Parliamentary discourse and parliamentary rhetoric have been for some time the object of a great deal of research in the fields of sociology and political sciences […], but only recently have they become a truly interdisciplinary concern through the involvement of linguistic and rhetorical scholarship. The linguistic perspective has brought to the fore not only crucial aspects of parliamentary interaction patterns, institutional role shifts, confrontational speech acts, metadiscursive strategies, but also important clues about moral and social standards, prejudices, taboos, as well as value judgements of different social—political groups in particular societies. In applying discourse analytical methods to parliamentary contexts, the kind of studies described by Ilie seems to look for the interactional aspects of exchanges inside of, or related to parliaments, in order to connect what goes on inside parliaments with larger political discourse -105-

aspects and affordances that permeate the whole public sphere (in the more broad habermasian sense2; cf. Habermas, 2001). If we are to understand the role of parliamentary practices in identifying, defending or discarding topical political issues, we need to explore recurring institutionalised linguistic patterns, rhetorical devices and argumentation strategies preferred by MPs [Members of Parliament], which can reveal hidden agendas, ideological beliefs and persuasion/dissuasion strategies. (Ilie, 2010: 880) In studying parliamentary discourse, therefore, discourse analysis has so far mostly connected it to political discourses in general, seeing parliaments as advantage social loci for data harvesting and methodological expansion of other studies in political discourses that focused in other loci. Seeing parliaments and legislative bodies in general as social contexts for the production of legal norms imposes a shift in the way in which parliaments are studied, through a discourse analytical point of view. First of all, it needs recognizing that parliaments are a social place where an intrinsic institutional hybridity (Sarangi, 2000) determines itself. As noted by political philosophers (Habermas, 2001), parliaments in democratic regimes have, as their ultimate institutional goal, the mediation of the political formation of the citizen’s opinions and will and the designing and enacting of general norms that express those opinions and will. In so doing, parliaments coordinate a hybridization between political constraints and communicative goals, on the one hand, and juridical constraints and communicative goals, on the other. Although this mixture (i.e. hybridity) should be pronounced through all discursive interaction inside parliaments (in such aspects as roles and identity; display of specific communicative goals; social presuppositions invoked; facework; message framing and categorization, and the like—Sarangi, 2000), it should be even more prominent in the norm enacting activities. That is because here the ultimate juridical goal of producing a written normative text (the legal norm) is the very institutional goal that gives the activity its specific social reason (and, thereby, its communicative motif). As such, it also serves as the ultimate source of institutional constraints and presuppositions (Goffman, 1983; Levinson, [1978] 1998), both of which serve to mark the activity with its juridical aspect. It is also for this reason that this specific activity type is particularly important for forensic linguistics. 3. Genre (activity type) analysis in forensic linguistics Although a relatively new field of inquiry and research, forensic linguistics has, since its beginning, concerned itself with establishing particular aspects of the use of language inside legal contexts. Early accounts of a legal language were specifically aimed at describing regularities and patterns that served to differentiate (and sometimes oppose) this language use in juridical documents and juridical activities and settings from the general aspects of language use in everyday contexts and activities (or other institutional contexts). Those accounts were directed both at the historical changing and shaping of this legal language and at synchronic descriptions of language use in legal settings (Coulthard and Johnson, 2007; Tiersma, [1999] 2000). Starting from those early studies, forensic linguistics still devotes a big part of its studies to describe, map and analyse language constituted and mediated activities and setting within the legal phenomena. Coulthard and Johnson (2007), trying to provide a broad systematization and grouping of the studies conducted at the interface of linguistics and law, divided them among two big groups: studies aimed at describing and analysing the ‘language of the legal process’ and 2

For what matters here, Habermas divides the public sphere, as a categorical concept, in two parts, linked to each other through various direct and indirect methods of political participation and norm imposing: the formal public sphere, that relates to those holding political offices and, so, political power; and the informal public sphere, that relates to general forums and contexts where public discourses are produced with the ultimate aim of political opinion and will formation.

-106-

studies that use linguistic analysis tools to produce or improve evidence gathering and analysis inside legal processes (which the authors term ‘language as evidence’) (Coulthard and Johnson, 2007: 8)3. Studying the ‘language of the legal process’—or more broadly, studying the legal language—implies studying how language organizes and is organized by legally relevant settings. As Maley (1994: 11) puts it: Language is medium, process and product in the various arenas of the Law where legal texts, spoken or written, are generated in the service of regulating social behavior. Particularly in literate cultures, once norms and proceedings are recorded, standardized and institutionalised, a special legal language develops, representing a predictable process and pattern of functional specialisation. Coulthard and Johnson (2007: 55ff.) gloss these studies as studies of legal genres, which they define as such: Genre can be defined simply: conventional, repeated and distinctive features of text that arise from its communicative purpose. […] Genres are therefore domain specific; they have lexico-grammatical features associated with them; they have specific stages, often named and usually sequenced that contribute to a structural whole; while each stage has a purpose in the text as a whole, some of them may be optional […]. (p. 55-56) Studies of legal genres, then, are already a big part of forensic linguistics. These studies help determine specific structural and inferential frameworks that people—both lay and professional—use when participating in an interaction that take place inside a legally relevant settings. As Levinson ([1978] 1998: 97) puts it (using the intimate related concept of ‘activity type’, which he created): […] Activity types […] play a central role in language usage. They do this in two ways especially: on the one hand, they constrain what will count as an allowable contribution to each activity; and on the other hand, they help determine how what one says will be ‘taken’—that is, what kind of inferences will be made from what is said. In utilizing linguistic analysis of institutional contexts, the mapping and structuring of genres or activity types is all the more important, because the researcher needs to differentiate what is going on inside those activities and the specific accounts its members make of them as a takenfor-granted knowledge for that activity, from the general common and practical knowledge about everyday activities that is taken for granted by the generality of people (Garfinkel, 1967). Although people in general (including the analyst) do know the tenets and patterns of this latter kinds of activities, they do not (necessarily) know the tenets and patterns of specific institutional activities. Uncovering those activity specific aspects is needed to determine exactly what it is that is going on (Goffman, [1974] 1986) in such activities and to make reflexive sense of them. Forensic linguistics has profited from this kind of work in activities such as police interviews, courtroom trials, mediation, witness questioning, judicial opinion writing, and expert evidence presenting (cf., for a recent review, Coulthard and Johnson, 2010). Stemming from 3

Coulthard and Johnson (2010) revise this early twofold division, postulating a new threefold division instead: ‘i) the study of written language of the law; ii) the study of interaction in the legal process […] and; iii) the description of the work of the forensic linguist when acting as an expert witness’ (loc. 788 [of the Kindle® edition]). Since both divisions have aimed mostly at systematic and organizational purposes, for this paper, the first twofold division seems more clear-cut. Coulthard and Johnson (2010) concede that ‘this [division] is simply a convention’.

-107-

those studies, a whole range of other work could be done in the intersection of linguistics and law, including assessments about the efficacy of legal interviewing (Rock, 2001), assessment of statute interpretation (Kaplan and Green, 1995), assessment of equality and fairness in trials (Solan and Tiersma, 2005) and police operations (Shuy, 2005), just to cite a few examples. So far, however no similar attention has been paid to norm-enacting activities. 4. Analysing norm-enacting activity Just as forensic linguistics has profited from the specific analysis of a number of activity types that help construct and get constructed by specific legal settings, applying the same principle to the norm-enacting activity should provide an additional setting to look for, in order to try and understand what goes on inside legal contexts. So it will be presented, now, some sample data and some preliminary findings, in order to begin mapping norm-enacting activity as a specific activity type inside the legal domain (but with various political tenets). All the examples presented hereafter were extracted from a set of lay transcripts of faceto-face interactions of a series of Committee meetings 4 that took place inside the Brazilian Congress, during the National (Brazilian) Constitutional Assembly of 1987-88. The Committee I am citing from was located at the beginning (first phase: from April, 1st through June, 15th, 1987) of the Constitutional process and had, as its express institutional goal (as set by the Constitutional Assembly Bylaws), to present a written document (a project) that would contain proposition of legal norms to be incorporated, in the second phase of the Constitutional Process (from July 1987 on), in a unified document that would become the Constitution (eventually enacted in October, 15th 1988 and in force, as amended, ever since). For this particular committee, the proposition they needed to present concerned individual, political and social rights, as its main part. Example 1: MRS. CHAIRWOMAN (Anna Maria Rattes): — The next speaker to have the floor is Constituent Franscisco Rollemberg. MR. CONSTITUENT FRANCISCO ROLLEMBERG: — Mr. Chairwoman, Mr. Reporter, I am analyzing very carefully the bill [anteprojeto] that Your Excellency offers us this afternoon. Being the last debater, almost nothing is left for me to analyze, except what has already been analyzed. Initially, I would like to say that your work is full with humanism, very well written, with correct Portuguese, but it sins, sometimes, for wanting to explicit too much. Allow me, Mr. Reporter, to make some comments on some of the articles of this body of norms [articulado] that you present to us in your bill report [anteprojeto de relatório] [nine paragraphs of considerations on the text of the bill omitted] Now, in the letter ‘g’ of article 3, item I, Chapter I, it is said that ‘because of absolute incapacity for paying, nobody can be deprived of public services of water, sanitation and energy’ Mr. Reporter, in actual practice, those who pay and those who cannot pay, in almost every city in Brazil, do not count on public services of water and sanitation. Some have electric energy, because we are building rural electrification, we are electrifying slums. 4

The corpus used to this paper—from which these example stem—consist of the transcripts of approximately ten Committee meetings and are all available online, through a public website, maintained by the Brazilian Congress (Retreived from http://www2.camara.gov.br/atividade-legislativa/legislacao/Constituicoes_Brasileiras/constituicaocidada/publicacoes/anais-da-assembleia-nacional-constituinte). All translations from the original data in Portuguese are mine

-108-

Maybe you could merge the ‘f’ with the ‘g’ and write as so: ‘because of absolute incapacity for paying, nobody can be deprived of public services of water, sanitation and energy, with its victims having right to the assistance of the State’. Then we would have one sole item—maybe it could even be better written then my suggestion here. There is another paragraph that scares me a little. It is the item III, letter ‘e’, that establishes ‘men and women are equal in rights and obligations, including those of a domestic and familiar nature...’ I think it should end there, not continue like this: ‘... with the only exception of those relative to pregnancy, labour and milking’. […]. This first example is a standard (unmarked), typical sequence of interaction in norm-enacting activity in the Brazilian Congress. From a structural standpoint, the data shows a long, typically non-interrupted turn of talk, that has a very ritual start from the moment the chair of the session concedes the floor to a previously inscribed speaker. The ritual act of conceding the floor is matched by a ritual first act of marking its receiving, evident in the first two complimentary addresses (‘Mr. Chairwoman, Mr. Reporter’) the speaker is compelled to make before he can start conducting his substantive message. Turn design, ritual openings and closings and specific rules for the exchange of talk are some of the aspects that establish this activity as a specific example of institutional interaction (Drew and Heritage, 1998). Also in this example some salient features of role attribution and enacting (Sarangi, 2010) are visible. In terms of discourse roles (Sarangi, 2010, drawing heavily on Goffman, [1979] 2002), the Chair (in this case, a chairwoman, who is temporary replacing the chairman) and the Reporter are the addressees of the message, a fact displayed not only by the ritual opening sequence of the turn, but also by the vocatives used throughout the text, and by the secondperson deictic terms (‘your work’; ‘you’) directed primarily at the reporter, and contrasted with other deictic fields (Hanks, 2005) such as that of the speaker (‘allow me’, ‘I would like’) and that of a generalized audience (‘those who pay and those who cannot pay’). There is also a host of ratified hearers who are the other Constituents present at the same session, but are nevertheless not addressed by the speaker directly. This example also shows the interesting activity role of the Reporter, on which stands its discourse role. The Reporter, who is also a Constituent5 (in the case of the data, a Senator elected by a state in Southern Brazil), was chosen by Party leaders in the first steps of the instalment of the Constitutional Assembly to occupy this special role. This additional role (besides being a Constituent), relates to, and derives from, the additional action (cf. Goffman, 1961, on the essential relation between role and social action) the Reporter must perform. As a Reporter, he must, first, write a bill proposing how the normative text that will result from the works of the Committee would result, in order to give an object for the (oral and written) debates of the Committee. Then, after the oral debates inside the face-to-face meetings at the Sessions of the Committee, and the parallel presentation of written amendment propositions by the members of the Committee, the Reporter must present a final report that contains a definitive written bill, that should reflect both the oral debates and the written amendments, whose approving is recommended to the Committee by the Reporter. It is this final bill report that will be put up for voting by the Committee members.

5

I use the term ‘Constituent’ to identify the members of Congress, following the actual Portuguese word (Constituinte) used by the participants themselves to refer and address one another. All Constituents are, therefore, members of Congress and, besides being Constituents (an identity (or role) name derived by their being part of a Constitutional Assembly), they are also (and originally) Representatives or Senators. Roughly speaking, they are equivalent to what discourse studies of parliamentary context usually call ‘MPs’ for Members of the Parliament.

-109-

As a consequence the oral debates are centred on, and refer constantly to that first written bill, whose Animator, Author and Principal (in goffmanian terms) is the Reporter, in order to build the final bill report, whose Animator and Author is still the Reporter, but whose Principal, after the vote, is the entire Committee. Not all exchange sequences found in norm-enacting activity follow this unmarked type. In the next example we will look into a very common marked variation of that standard, which occurs when one of the participants in the session starts a direct debate with the person who has the floor, as if the activity were closer to an ordinary conversation. Example 2 [the Constituent Joao Paulo was the original enrolled speaker, but was first interrupted by the Constituent Samir Achoa some 28 turns before the first turn in this excerpt, whereafter a long exchange of turns ensued]: MR. CONSTITUENT JOÃO PAULO: — The worker is not irrational, Constituent Samir Achôa. MR. CONSTITUENT SAMIR ACHÔA: — There are workers and workers; there are Congressmen and Congressmen. MR. CONSTITUENT JOÃO PAULO: — There are workers and exploiters; workers and dominators. MR. CHAIRMAN (Mário Assad): — This Chair will not allow parallel debates. MR. CONSTITUENT JOÃO PAULO: — Your Excellency, Constituent Samir Achôa, takes responsibility for your part, and we take responsibility for ours. MR. CONSTITUENT SAMIR ACHÔA: — There are workers, in their great majority responsible, and there are those irresponsible. MR. CONSTITUENT JOÃO PAULO: — Your Excellency would not want to punish the irresponsible by punishing the majority, punishing all of them. MR. CHAIRMAN (Mário Assad): — This Chair will not allow the parallel debate and the intervention will only be conceded with the permission of the speaker. In terms of the structuring of the activity, this example shows participants’ comprehension of the rigid speech-exchanging structure as a normative expectation of the framing of the activity (Goffman, [1974] 1986). The direct intervention of another participant during one participants’ turn of talking is not a regular feature of the activity—which is compatible with the fact showed in Example 1, that turn assignment follows previous inscription. Also regarding activity structure, this example shows that, besides the discourse role of addressee, the Chairman has an activity role of controlling the structure of turn assignment and exchange. In doing this, the Chairman functions as a sort of gatekeeper to the conversational floor, a role afforded to him by the Assembly’s Bylaws, and locally confirmed through the ongoing construction of the interaction. In terms of the social presuppositions (Goffman, 1983) brought by the participants to the content of their turns and the contribution of these presuppositions to the formation of their roleidentity (Sarangi, 2010) as constituents (and congressmen, more generally), this sequence demonstrates that certain topic bring to the fore a necessity for immediate reaction. The Constitutional Assembly was a symbolic marker of the passage from an Authoritarian Military State to a Democratic State in Brazil, after more than 20 years. The primary justification that the Military regime used to sustain its right to govern was the threat of communism presented by President João Goulart, deposed by the Military Coup of 1964. So categories such as ‘workers’ and related semantic units such as ‘strike’, ‘assembly’, ‘protest’, ‘(private) property’, ‘production’ evoked some heavy social presuppositions derived from the communism-capitalism divide, still very socially relevant back in 1987 Brazil. -110-

Before the sequence presented in this example, Constituent João Paulo was praising the Reporter’s bill for including expressly the right of the workers to organize strikes. Constituent Samir Achôa, sensing the sensitiveness of the matter, began interrupting several turns before the sequence presented above. In this sequence both constituents assumed polarized activity roles, related to specific social identities, which are themselves tied to their discourse roles as animators of specific social discourses: the defender or the workers (João Paulo) versus the defender of the economic and industrial production (Samir Achôa). This kind of polarized direct debate appeared in the data exactly when a more sensitive issue was being addressed in a otherwise quasi-monological turn of talk, that was bound to go on as such by the unmarked structure of exchange of turns of talk, had a sensitive issue not appeared within it. This kind of debate sequence is also a marker of the hybridity of the norm-enacting activity type: although established to discuss a normative (juridical) final text, during the discussion, various sequences assimilated to direct political debates emerge that resemble other kind of discourse activities within parliaments (plenary speeches, for one) and other settings (political talkshows, for instance). 5. Final and concluding remarks The foremost aim of this paper was to show that not only norm-enacting activity within parliamentary contexts pertain, as an object of study, to the field of forensic linguistics, but also that the same methods used to analyse and uncover a variety of features of various types of activity within the juridical (social) phenomenon could and should be applied to norm-enacting activity. Mapping the particular configuration of norm-enacting activity, in terms of activity framing, participant structure, communicative goals mobilized by participants, roles and identities, talk-exchange structure, facework and so on (Sarangi, 2000) help establish the kinds of particular inference procedures and schemata used by participants, as well as the particular structural constraints on what counts as a valid move inside this particular (language) game (Levinson, [1978] 1998). By uncovering these particulars of the practical reasoning (Garfinkel, 1967) involved in the conception, reproduction and accounting of the activity, we are also able, on a more practical note, to facilitate and further the comprehension of a host of pressing theoretical and practical questions, especially inside the legal domain itself. Just to mention a few, by focusing on norm-enacting activity and its (discursive) tenets, we can help clarify: a) The issue of the so-called ‘legislative intent’ (Solan, 2004; Waldron, 2004). Still a very pressing issue in the Law debate (and an issue that also touches other very important questions such as the separation of powers, democracy, and law interpretation), the question of legislative intent discusses, in simplified terms, whether legal norms (statutes) have a retrievable (meaning) intention behind the written normative text itself, that could be invoked or discovered when there is some apparently undecidable interpreting issue to be resolved by the courts. So far, lawyers and jurists are just discussing if there is a legislative intention from a more theoretical point of view, even when some linguistic insights are brought to the fore (Solan 2004; Struchiner, in press; Tiersma 2001). There is also no questioning of how this legislative intent could be retrieved, should there be one (but see Tiersma, 2001: 438). Analysing actual data from norm-enacting activities can help answer both questions: if there is anything resembling an intent being constructed by the interactions that end up in a particular normative text enacted and, if so, how can this intent be found (or retrieved) after that normative text is enacted and becomes law in its own right. -111-

b) The issue of the difference between norm creating and norm applying (and/or interpreting). It is generally accepted in legal theory and practice that there is some kind of qualitative difference between the activity of creating a norm and the activity of applying a norm (Habermas, 2001: 168ff., who even proposed the concepts of ‘discourses of justification’ and ‘discourses of application’). However, while the application of the laws (through the courts and legal processes) has been a central focus of attention and research within the Law, the same cannot be said about Law creation. We still do not know exactly if and in what respects the legislative process is different from the judicial process, especially the judicial processes of appeal courts, that do not directly involve themselves with fact-finding and evidence gathering. This issue is not only important for the understanding of interpretation and application of laws, but also has been a central point of argument between defenders and opponents of judicial review (Tate and Vallinder, 1995; Waldron, 2006). c) Finally, the issue of legislative drafting. Much of the work carried out on the subject is dedicated to the end product of norm-enacting activity and aims at trying to describe textual patterns and cognitive properties associated with them (Bhatia, 1994; Maley, 1994). However, without knowing the interaction processes and contexts (and, therefore, activities) involved in arriving at those texts, there is a big knowledge gap in the object of study and the length of the conclusions that the research can reach. Also, initiatives such as the plain language movement (e. g., Tiersma, [1999] 2000) seem to lack this very knowledge of the (discursive) process of bringing about a legal norm, and, for that reason, tend to get stuck at proposed solutions such as lexical choice and syntax construction improvement, that are not only partial, but of unknown applicability in reality—compare how the efforts about jury instructions improvement are far more concrete, since we do know a lot about how these instructions are conceived and brought about in practical activities of trials (Tiersma, [1999] 2000 and references therein). This issue also relates to advocacy and influencing of public policy, a very important issue for democracy (see, from a theoretical standpoint, Habermas, 2000). It should be clear, then, that the field of forensic linguistics can both profit from the study of norm-enacting activity and also that it already has the means and methods to do so. Moreover, it should be clear that, by doing this kind of study, forensic linguists can help to deal more adequately with a series of very important issues that draw on discursive aspects of the normenacting activity, but which also have pressing consequences for both juridical and political domains. Being an interdisciplinary field in itself, it seems that this kind of work is what forensic linguistics is trying to accomplish. References Bhatia, V. (1994) Cognitive structuring in legislative provisions. In J. Gibbons (ed.) Language and the Law. London and New York: Longman, 136—155. Bobbio, N. ([1985] 2000) Estado, poder e governo [State, power and government]. In N. Bobbio. Estado, governo e sociedade: Para uma teoria geral da política [State, government and society: Towards a general theory of politics]. Transl. Marco Aurélio Nogueira. São Paulo: Paz e Terra. Coulthard, M. & Johnson, A. (2007) An Introduction to Forensic Linguistics: Language in Evidence. New York: Routledge. -112-

Coulthard, M. & Johnson, A. (2010) Introduction: current debates in forensic linguistics. In M. Coulthard and A. Johnson (eds). The Routledge Handbook of Forensic Linguistics. London and New York: Routledge. (Kindle® edition) David. R. (1998) Os grandes sistemas do direito contemporâneo [The big systems of contemporary Law]. Transl. Hermínio A. Carvalho. São Paulo: Martins Fontes. Drew, P. and Heritage, J. (1998) Analyzing talk at work: an introduction. In P. Drew and J. Heritage (eds) Talk at work. New York: Cambridge University, 3—65. Gales, T. (2009) ‘Diversity’ as enacted in US immigration politics and law: a corpus-based approach. Discourse and Society 20(2): 223—240. Garfinkel, H. (1967) What is ethnomethodology? In H. Garfinkel (1967) Studies in ethnomethodology. Engelwood Cliffs: Prentice-Hall, 1-34. Kaplan, J. P. & Green, G. M. (1995) Grammar and inferences of rationality in interpreting the child pornography statute. Washington University Law Quarterly 73(3): 1223—51. Goffman, E. ([1974] 1986). Frame analysis. Lebanon, NH: Northeastern University. Goffman, E. ([1979] 2002) Footing. In B. T. Ribeiro and P. M. Garcez (eds) Sociolingüística interacional [Interactional Sociolinguistics]. São Paulo: Loyola, 107—48. Goffman, E. (1983) Felicity´s condition. The American Journal of Sociology 89: 1—53. Habermas, J. ([1998] 2000) La constelación posnacional y el futuro de la democracia [The postnational constelation and the future of democracy]. In J. Habermas. La constelación posnacional [The post-national constellation]. Barcelona: Paidós. Habermas, J. (2001) Between Facts and Norms: Contributions to a Discourse Theory of Law and Democracy. 4th Printing. Trans. William Rehg. Cambridge: MIT. Hanks, W. F. (2005) Explorations in the deictic field. Current Anthropology 46: 191—220. Ilie, C. (2010) Analytical perspectives on parliamentary and extra-parliamentary discourses. Journal of Pragmatics 82: 879—84. Kingdon, J. W. (1999) America: The unusual. New York: Worth. Levinson, S. ([1978] 1998) Activity types and language. In P. Drew and J. Heritage. Talk at Work: Interaction in Institutional Settings. New York: Cambridge University, 66—100. Montesquieu ([1748] 1973) Do espírito das leis [The spirit of laws]. São Paulo: Abril Cultural. Maley, Y. (1994) The language of the Law. In J. Gibbons (ed.) Language and the Law. London and New York: Longman, 11-50. Rock, F. (2001) The genesis of a witness statement. Forensic linguistics 8(2): 44—72.

-113-

Sarangi, S. (2010) Reconfiguring self/identity/status/role: the case of professional role performance in healthcare encounters. In J. Archibald and G. Garzone (eds) Actors, Identities and Roles in Professional and Academic Settings: Discursive Perspectives, 27—4. Shuy, R. (2005) Creating Language Crimes: How Law Enforcement Uses (and Misuses) Language. New York: Oxford University. Solan, L. (2004) Private language, public laws: the central role of legislative intent in statutory interpretation. Brooklin Law School Public Law and Legal Theory Research Paper Series 5: 1—82. Solan, L. & Tiersma, P. (2005) Speaking of Crime: The Language of the Criminal Justice System. Chicago and London: University of Chicago. Struchiner, N. (in press) Indeterminação e objetividade: quando o direito diz o que não queremos ouvir [Indetermination and objectivity: when Law tells what we do not want to hear]. In Direito e interpretação: Racionalidades e instituições [Law and interpretation: Rationality and Institutions]. Rio de Janeiro: Renovar. Tate, C. N. & Vallinder, T (1995) The global expansion of judicial Power: the judicialization of politics. In C. N. Tate and T. Vallinder. The global expansion of judicial power 1-10. New York and London: New York University. Tiersma, P. ([1999] 2000) Legal Language. Chicago: University of Chicago. Tiersma, P. (2001) Message in a bottle: Text, autonomy and statutory interpretation. Tulane Law Review 76(2): 431—482. Tiersma, P. (2005) Writing, text and the Law. Loyola Legal Studies Paper n. 2005—31. Retrieved on 25 July 2010 from http://ssrn.com/abstract+850305. Van Dijk, T. (2004) Text and context of parliamentary debates. In P. Bayley (ed). Cross-cultural Perspectives on Parliamentary Discourse. Amsterdam: John Benjamins, 339—372. Waldron, J. (2004) As intenções dos legisladores e a legislação não-intencional [The legislators intentions and the unintentional legislation]. In A. Marmor (ed). Direito e interpretação [Law and interpretation]. São Paulo: Martins Fontes, 495—536. Waldron, J. (2006) The core of the case against judicial review. Yale Law Journal 115: 1346— 1406.

-114-

Non-adversarial justice: the changing role of courtroom participants in an Indigenous sentencing court Natalie Stroud Faculty of Law, Monash University, Melbourne, Australia [email protected] Abstract This paper forms part of a wider sociolinguistic study of cultural and language disadvantage for Indigenous offenders in the criminal justice system. It examines the changing role of courtroom participants in an Indigenous sentencing court such as the Koori Court of Victoria, and how this may lead to a better quality of justice for this disadvantaged cultural group. The problem under review is the high percentage of Indigenous offenders who continue to come in contact with the law. The key aim of the research is to determine if issues of miscommunication identified by academics over the past three decades continue to be reflected in the court process, or whether an awareness of cultural and language difference by participants at the Koori Court hearing leads to a more restorative and therapeutic outcome for Indigenous offenders. Taking an interdisciplinary approach, the study investigates the role of participants and the communicative process in a conventional court and compares this with the non-adversarial Koori Court. There are now seven adult and two children’s Koori Courts in Victoria, established under the jurisdiction of the Magistrates’ Court of Victoria, and one County Koori Court, Australia’s only Indigenous court in a higher jurisdiction for more serious cases. This paper will illustrate how new forms of court practice lead to improved interaction between courtroom professionals and other key participants such as the offender and Indigenous Elders, in order to best achieve a positive outcome. Keywords:

NON-ADVERSARIAL SOCIOLINGUISTICS; COMMUNICATION

INDIGENOUS COURTROOM

JUSTICE; COMPARATIVE LAW; DISCOURSE; CROSS-CULTURAL

1. Introduction There continues to be a high percentage of Indigenous offenders in the Australian criminal justice system, with many reoffending throughout their life. In spite of measures aimed at lowering the recidivism rate, mainstream methods of punishment have not been particularly effective in rehabilitating a cultural group which has been seriously disadvantaged for over 200 years. This paper will examine some of the innovations designed to address this problem. According to the Australian Bureau of Statistics, the current estimated number of people identified as Indigenous Australians is 582,000, or 2.6% of the total Australian population of 22,400,000. When this is compared with the number of Indigenous offenders in the corrections system, a total of 7,584, or 26% of the total prisoner population of 29,700, it is clear that there continues to be a disproportionate representation of Indigenous offenders in the prison system (ABS cat.no.4517.0, 2010). Although Victoria has one of the lowest imprisonment rates of Indigenous (or Koori) defendants, Koories are 12 times more likely to be placed in an adult prison compared with non-Indigenous prisoners. Many Koori offenders are young, with more than half aged under 25 years, and more than a third under 15 years. Nearly 40% of Indigenous families are sole parent families, compared with 15% of nonIndigenous families. (Indigenous Affairs Report 2007—08:52). A recent change of government in Victoria has led to a shift in public and government attitude to crime, with -115-

legislation proposed to increase punitive measures for punishable offences, such as the abolition of suspended sentences and mandatory minimum sentences, together with an increased number of police and expanded prisons (Gregory, 2011:85). It is feared that this will only further increase the disadvantage experienced by this cultural minority group. The key aim of this study is to determine if issues of miscommunication identified by academics over the past three decades continue to be reflected in the court process, or whether an awareness of cultural and language difference by participants at the Koori Court hearing leads to a more restorative and therapeutic outcome for Indigenous offenders. In this regard the Koori Elders’ participation in the administration of the law is paramount. The court process and the role of courtroom participants in the mainstream County Court and Magistrates’ Court of Victoria will be examined and compared with the court process in the alternative sentencing Koori Court. 2. Context Miscommunication can occur between legal professionals and Indigenous offenders in a mainstream court when there is little understanding or awareness of cultural or language disadvantage. Difficulties may arise when an assumption is made that, because a person speaks English, they understand what is being said. The legal domain may be intimidating for any person unfamiliar with the court system, even more so for a person from an oral-based culture of group consensus. Cultural disadvantage may also lead to a breakdown in communication when the accused is unable to tell their story in their own words. When differences in pragmatic and communicative styles of speaking are not understood, the right to a fair hearing for Indigenous offenders may therefore be compromised (Eades 2008a, 2008b, 2010; Stroud 2006, 2010). Australian criminal courts of law operate within an adversarial, common law framework, maintaining observance of the rule of law, but not always with consideration of underlying cultural or social factors. In recent years, following recommendations made by the Royal Commission into Aboriginal Deaths in Custody (1991), the changing paradigm of criminal justice has led to the establishment of a number of non-adversarial Indigenous courts throughout Australia in an attempt to redress the disadvantages experienced by Indigenous offenders and reduce the rates of reoffending. In discussing the process of non-adversarial justice, King notes that this is ‘built on notions of cooperation rather than conflict’ (King, Freiberg, Batagol, and Hyams 2009:15). The Koori Court is an Indigenous sentencing court, with the same powers as the mainstream court. It was first established in 2002 under the Magistrates’ Court (Koori Court) Act 2002. There are now seven adult Koori Magistrates’ Courts operating throughout Victoria, and two Children’s Koori Courts for 10—17 year olds, established under the Children and Young Persons (Koori Court) Act 2004. Following the success of these courts, the County Koori Court was established under the County Court Amendment (Koori Court) Act 2008. This court began operations in regional Victoria in 2008, and is Australia’s first sentencing court for Aboriginal offenders in a higher jurisdiction. The Koori Court model, while similar in many respects to other specialist Indigenous courts in operation throughout Australia, such as the Nunga Court in South Australia, the Murri Court in Queensland, and Circle Sentencing Court in New South Wales, is unique to Victoria, incorporating the best features of existing models. It is designed to be non-adversarial in nature, and has a ‘solutionfocused’ approach to the court process, with broad aspects of therapeutic and restorative justice, in order to address any underlying issues behind the offence and bring about behavioural change and rehabilitation of the offender. The Koori Court program was initiated by the Victorian Government in partnership with the Indigenous community under the -116-

Victorian Aboriginal Justice Agreement (AJA2, 2005). Prior to the establishment of each court, a key factor in the success of the program was the involvement of the local Indigenous community in partnership with the government, in assessing the needs of the community and ensuring appropriate support services were in place. To be eligible for a case to be heard in the Koori Court a person must be Aboriginal or Torres Strait Islander, plead guilty to the offence and be willing to take responsibility for their actions. Sexual assault and breaches of family violence orders are not heard in this court. 3. A comparison of mainstream courts with the Koori Court of Victoria Using a descriptive approach, this qualitative study draws on comparative law and interactional sociolinguistics as informed by the interpretive approach to social interaction of Gumperz (1982) and participation framework of Goffman (1981). Principles of restorative justice are informed by Braithwaite (2002) and therapeutic jurisprudence by Winick and Wexler (2003). Initial observational case studies conducted at selected courts reveal some interesting patterns in the communicative process. Although detailed data is yet to be collected and evaluated, similarities and differences in courtroom formality, the role of participants and the language used in the courtroom are examined in order to understand how new ways of thinking about communication, and new approaches in court process may have an impact on justice for Indigenous Australians. 3.1. Formality A wide variation in formality was found across the mainstream spectrum. In the County Court, the courtroom design is formal, with the raised Judge’s bench a marker of power and prestige, and all legal professionals formally attired in wigs and gowns, similar to those worn in the United Kingdom. The jury box, witness’ stand and defendant’s box are located in separate areas of the courtroom, strategically placed for maximum eye contact between participants. Prosecuting and defence counsel are seated at the bar table facing the Judge with instructing solicitors seated opposite. The Tipstaff in charge of court proceedings, together with the Judge’s associate, court registrar and police presence complete a rather intimidating picture. The public are non-participatory observers. All present must stand and bow when the Judge enters or leaves the court. The language of the court is formal with honorifics used between speakers. The Magistrates’ Court is less formal than the County Court, and wigs and robes have been dispensed with. However, the process of the court remains similar to the County Court and follows the rule of law, with the Magistrate seated at the raised judicial bench, counsel at the bar table and defendant in the dock with security guards. There is no jury. This court deals with all summary offences and some indictable offences. This is a very busy court (with one Magistrate scheduled to hear 80—90 cases in a day), with visible security and armed police. The defendant is represented by their lawyer and is largely silent. Some judicial officers prefer more formality than others, and some are more culturally aware than others. The Koori Court is a division of both the County Court and Magistrates’ Court, and variation in formality in the Koori Court extends across both jurisdictions. All Koori Courts are designed in a culturally appropriate manner, with three flags standing at the front of the court, Aboriginal, Australian and Torres Strait Islander, emphasizing the cross-cultural nature of the court. Aboriginal art is displayed on the walls. Koori Courts are informal, and no wigs or gowns are worn during the sentence hearing. At the start of all hearings, the judicial officer pays respect to the original inhabitants of the land on which the court stands, also Elders past and present. -117-

In the County Koori Court, the courtroom is dominated by a large elliptical table, not only a work of art, but an example of Koori communities working together, as the 10,000 year old ancient red gum timber was brought from Yorta Yorta country in northern Victoria and crafted with care in the Ganai-Kurnai country in Eastern Victoria, for use in the Morwell County Koori Court. The Judge and all participants are seated around this table for the ‘sentencing conversation’, with the Indigenous Elder and Respected Person (called Aunty or Uncle as a sign of respect) seated either side of the Judge. The offender sits with a family member or support person directly opposite the Judge and Indigenous community Elders. The Defence Lawyer, Prosecuting Lawyer, Koori Court Officer and Corrections Officer complete the full circle. Family and friends are seated close by in the public area of the court. Although the County Koori Court is less formal than the mainstream County Court, some parts of the court process are similar, with hearings conducted in three stages. Stage 1 is the formal arraignment with the defendant agreeing to a guilty plea. The Elders are not present at this stage. Stage 2 is the ‘Sentencing Conversation’, which is informal and interactive, with all participants including the Elders seated around the oval table. Stage 3, usually conducted on another day, is more formal, with the Elders returning to the back of the court in order to show that the sentence is the decision of the Judge alone. The Koori Magistrates’ Court is less formal again, with the Magistrate and all participants seated around the oval table, and family and support persons seated close by in the body of the court. The Magistrate greets the offender informally by their first name, and introduces them to all participants. Proceedings are explained to the offender in plain English without legal jargon, with the Magistrate checking for understanding throughout the hearing. The Elders participate in the process, and contribute cultural knowledge. The offender is often known to the Indigenous community Elders and many in the courtroom. Any person in the body of the court may contribute to the conversation at an appropriate point. 3.2. Role of courtroom participants There is wide variation in the role of participants in all courts reviewed, the most marked differences lying in their interaction or lack of interaction in the court process. The Judge in the formal mainstream County Court is an impartial observer, hearing matters from the bench and ensuring proper procedures are followed prior to consideration and sentencing. The defendant is mostly silent and speaks through the defence lawyer. Participants in the mainstream Magistrates’ Court follow the above proceedings, but in a less formal manner, with the Magistrate, similar to the Judge, having an impartial role. Marchetti and Daly (2007:440) note that ‘key people who are normally involved in court hearings, such as defence lawyers, prosecutors and judges, are not necessarily the only people who should be determining what happens to offenders’. In the non-adversarial Koori Court, all participants are interactive in the court process, and the addition of the Indigenous Elder and Respected Person has a marked effect on both the interaction in the court process and the relationship between the justice system and the Indigenous community. The Koori Court Officer liaises with all parties involved in the hearing, and it is interesting to observe the Police Prosecutor working with the Defence Lawyer and all at the table to reach a positive outcome for the offender. This contrasts with the combative style of the adversarial system. Other active participants are the Corrections Officer, Support Services, family members, community members and general public. Any person in the court may add to the discussion. According to Marchetti and Daly (2007:440), these additional participants at the court hearing change the focus of sentencing ‘to one which is more negotiated, rehabilitative or reconciliatory’. The Koori Court redefines the court process by reframing the role of participants in the courtroom. Participants become much more interactive and cooperative, in order to carry -118-

out the philosophical aims of therapeutic jurisprudence and restorative justice, to bring about behavioural change in the offender and restore balance in the community. The inclusion of the Indigenous Elder and Respected Person in the administration of the law is a significant difference to the mainstream court. The Koori Court Officer liaises with all participants, and the presence of a family member or support person also changes the dynamics of the legal process. The Koori Court’s broad interpretation of the principles of restorative and therapeutic justice is both relevant and successful in changing behaviour and restoring balance in the life of the offender and the community. The judicial officer is aware of the therapeutic value of rehabilitation, and support groups take an active role in courtroom proceedings and follow up after sentencing. The offender must demonstrate willingness to change their behaviour and attend rehabilitation programs if required. For a successful outcome, the judicial officer is then able to determine the appropriate sentence with consideration to underlying cultural and social factors. In spite of there being no specified role for the victim in the Koori Court, they are welcome at the hearing, and any impact statements are taken into account prior to sentencing. Three of the main roles taken by courtroom participants illustrate the variation in practice between the adversarial mainstream court and the alternative sentencing Koori Court - the Judicial role, the Elders’ role and the role of the offender. 3.2.1. Judicial role Judicial officers in all courts must be fair, impartial, and follow the rule of law. According to King (2010:133), in a non-adversarial courtroom the judicial role changes from a neutral, largely uninvolved umpire seeking to ensure the fairness of a process mainly conducted by the parties, to a collaborative process and increased interaction between the judicial officer, participants, court team members and community members, informed by therapeutic jurisprudence principles. Some specially selected Judges and Magistrates preside in both the Koori Court and mainstream courts, resulting in a ‘cross-over’ of culturally aware practices. Organizations such as the Judicial College of Victoria, Australian Institute of Judicial Administration and the National Judicial College of Australia, run cultural awareness programs, workshops and visits to prisons for the judiciary and legal professionals, and these are generally well attended. One problem in the mainstream Magistrates’ Court is that, in spite of Magistrates who are sympathetic to cultural issues, the high number of cases heard per day prohibits time being taken to observe any language and cultural difficulties that may arise. In the Koori Court however, the lower number of cases heard in a day allows time for the judicial officer to hear from all participants and address some of the underlying social problems. Some detractors have criticized judicial officers for being too ‘soft’ and label therapeutic judging as ‘social work’. It is clear, however, that judicial officers who apply principles of therapeutic jurisprudence in their court find that an approach that ‘respects, empowers and involves the participant is more effective than a paternalistic response’ (King et al., 2009:211). The outcome of one hearing observed by the researcher illustrates the success of this approach. Case Study 1 The accused, a young man in his twenties, on a charge of alcohol related assault, entered the courtroom with a confident stride and sat at the table opposite the Judicial Officer and the Elders. By the time the Elders had spoken about the shame he had caused his family and the community because of his alcohol and drug problems, his body language had changed. As he told his story, the court heard of a life coping with disability, low literacy skills, crowded -119-

housing and sadness at the death of his father. He showed remorse at his actions, and shame when thinking of what his father would have thought of him. He vowed to turn his life around. The court then heard from participants at the hearing, such as the Koori Court Officer, the Drug and Alcohol Counsellor and family members, and a decision was made regarding available courses the offender could attend, with a work component and follow up support services. He was given a deferred sentence so that he could attend rehabilitation, with the instruction that he would have to serve the full sentence if he got into trouble again. The Judicial Officer concluded the case by acknowledging the collaborative input of participants and noting that the absent victim should also be remembered. 3.2.2. Elders’ participatory role There is no designated role for Indigenous Elders or community members in the mainstream court, although the Magistrates’ Court Integrated Services Program (CISP) includes a Koori Liaison Officer to work with Koori defendants. In the Koori Court, however, the participation of Elders at the court hearing is one of the main tenets of the court process. Elders and Respected Persons often have personal knowledge of the offender. This is especially so in regional community areas. In the busy urban Koori courts, the Elders may not know the background of the offender but are able to show the offender that their behaviour brings shame to the community. While expressing horror at the offence, the Elder often shares their own story of disadvantage and heartbreak, and offers encouragement and hope to the offender that the court and community is there to help. Case Study 2 At one hearing, the Elder spoke directly to the offender, saying ‘look at me and give me respect’. She did this until he met her eye gaze, when she then expressed her disapproval of his bad behaviour. She told him he had shamed his community and now must take responsibility, turn his life around, return to study and be proud of and respect his cultural heritage. Case Study 3 At another day of hearings at a regional court, case after case was adjourned due to absence of each offender. The Elders were able to impart cross-cultural knowledge to the Magistrate regarding the importance of kinship obligation, which meant that the offenders were obliged to attend the funeral that day of a member of the local Indigenous community. Without this knowledge, they would be recorded as ‘failed to appear’ and receive a penalty. The Elders’ role is a vital one for the success of the courts, and it is important that they receive support from the Court. Although they are held in great regard and have the highest respect of their community, many of them have poor literacy and numeracy skills. The Magistrate spends time with the Elder and Respected Person prior to the hearing outlining the background of each case. 3.2.3. Offender’s Participatory Role One of the main variations between the role of offender in mainstream courts and Koori Courts is in their participation and interaction in the court process. Whereas in the mainstream courtroom, the offender is largely silent and only speaks through their lawyer, in the Koori Court they are encouraged to participate and are given the time to tell their story. -120-

They have a ‘voice’ in the court process. It is quite confronting for them to sit directly opposite the Judge or Magistrate and respected Elders and be told that they have let down their community by their actions. If one of their underlying problems is alcohol abuse, drugs, or anger management, they are referred to support services and given the opportunity to enter a rehabilitation program. The offender may also be a victim, and Elders of the Indigenous community have a positive impact in helping to break the cycle and stop the ‘revolving-door’ effect of reoffending. Case Study 4 A young woman with a long standing drug and alcohol dependency came before the court charged with burglary and theft. The Elders spoke to her about her future and the care of her children. The Koori Court Officer linked her with a Koori drug and alcohol worker, who arranged follow up visits with her at home. She was given a suspended sentence, and came back to court every two months for 12 months so that the Elders could monitor her progress. At the end of this time she was drug and alcohol free, coping well and a devoted mother to her children. The Magistrate took the view that a suspended sentence of imprisonment was appropriate (Magistrates’ Court of Victoria Annual Report, 2007-08:55). 3.3. Courtroom Language Language in the mainstream courtroom is formal, specialized, structured and precise, with complex terminology. There are certain norms and court practices which must be followed. This is an adversarial court which uses a question/answer communicative style. Indigenous participants in a mainstream court may ‘appear’ competent in the language, but cultural differences may lead to a breakdown in communication between speaker and hearer and misunderstandings can occur in proceedings. Honorifics used by legal counsel emphasize cultural and social distance and are markers of power and prestige, creating further disadvantage for an Indigenous speaker who is already less powerful. In mainstream adversarial courts the legal counsel dominate the courtroom discourse. The use of complex legal terminology and possible manipulation of language by some lawyers may cause misunderstandings and confusion for a defendant. In addition to assumptions of language competence there may be other unrecognized cross-cultural differences. Different pragmatic and communicative styles of speaking may also cause miscommunication, such as differences in speaking about time and place, inability to cope with the question/answer format or gratuitous concurrence by the defendant, regardless of understanding. 3.3.1. Specific linguistic features addressed in the Koori Court In comparison, one of the main differences in the Koori Court is in the interactive role of participants in the ‘sentencing conversation’. Auty (2006:118—119) observes that the role of legal professionals in the court process is not as prominent in this court. The legal environment is modified, and all participants, including the offender, are given respect and encouraged to speak during the hearing. Indigenous speakers come from a rich cultural heritage of oral languages which involve a different way of thinking, and this is respected. Cultural and language heritage remains embedded in their discourse, with variation in politeness, taboos and different terminology used for specific occasions and kinship relationships. Time and place are described in different ways. The use of plain English, together with an awareness by legal professionals that if certain pragmatic, semantic or -121-

syntactic features are avoided (such as tag questions, question/answer, specialized legal jargon), communication is improved between all participants. In this court there is recognition of the importance of silence and the use of the Indigenous communicative style in the discourse. The offender has time to tell their story instead of speaking through their lawyer. They are part of the process. Language accommodation appears to be both ways between speakers in this court, rather than one way towards the more powerful speaker. Turn-taking of speakers is quite lengthy compared with a more formal court with rigid control of discourse. The tenor of the discourse is informal, and speaker/hearer inconsistencies are kept to a minimum. One interesting feature in the Swan Hill Koori Court is its ‘Wamba Wamba Language Initiative’, launched in July 2009. The local Aboriginal Wamba Wamba language is spoken by the Elders and Respected Persons at the opening and closing of each matter heard. This is a significant innovation, and an example of the court and Indigenous community working together. (Magistrates’ Court of Victoria Annual Report, 2009—10:49). 4. Discussion Since the establishment of the first Koori sentencing court in 2002, many innovative partnership, education and support programs have been put in place by the Victorian Government together with the Courts and Aboriginal community. The aim of these programs is to deal with underlying problems, rehabilitate offenders following their court hearing, and reduce reoffending. Some collaborative and innovative programs include diversionary programs for Indigenous offenders; the Court Integrated Services Program (CISP), a support service which addresses individual needs such as drug and alcohol addiction to reduce the chance of reoffending; and the Credit/Bail Support (CBS) for early intervention. For young Koories there are education and training programs, mentoring programs and apprenticeship and job assistance. Many of these programs have had positive results in addressing the longterm needs of Indigenous Australians. One illustration of this is the Wulgunggo Ngalu Learning Place, a voluntary residential program built on the site of the old Won Wron Prison in Gippsland for young male adults placed on Community Based Orders (CBO’s). This is a long term solution to recidivism where they can learn a trade or land management with cultural input such as the making of boomerangs and digeridoos. Cultural awareness programs and workshops are run every year by both the judicial colleges and the courts for the judiciary and legal professionals, with support programs for Elders to enhance dialogue. The Neighbourhood Justice Centre, established in 2007 as a community court, delivers therapeutic and restorative justice and provides alternative dispute resolution and crime prevention. The aim of the Koori Court is for a ‘solution-focused’ outcome for Indigenous offenders, at the same time delivering greater fairness to Koori offenders involved in the criminal justice system. The Koori Court program is consistent with the Victorian Charter of Human Rights and Responsibilities (2006), as the Charter requires that ‘a person charged with a criminal offence has the right to a fair hearing’. The Koori Court is not a ‘soft’ option or ‘apartheid’ justice. Rather, it represents an attempt at a more holistic approach, which considers the wider picture and involves the entire community in the legal process. In this court the legal system is adapted with greater focus on therapeutic and restorative justice. The informal court hearing enables increased interaction and cultural and language awareness and understanding of participants. The inclusion of community Elders changes the dynamics of the courtroom. The offender is given a ‘voice’ in proceedings.

-122-

Evaluations of the Koori Courts have shown that this specialist court is a culturally appropriate and effective sentencing option of reducing recidivism in Indigenous offenders. In his evaluation of the Koori Court pilot program 2002—2004, Harris (2006) reported a reduction in the level of recidivism and a ‘high level of support for the Koori Court model’. Although Harris’ findings were later queried by Marchetti and Daly (2007) and Fitzgerald (2008), they remain a comprehensive evaluation of the first years of the Koori Court program, and many of his 19 recommendations have since been implemented. An evaluation of the Children’s Koori Court was carried out by Borowski (2009), and an evaluation of the County Koori Court is nearing completion. The Sentencing Advisory Council has also released a statistical report of the Koori Courts (2010). All reports highlight the importance of the active involvement of Elders and Respected Persons and other members of the Indigenous community in the hearing process, but note that there is no ‘quick fix’ to the problem of overrepresentation of Koories in the justice system. Statistics do not reflect the steady improvement in the lives of many defendants who have passed through the Koori Court, and should be measured in decades rather than in years. It is important not to view communication difficulties solely through a linguistic lens. Many other factors such as systemic disadvantage; the social imbalance of power; the need for more programs to improve literacy and numeracy; better housing; improved mental health and drug and alcohol addiction programs; are all necessary to address and improve the lives of Indigenous offenders and reduce recidivism. These are the human elements that must be taken into consideration if disadvantage is to be addressed and a reduction in crime is to be achieved. However change takes time and it may take generational change to redress the imbalance of opportunity experienced by this minority group. The vision for Koori Justice in Victoria, under the Koori Justice Agreement, is to continue to raise cultural and language awareness and restore balance in the community. The interactive role of all courtroom participants in the specialist Koori Court leads to improved communication and brings about a better quality of justice for Indigenous offenders. In this regard the Koori Elders’ participation in the administration of the law is paramount. 5. Conclusion The Koori Court model is an effective method of dispute resolution, incorporating the philosophy of restorative and therapeutic justice designed to achieve lasting outcomes for the wider community, the Indigenous community and the Aboriginal offender. These courts have effectively lowered recidivism as they engage in culturally appropriate forms of communication and adapt the formal court roles accordingly. It is pleasing to note that some specially selected Judges and Magistrates preside in both the Koori Courts and mainstream courts, resulting in a ‘cross-over’ of culturally aware practices. Education appears to be the key in changing the dynamics of the courtroom. It is hoped that the excellent professional development programs in cultural awareness run by the judicial colleges continue, together with education programs for lawyers and court staff; professional development for Elders; and that non-adversarial components of therapeutic and restorative justice are included in the curriculum for all law students, as it is at Monash University. The model of this innovative alternative sentencing court may be applied in the future to other specialist courts. Both adversarial and non-adversarial practices have their place in the criminal justice system, and it may be that real change will only come for Indigenous Australians when there is a community awareness and acceptance of difference, and mutual respect between cultures.

-123-

References Australian Bureau of Statistics. (2010) Prisoners in Australia, 2010. Cat No.4517.0. http://www.abs.gov.au/ausstats/[email protected]/Products/902DD677C4113895CA2577F300 Auty, K. (2006) We teach all hearts to break—but can we mend them? Therapeutic Jurisprudence and Aboriginal Sentencing Courts. eLaw Journal (special series)1: 101—128 https://elaw.murdoch.edu.au/archives/special_series.html. Borowski, A. (2009) Courtroom 7: An Evaluation of the Children's Koori Court of Victoria. Melbourne: La Trobe University. Braithwaite, J. (2002) Restorative Justice and Responsive Regulation. New York: Oxford University Press. Department of Justice (1991) Royal Commission into Aboriginal Deaths in Custody. National Report 5: 31. Eades, D. (2008a) Courtroom Talk and Neocolonial Control. Berlin: Mouton de Gruyter. Eades, D. (2008b) Language and disadvantage before the law. In J. Gibbons and M. T. Turell (eds) Dimensions of Forensic Linguistics. Amsterdam: John Benjamins. Eades, D. (2010) Sociolinguistics and the Legal Process. Bristol: Multilingual Matters. Fitzgerald, E. (2008) Does circle sentencing reduce Aboriginal offending? Crime and Justice Bulletin: Contemporary Issues in Crime and Justice 115. Goffman, E. (1981) Forms of Talk. Philadelphia: University of Pennsylvania Press. Gregory, J. (2011) Stepping out of the shadow. Law Institute Journal 85(3): 18. Gumperz, J. (ed.) (1982) Discourse Strategies. Cambridge: Cambridge University Press. Harris, M. (2006) “A sentencing conversation”: evaluation of the Koori Courts pilot program October 2002—October 2004. Melbourne: Department of Justice. King, M., Freiberg, A., Batagol, B., and Hyams, R. (2009) Non-Adversarial Justice. Annandale, NSW: The Federation Press. King, M. (2010) Judging, judicial values and judicial conduct in problem-solving courts, Indigenous sentencing courts and mainstream courts. Journal of Judicial Administration 19 (3): 133. Marchetti, E., and Daly, K. (2007) Indigenous sentencing courts: towards a theoretical and jurisprudential model. Sydney Law Review 29(3): 415—443. Magistrates’ Court of Victoria 2007—2008 Annual Report. Melbourne: Magistrates’ Court of Victoria.

-124-

Magistrates’ Court of Victoria 2009—2010 Annual Report. Melbourne: Magistrates’ Court of Victoria. Sentencing Advisory Council. (2010) Sentencing in the Koori Court Division of the Magistrates’ Court: A Statistical Report. Melbourne: Sentencing Advisory Council. Stroud, N. (2006) Accommodating language difference: a collaborative approach to justice in the Koori Court of Victoria. In K. Allan (ed.) Selected Papers from the 2005 Conference of the Australian Linguistic Society, Melbourne. http://www.als.asn.au/proceedings/als2005/stroud-koori.pdf Stroud, N. (2010) The Koori Court revisited: a review of cultural and language awareness in the administration of justice. Australian Law Librarian 18 (3): 184—192. Victorian Aboriginal Justice Agreement Phase 2 (AJA2) (2005). Melbourne: Victorian Department of Justice. Victorian Government Indigenous Affairs Report 2007/08. Melbourne: Aboriginal Affairs Victoria 2007—08: 52. Winick, B and Wexler, D (eds) (2003) Judging in a Therapeutic Key: Therapeutic Jurisprudence and the Courts. Durham: Carolina Academic Press. Legislation Magistrates’ Court (Koori Court) Act 2002 Children and Young Persons (Koori Court) Act 2004 County Court Amendment (Koori Court) Act 2008 Charter of Human Rights and Responsibilities Act (Vic) 2006

-125-

Sincere and deceptive statements in Italian criminal proceedings Tommsso Fornaciari1 and Massimo Poesio2 University of Trento, Italy 1 [email protected] 2 [email protected] Abstract Identifying false or deceptive statements in testimonies is a difficult challenge in criminal proceedings because it is not a task humans find easy. Text classification techniques have shown promise at this task—but so far, they have mainly been tested with laboratory produced data rather than authentic, real life data. We collected what is the first Italian corpus of hearings from criminal proceedings in which the defendant was found guilty of false testimony. In such cases, the transcriptions of each hearing report the words exactly as told by the subjects, and the judgment points out the statements found by the Court to be false. This characteristic makes it possible to annotate sincerity and deception of statements in such data on the basis of unusually solid objective information. We used these data to train models to classify statements as sincere or deceptive, showing that in spite of the difficulty humans have at this classification task, it is possible to obtain a performance well above chance level from automatic classifiers using very simple surface linguistic features. Keywords:

FORENSIC LINGUISTICS; DECEPTION DETECTION; TESTIMONY IN COURT; TEXT CLASSIFICATION

1. Introduction 1.1. Detecting deception Identifying deceptive statements in testimonies could provide very useful support to investigative work, particularly when other kinds of evidence are scarce or absent. In spite of this, modern studies demonstrate that human performance in recognizing deception is not much better than chance (Bond and De Paulo, 2006). Furthermore, in some studies human skills seem to be not particularly improved even after specific training (Levine, Feeley, McCornack, Hughes, and Harms, 2005). Other studies instead try to demonstrate that the ability of humans as lie-detectors is underestimated (Frank and Feeley, 2003). In any case, even in papers in which positive effects of training are found, the difficulty of the task is openly recognized (Porter, Woodworth, and Birt, 2000). Probably the difficulty in recognizing deceptive statements has led to the development of a wide variety of approaches to discover deceptive statements. They can be very different from each other, but all of them typically involve two steps: • To identify some clues of deceptiveness in the communicative act; • To verify if the statements held as false are actually false. The choice of clues to be considered in the analysis determines the strategy in trying to detect deception. Several authors try to put together different analysis techniques, hoping to optimize the accuracy in detecting falsehoods. This is the case with De Paulo, Lindsay, Malone, Muhlenbruck, Charlton, and Cooper (2003), who consider more than 150 cues, verbal and non-verbal, directly observed through experimental subjects. Also Jensen, -126-

Meservy, Burgoon, and Nunamaker (2010) recently focused on cues coming from audio, video and textual data, with the aim of building a paradigm useful to identify deceptiveness. 1.2. Stylometry With the contribution of modern linguistics and psychology, the analysis of language has become increasingly effective and has been applied to specific aspects of the discourse. In recent years stylometric methods which typically analyse linguistic style in text through statistical techniques, have in fact been demonstrated to be effective in several forensic tasks, such as author profiling (Coulthard, 2004; Solan and Tiersma, 2004), including deducing the age and sex of authors of written texts (Koppel, Schler, Argamon, and Pennebaker, 2006), author attribution (Luyckx and Daelemans, 2008; Mosteller and Wallace, 1964) and plagiarism analysis (Stein, Koppel, and Stamatatos, 2007). Stylometry is also becoming more and more important in Deceptive Language Analysis. Stylistic features have emerged as useful markers to evaluate the truthfulness of the speakers (or writers). A lot of studies have been carried out following this path (for example Porter and Yuille, 1996), in a variety of contexts. For example Adams (1996), working in the context of Police Forces, asserted the necessity to take into account the personal style of communication together with the content of the testimonies. In Italy Anolli, Balconi, and Ciceri, (1999), working on Italian linguistic data, tried to identify styles of communication which are specific to deceptive language. 1.3. Deceptive language analysis 1.3.1. Field and laboratory studies Regarding deceptive language, the existing papers can be roughly divided in two main families: field studies and laboratory studies. Field studies, such as those using Criteria Based Content Analysis (Vrij, 2005), one of the foremost techniques for the evaluation of children's statements in cases of suspected sexual abuse, are interesting for forensic practice but, as noticed by Vrij himself (2005), it is often difficult to verify the sincerity of the statements. Typically, in practical cases the content of the testimonies themselves and non-verbal cues play an important role in the assessment of sincerity. Such methods of research are quite different from those employed in our paper, which relies instead on stylometric analyses. Laboratory studies (Newman, Pennebaker, Berry, and Richards, 2003), on the other hand, focus on mock lies, produced by experimental subjects under laboratory conditions. These studies result in the creation of balanced data sets that typically allow stylometric analyses through machine learning algorithms. Nevertheless, the artificiality of the conditions means that the findings of such studies may not be applicable to real life cases. As Koppel et al. (2006) point out, the features used in stylometric analysis belong to two main families: • surface-related features; and • content-related features. The first type of feature includes the frequency and use of function words or of certain ngrams of words or part-of-speech (POS). The second kind of feature specifies information about the semantic content of words, accessed from dictionaries and lexical resources.

-127-

1.3.2. The Linguistic Inquiry and Word Count (LIWC) Perhaps the best-known lexical resource for deception detection is the Linguistic Inquiry and Word Count (LIWC), created by Pennebaker, Francis, and Booth (2001). This was applied, among other things, to the evaluation of deceptive language. For example Newman et al. (2003) reached an overall accuracy of 60% in classifying deceptive vs. truthful texts. In addition LIWC has been employed in studies on deceptive language carried out by other groups, such as the work by Strapparava and Mihalcea (2009), who obtained results similar to Newman et al. (2003) at classifying into ‘sincere’ or ‘deceptive’ texts collected with the Amazon Mechanical Turk service. Strapparava and Mihalcea actually used surface features in order to classify their texts, but also used the LIWC, even if for post-hoc analysis only, to measure several language dimensions, such as positive or negative emotions, cognitive processes, and so on. In this way, they were able to identify some distinctive characteristics of deceptive texts. Moreover, the opportunity to work with data in electronic format, and the increasing relevance of Computer Mediated Communication, has contributed to an increase of studies in which deceptiveness is produced through the use of computers: Hancock, Curry, Goorha, and Woodworth (2008), for example, employed LIWC for research about dyadic communication in a synchronous text-based setting. Making use of different variables, Zhou carried out an analogous study of both synchronous (2005), and asynchronous (Zhou, Burgoon, Nunamaker, and Twitchell, 2004) Computer Mediated Communication. 1.4. Our research Our paper also aims to develop machine learning models of deception detection. However, we aim also to fill a research gap identified most recently by Zhou, Shi, and Zhang (2008), who highlighted the lack of ‘data sets for evaluating deception detection models’ (p. 1078). Our goal is to contribute to research knowledge by analyzing transcriptions of false and true testimonies presented during Court hearings and to distinguish true testimony from false on the basis of stylometric differences. The theoretical assumption on which this paper is based, historically known as the Undeutsch hypothesis (1967), is that the cognitive elaboration of untruthful statements differs from the elaboration of truthful ones, so that differences should be traceable in the features of the statements themselves. In order to study this hypothesis it is necessary, on one hand, to collect testimonies containing real life linguistic data; and on the other, to know with certainty if statements are sincere or deceptive. There currently exists a lack of research in which both of these prerequisites are satisfactorily met. With respect to the kind of data collected, the two studies of Fitzpatrick's group (Bachenko, Fitzpatrick, and Schonwetter, 2008; Fitzpatrick and Bachenko, 2009) are the most similar to our research activities. They collected a corpus of criminal statements, police interrogations, and civil testimony. On the other hand, as deception cues to analyze their data they choose several ‘linguistic phenomena’ such as preference for negative expressions in word choice, inconsistencies between verb and noun forms and so on, and their texts were annotated manually on the basis of these ‘phenomena’. They obtained accuracy close to 75% in the classification task. We believe we are the first to apply a stylometric approach to Italian language to detect deception. The structure of the paper is as follows: In Section 2 we discuss the method used to collect the data for our study. In Section 3 we discuss the methods used to build the models and in Section 4 we present our results. -128-

2. The data 2.1. Finding suitable data In criminal proceedings, investigators interview numerous witnesses, who can produce true or false statements. In many cases the investigators do not know which statement is true or false, and in most cases the transcripts of these testimonies do not reproduce verbatim what the subjects said. Instead, they are simply a synthesis of the witnesses' declarations, carried out by the police officer who produces the transcript. Such reports are not a faithful mirror of the linguistic behavior of the subjects; therefore they are not useful from the purposes of the present paper. However in Italy there is a specific case of testimony that is reported verbatim: hearings that take place during a debate in front of the judge. Focusing on this aspect of the criminal procedure is therefore the most promising way of studying deception production. Furthermore, to focus on the debate is a convenient choice from the point of view of the homogeneity of data. It is an event strongly ritualized, in which actors and acts recur in a standard way, and it guarantees a certain regularity of conditions in different hearings. In addition, there is a type of criminal proceedings in which the truthfulness or deceptiveness of testimonies is easily verifiable. This is the case of criminal proceedings concerning violations of articles 368 and 372 of the Italian Criminal Code1 that codify the crimes of ‘calumny’ and ‘false testimony’, respectively. They are typically proceedings that originate when statements, issued in hearings related to any crime, are found unreliable, and therefore the statements themselves become the object of a further criminal proceeding for ‘calumny’ or ‘false testimony’. Because these proceedings are related to the lies, necessarily they end with a judgment that points out in a certain, organic and exhaustive way, the lies told by the defendant. 2.2. Data collection Our first step was to contact the Courts in several Italian towns, in order to receive authorization to examine their dossiers and extract information from them for research purposes. The three Presidents of Court to which the research project has been presented, allowed the collection of the data, with the restriction of publishing them in anonymous form, respecting the privacy of the subjects involved. This paper is based on the data collected in the Courts of Trento, Bolzano and Prato. Eighteen hearings with false testimonies were identified, issued by a total of seventeen subjects, one of whom was interrogated twice, who appeared in the hearings as defendant, witness or expert witness.

1

To be precise, art. 368 reads: Chiunque, con denunzia, querela, richiesta o istanza, anche se anonima o sotto falso nome, diretta all'Autorità giudiziaria o ad altra Autorità che a quella abbia obbligo di riferirne, incolpa di un reato taluno che egli sa innocente, ovvero simula a carico di lui le tracce di un reato, è punito con la reclusione da due a sei anni. In brief, it punishes whoever tries to charge the responsibility of some crime on someone who he knows is innocent. Art. 372 instead reads: Chiunque, deponendo come testimone innanzi all'Autorità giudiziaria, afferma il falso o nega il vero, ovvero tace, in tutto o in parte ciò che sa intorno ai fatti sui quali è interrogato, è punito con la reclusione da due a sei anni. This article punishes someone who, in front of the Judicial Authority, says a falsity or denies the truth, or does not reveal what he knows about the investigated facts.

-129-

2.3. Pre-processing Each transcript was converted to XML format according to a coding scheme where each intervention of the heard subject, in between the interventions of some other individual, is classified as a turn. Each turn can be constituted by one or more utterances—delimited by terminal punctuation marks—that are the main units of analysis for this paper. This resulted in 1437 utterances available for analysis from the complete corpus of 18 transcripts. Each utterance of a witness was assigned a label that specifies the truthfulness or truthlessness of the utterance itself. This annotation was carried out by hand, on the basis of information found in the Court's judgment relative to the testimony. Between the white of the truth and the black of the falsity, however, there are wide gradations of gray, and the judgment that describes the facts and points out the lies told, cannot specify the truth of each statement issued in the courtroom. To label the utterances is therefore a complex task, as discussed in the following annotation scheme: ‘False’: the utterance is clearly pointed out in the judgment as false, or the falsity is a logic consequence of some ascertained lie. ‘True’: the utterances that are coherent with the reconstruction of the facts contained in the judgment are considered true. Also the utterances that explain something not considered in the judgment because they not influential with respect to the investigated facts, are generally considered true. ‘Not reliable’: an utterance is considered not reliable if it is related to the investigated facts, but the judgment does not prove its deceptiveness. ‘True or not reliable’: like the ‘not reliable’ utterances, the ‘true or not reliable’ ones are related to the topic of investigation, and the judgment demonstrates nothing about them. Nevertheless, according to the event and to other statements certainly true or false, and/or on the basis of a weak connection with the interests that the subject tries to defend, it is logical to suppose that they are probably true. In brief, according to common sense, those utterances should be true, but the fact is not demonstrated, and ultimately questionable. ‘False or not reliable’: this is the specular situation with respect to the previous point. According to the interests of the subjects, and to the economy of the event and of the testimony, it is reasonable, but not demonstrated, that these utterances are false. In these cases, the final evaluation is not certain, and a note is made about the ‘hue’ of the statement. ‘Undecidable’: the utterances that, from a logical point of view, cannot be either true or false, are considered undecidable. This is the case for many questions (like ‘Excuse me, can you repeat?’), but also for several utterances stopped in mid-sentence, that do not have a complete sense. This is also the case for utterances that have a meta-communicative function, and regulate the relations between actors, like ‘Now I'll explain.’ or ‘If you think so...’ and so on. The corpus was tokenized and anonymized in accordance with the agreements with the Courts. Stop words were not removed: on the contrary, in stylometric analysis function words are considered crucial. Blocks of punctuation marks were considered as one token. For example, a single comma was considered a token, and three suspension points were also considered a single token. Finally, the corpus was lemmatized and POS-tagged using a

-130-

version of TreeTagger2 (Schmid, 1994) trained for Italian. While the utterances of other participants in the courtroom are not considered, the 1437 utterances of the heard subjects have been labeled according to this coding scheme. Corpus statistics are provided in Table 1. Table 1: Corpus statistics

Table1: Corpus statistics Utterances Tokens with without punct. punct. False 333 5778 4802 True 537 7908 6628 Not reliable 225 3351 2746 True or not reliable 83 1758 1452 False or not rel.reliable 78 1648 1360 Undecidable 181 1146 886 Total 1437 21589 17874 Label

The utterances labeled as ‘True or not reliable’, ‘False or not reliable’, ‘Not reliable’ and ‘Undecidable’ have been discarded, and the analyses concern only the part of the corpus constituted by ‘True’ and ‘False’ utterances: a total of 870. 3. Methods 3.1. Features Each utterance was described as a feature vector. The features come from a training set made by ten of our eighteen hearings. This subset of hearings provides 623 utterances labeled as ‘true’ or ‘false’ (about 72% of the utterances in our corpus). These features were selected by looking at the most distinctive features of true and false utterances, derived from the following approach. Frequency lists of all lemmas arising from both true and false utterances were created separately. The 200 most frequent lemmas for each type of utterance were selected and afterwards merged into a single list containing the most frequent lemmas of both classes of utterances. Theoretically, this list could have had a minimum of 200 items, in case of completely identifying the two previous lists, and a maximum of 400 items, in the case of no overlap. The same procedure was applied to collect the following features, independently for each class, as shown in the Table 2.

2

http://www.ims.uni-stuttgart.de/projekte/corplex/TreeTagger/DecisionTreeTagger.html

-131-

Table 2: Selected features

Table 2: Selected features Features Selected first 200 Lemmas first 200 Bigrams of lemmas Trigrams of lemmas first 200 first 25 POS Bigrams of POS first 25 first 25 Trigrams of POS Total 675

The last features in the vector were the length of each utterance, with and without punctuation. Therefore, the theoretical minimum length of the final vector was 677 features (the lengths of the utterances, plus the 675 features stated above) and the theoretical maximum 1352. In the end, the feature vector had 1021 features. This suggests that the features of true and false utterances are quite different, and this is promising for the following analyses. 3.2. Baseliners Before evaluating the results of the analyses, it was necessary to compute a baseline with which to refer. This was achieved through a simulation using a Monte Carlo technique. First, four hearings were used as a test set, for a total of 148 utterances, about 17% of the total amount of utterances in our corpus. This test set had 81 utterances labeled as ‘true’ and 67 as ‘false’, about 54.73% and 45.27% of ‘true’ and ‘false’ utterances, respectively. Then, 10,000 simulations were carried out, in which a classifier tried to guess the class of each entity of the test set, simply on the basis of the fact that 54.73% of the entities belong to the class ‘true’, and 45.27% belong to the class ‘false’. The result was that more than 99% of simulations did not exceed 60% of correct answers. Therefore 60% of correct classifications was assumed as the threshold for our test set. 3.3. Training Using the training set mentioned above, models were built using the Naïve Bayes and SVM classifiers in the Weka package3. In order to evaluate the models' effectiveness in classification task, the said test set was employed. 4. Results 4.1. Modal performances The results of the classifiers on the supplied test set are shown in Tables 3 and 4. While SVM4 performs clearly better than the random classifier, with a remarkable precision detecting deception, Naïve Bayes barely exceeds the baseline of the 60%. 3 4

http://www.cs.waikato.ac.nz/ml/weka/ The algorithm for training Support Vector Machines was Sequential Minimal Optimization—SMO.

-132-

Table 3: Naïve Bayes performance—supplied test set

False utterances True utterances Total Total %

Table 3: Naive Bayes performance - supplied test set Correctly Incorrectly classified entities classified entities Precision Recall 18 49 0.75 0.269 75 6 0.605 0.926 93 55 62.84% 37.16%

F-Measure 0.396 0.732

Table 4: SVM performance—supplied test set

False utterances True utterances Total Total %

Table 4: SVM performance - supplied test set Correctly Incorrectly classified entities classified entities Precision Recall 33 34 0.917 0.493 78 3 0.696 0.963 111 37 75.00% 25.00%

F-Measure 0.641 0.808

4.2. Deceptive language Overall, the performances of the SVM models are well above the chance level. Concerning false utterances, which are the target we have to detect, the precision is more than 90%. Instead, the recall needs to be improved, being slightly lower than 50%. For the next stage of the analysis, it was necessary to determine which kinds of utterances were easier or, conversely, more difficult to classify. To answer this question, the test set was examined. In general, the statements are very brief. From a total of 148 utterances: 95 contain 5 tokens or less (considered without punctuation): that is 64.19% of the test set; 27 have from 6 to 10 tokens; 26 are longer than 10 tokens. Figure 1 represents the distribution of the length of the utterances, showing separately the true and the false ones.

-133-

Figure 1: Distribution of length of utterances

Given this distribution, the accuracy of our models was examined on the basis of the length of the utterances. The results show that there is an improvement in performance with shorter statements compared to longer statements, as shown in Table 5. Table 5: Accuracy of SVM models according to utterance lengths

Length Accuracy False utterances True utterances

Table 5: Accuracy of SVM models according to utterance lengths 1-5 tokens 6-10 tokens 11 tokens 83.20% 59.20% 61.50% Precision Recall Precision Recall Precision Recall 1

0.61

1

0.267

0.571

0.364

0.771

1

0.522

1

0.632

0.8

In particular, maybe contrary to what could be thought, the utterances equal to, or shorter than 5 tokens, are classified with accuracy higher than 80%, while the accuracy for the longer utterances corresponds more or less to the chance level. Short statements are typically conventional, that is made by stereotyped linguistic formulas, which could be relevant in order to classify statements as true or false. To explore that idea, correspondence analysis has been carried out on the entire corpus (Baayen, 2008). The results are shown in Figure 2.

-134-

Figure 2: The correspondence analysis

The most ‘extreme’ features, useful to classify the utterances, are just brief and highly conventional expressions including, for example: ‘(Do not) know’ - ‘(Non) ricordare’; ‘(Do not) remember’ - ‘(Non) sapere’; ‘Yes’ - ‘Sì’; ‘Not’ - ‘No’; ‘Sure’ - ‘Certo’, and so on... Therefore, it is possible to suppose that, when the language is more conventional, is easier to be recognized as true or false. 5. Conclusions Our data show that it is possible to train models to classify statements as true or false, with performances clearly above the chance level. However long utterances are more difficult to classify, probably because their complexity represent noise for the models. This suggests that a future research direction may be to employ vectors containing less features, but constituted by longer n-grams, to detect expressions longer than three lemmas. Deception is generally accepted to create an increase in cognitive load (Vrij, A., Fisher, R., Mann, S., and Leal, S., 2006). This increased cognitive load required to produce a deceptive statement, in culmination with the stress related to being involved in the hearings themselves, could account for the shorter, more conventional utterances identified in Section 4.2 which require less cognitive load to produce. In a not yet published paper by Tomblin et -135-

al. (in preparation) ‘Formulaic Language’ is used as a marker of deception and they arrive at analogous conclusions. According to our data, in any case, it seems that commonly used expressions can be useful in identifying not only deceptive, but also sincere statements. In fact, if on one hand short negative answers, as well as brief denials of knowing or remembering, are typical of deceptive language, on the other hand short affirmative statements are generally truthful. It depends, at least in part, on the dynamic of the event of the hearing itself: prosecutors pose questions about facts which are the object of investigation, with the aim to verify the information collected during the inquiry. Therefore it is possible in the hearings to find several questions to which the interrogated subjects have to answer confirming or denying facts ascertained during the investigation activities. It is obvious that the subjects tend to be sincere when they confirm what the prosecutor already knows, and conversely they often lie through denials of responsibilities which are explicitly charged on them. Therefore, to have better insights about deceptive language, and to be more precise at recognizing it, it is necessary to carry out more refined analyses. Another possible way to improve analyses could be to employ linguistic tools for the lexical evaluation of the texts. As underlined above, a well known resource of this kind, already employed to detect deception in texts (Newman et al., 2003), is the Linguistic Inquiry and Word Count (LIWC). In fact, exploratory analyses seem to be promising, regarding the possibility to improve the model performances, employing a combination of surface and lexical features. On the other hand, short utterances are already well classified. Furthermore, the false utterances are recognized with a high degree of precision. It means that, at least for certain kinds of statements, deceptive language is clearly different to truthful language and it can be recognized. In a real life scenario such as the context described in this paper, the ability to confidently detect deception is an important contribution. Acknowledgements The data collection is a complex task that could not have been carried out without the help of a lot of people. Many thanks to Dr. Heinrich Zanon, President of the Court of Bolzano; to Dr. Sabino Giarrusso, President of the Court of Trento; and to Dr. Francesco Antonio Genovese, President of the Court of Prato. Many thanks also to Dr. Piero Tony, Chief Prosecutor of the Public Prosecutor's Office of Prato, to Dr. Sandro Pettinato of the Court of Trento, to Dr. Biagio Mazzeo, Prosecutor in the Public Prosecutor's Office of Genova, to Dr. Michela Guidi, Prosecutor in the Public Prosecutor's Office of Prato, and to Rita Fava of the Public Prosecutor's Office of Prato. Many thanks also to Dr. Marco Baroni of the University of Trento and to Dr. Carlo Strapparava of FBK - Fondazione Bruno Kessler. References Adams, S. H. (1996) Statement analysis: what do suspects’ words really reveal? The FBI Law Enforcement Bulletin, 65(10): 12—20. Anolli, L., Balconi, M., and Ciceri, R. (1999) Ulisse o Richelieu? Stili verbali della comunicazione menzognera. Lingua e stile, 34(3): 379—402. Baayen, R. (2008) Analyzing Linguistic Data: A practical introduction to statistics using R. Cambridge University Press.

-136-

Bachenko, J., Fitzpatrick, E., and Schonwetter, M. (2008) Verification and implementation of language-based deception indicators in civil and criminal narratives. In Proceedings of the 22nd International Conference on Computational Linguistics - Volume 1, COLING ’08, pages 41—48, Stroudsburg, PA, USA. Association for Computational Linguistics. Bond, C. F. and De Paulo, B. M. (2006) Accuracy of deception judgments. Personality and Social Psychology Review, 10(3): 214—234. Coulthard, M. (2004) Author identification, idiolect, and linguistic uniqueness. Applied Linguistics, 25(4): 4310—447. De Paulo, B. M., Lindsay, J. J., Malone, B. E., Muhlenbruck, L., Charlton, K., and Cooper, H. (2003) Cues to deception. Psychological Bulletin, 129(1): 74—118. Fitzpatrick, E. and Bachenko, J. (2009) Building a forensic corpus to test language-based indicators of deception. Language and Computers, 71(1): 183—196. Frank, M. G. and Feeley, T. H. (2003) To catch a liar: challenges for research in lie detection training. Journal of Applied Communication Research, 31(1): 58—75. Hancock, J. T., Curry, L. E., Goorha, S., and Woodworth, M. (2008) On Lying and being lied to: a linguistic analysis of deception in computer-mediated communication. Discourse Processes, 45(1): 1—23. Jensen, M. L., Meservy, T. O., Burgoon, J. K., and Nunamaker, J. F. (2010) Automatic, multimodal evaluation of human interaction. Group Decision and Negotiation, 19(4): 367—389. Koppel, M., Schler, J., Argamon, S., and Pennebaker, J. (2006) Effects of age and gender on blogging. In AAAI 2006 Spring Symposium on Computational Approaches to Analysing Weblogs. Levine, T. R., Feeley, T. H., McCornack, S. A., Hughes, M., and Harms, C. M. (2005) Testing the effects of nonverbal behavior training on accuracy in deception detection with the inclusion of a bogus training control group. Western Journal of Communication, 69(3): 203—217. Luyckx, K. and Daelemans, W. (2008) Authorship attribution and verification with many authors and limited data. In Proceedings of the 22nd International Conference on Computational Linguistics - Volume 1, COLING ’08, 513—520, Stroudsburg, PA, USA. Association for Computational Linguistics. Mosteller, F. and Wallace, D. (1964) Inference and disputed authorship: The Federalist. Addison-Wesley. Newman, M. L., Pennebaker, J. W., Berry, D. S., and Richards, J. M. (2003) Lying words: predicting deception from linguistic styles. Personality and Social Psychology Bulletin, 29(5): 665—675. -137-

Pennebaker, J. W., Francis, M. E., and Booth, R. J. (2001). Linguistic Inquiry and Word Count (LIWC): LIWC2001. Lawrence Erlbaum Associates, Mahwah. Porter, S., Woodworth, M., and Birt, A. R. (2000) Truth, lies, and videotape: an investigation of the ability of federal parole officers to detect deception. Law and Human Behavior, 24(6): 643—658. Porter, S. and Yuille, J. C. (1996). The language of deceit: an investigation of the verbal clues to deception in the interrogation context. Law And Human Behavior, 20(4): 443—458. Schmid, H. (1994) Probabilistic part-of-speech tagging using decision trees. In Proceedings of International Conference on New Methods in Language Processing. Solan, L. M. and Tiersma, P. M. (2004) Author identification in American courts. Applied Linguistics, 25(4): 448—465. Stein, B., Koppel, M., and Stamatatos, E. (2007) Plagiarism analysis, authorship identification, and near-duplicate detection pan’07. SIGIR Forum, 41: 68—71. Strapparava, C. and Mihalcea, R. (2009) The lie detector: explorations in the automatic recognition of deceptive language. In Proceeding ACLShort ’09 - Proceedings of the ACL-IJCNLP 2009 Conference Short Papers. Tomblin, S., Taylor, P., Vrij, A., Leal,. S., Mann, S., Nash, R. & Menacere, T. (in preparation) Formulaic language occurs more often in deceptive statements. Undeutsch, U. (1967) Beurteilung der Glaubhaftigkeit von Aussagen [Veracity assessment of statements]. In Undeutsch, U., editor, Handbuch der Psychologie: Vol. 11. Forensische Psychologie, 26—181. Hogrefe, Gottingen, Germany. Vrij, A. (2005) Criteria-based content analysis—A qualitative review of the first 37 Studies. Psychology, Public Policy, and Law, 11(1): 3—41. Vrij, A., Fisher, R., Mann, S., and Leal, S. (2006) Detecting deception by manipulating cognitive load. Trends in Cognitive Sciences, 10(4): 141—142. Zhou, L. (2005) An empirical investigation of deception behavior in instant messaging. IEEE Transactions on Professional Communication, 48(2): 147—160. Zhou, L., Burgoon, J. K., Nunamaker, J. F., and Twitchell, D. (2004) Automating linguisticsbased cues for detecting deception in text-based asynchronous computer-mediated communication. Group Decision and Negotiation, 13(1): 81—106. Zhou, L., Shi, Y., and Zhang, D. (2008) A statistical language modeling approach to online deception detection. IEEE Transactions on Knowledge and Data Engineering, 20(8): 1077—1081.

-138-

The false report during an emergency call: using discourse analysis to detect deceit Marty Laforest Université du Québec à Trois-Rivières, Canada [email protected] Abstract Presumably, the vast majority of individuals who dial an emergency phone number to report a robbery or an assault are sincere. However, police investigations have shown that a small number of callers disguise the facts they report, thereby concealing their involvement in the events concerned. Although the deception is usually discovered, such individuals make investigators lose precious time. The objectives of the study are 1) to determine whether emergency calls contain identifiable interactional/pragmatic indicators of deception by callers; and 2) if they do, to construct a model of analysis that investigators can use to assess callers’ sincerity. This study uses an interactionist, comparative perspective and is based on a corpus of 40 authentic emergency calls: 20 from deceitful callers and 20 from sincere callers. Weakly suitable answers to the call-taker’s questions and the absence of additional information that would help the call-taker better understand the caller's answers to her/his questions proved to be the best predictors of deception. A tree-based approach was used to generate a decision-tree that constitutes a promising model for determining whether or not a caller is deceitful. Keywords:

EMERGENCY CALLS; DECEIT; DISCOURSE ANALYSIS; TALK-IN-INTERACTION

Rationale1 Many police investigations begin with a phone call to an emergency service (number 911 in North America). In the vast majority of cases, the caller is completely sincere. But some investigations show that a small number of individuals, in one way or another, disguise the facts they report, thereby concealing or minimizing their involvement in the events concerned. One caller will say his wife has disappeared, when he has actually killed her; another will say he is the victim of a holdup, which he has, in fact, organized himself. Although the deception is usually discovered, such individuals, whom I characterize as deceitful, given that they present information so as to mislead the call-taker, complicate the task for investigators and make them lose precious time. Does a deceitful caller, as opposed to a sincere one, behave differently during the short amount of time in which s/he interacts with the call-taker? The aim of this paper is to establish whether this is the case. The question is of no importance to the call-taker, whose task is to quickly pick out the information that can be used to intervene as effectively as possible, not to determine whether the caller is telling the truth. But it does matter to the 1

This research project on emergency calls would not have seen the light of day without the support of the Québec provincial police (Sûreté du Québec [SQ]). I would especially like to thank Michel St-Yves and Brenda Mimeault, of the SQ behavioral analysis service (Division de l’analyse du comportement), for their help, as well as SQ telecommunications clerk Jenny Tanguay. Thanks go as well to Gaétan Daigle, of Université Laval, for his support with the statistical analysis, to Mark Dobbie, who translated the original French manuscript into English, and to the anonymous referees for their comments.

-139-

investigator in charge of clearing up the circumstances of a given event. If deceitful callers do behave differently, there is hope that analyzing emergency calls will help to identify them. Accordingly, the objectives are: 1) to determine whether 911 calls contain identifiable indicators of deception by callers; and 2) if they do, to construct a model of analysis that investigators can use to assess callers’ sincerity. State of research The detection of lies on the basis of verbal or vocal indicators has been dealt with in a great many studies (for a synthesis, see Vrij, 2008). According to Vrij, a number of these studies lack scientific rigour. Much of the academic work on the question has been conducted by psychologists, and the main weaknesses in many studies lie, in my view, in the methodology adopted, which is often based on the creation upon request of untruthful discourse spoken by subjects recorded in a laboratory (Burgoon, Blair, Qin and Nunamaker, 2003; Newman, Pennebaker, Berry and Richards, 2003; Sporer, 1997; Vrij, Edward, Roberts and Bull, 2000 among others)2. In such a context, the lack of real stakes for the liar and the fact that the discourse produced is (in many cases) a monologue, stand in contrast to real situations, in which deception occurs only in interactions, and therefore may be sources of bias. Such methodology also presupposes a narrow conception of lying that is evident in the instructions provided to subjects. Naturally, any assertion that wilfully contradicts the facts is untruthful, but the deception discussed in this paper includes any discourse aimed at misleading the interlocutor, which includes lying, but also all cases where a speaker simply leaves out information s/he could or should have given, or gives information that cannot be used to get the most accurate idea of the facts. Finally, the verbal indicators identified—for example, the weak lexical diversity mentioned by Colwell et al. 2002, among others—lack complexity in certain respects, because the analysis does not take into account the pragmatic and sociolinguistic dimensions of discourse. In this regard, the study by Shuy (1998) (not spoken of by Vrij) stands out for its wealth of content and sophisticated analysis, because it not only deals with discourse produced in real communication situations, but also rigorously takes into account all dimensions of discourse. Important studies have been conducted on emergency calls in general, notably by Zimmerman (1984, 1992a, 1992b), Whalen, Whalen and Zimmerman (1990), Tracy (1997), Tracy and Anderson (1999), and Drew and Walker (2010), but the studies on detecting lying in emergency calls are very few (Harpster, Adams and Jarvis, 2009; Laforest, Blais and Saint-Yves, [2007] 2009). Yet, as Harpster, Adams and Jarvis, following Shuy (1998), pointed out, a caller’s discourse produced in this kind of situation can be particularly revealing, insofar as it is less affected by the interviewer’s behaviour than it would be during an interrogation. This discourse contains the very first version of the facts, or, at least, the first mention of them by the caller. This first version is a response to a very open-ended question from the call-taker (‘What is your emergency?’). It provides access to a formulation that corresponds to the caller’s idea of an emergency call. 2

Bond and Lee (2005) and Mann, Vrij and Bull (2002) are notable exceptions, but the latter is a study of nonverbal liar’s behavior only.

-140-

Data Deceitful discourse is not easy to collect. However, it was possible to constitute a sample of 40 emergency calls using recordings provided by the Québec provincial police (Sûreté du Québec): 20 from deceitful callers and 20 from sincere callers. These characterizations of the callers were known to be correct as a result of now-completed police investigations. The calls were all made to the 911 emergency service. The types of cases are varied, but events (crimes or, at least, suspected crimes) involving persons (suspicious deaths, threats, various assaults) are much more numerous than events concerning property (thefts or fires). Initially, I thought the type of case would play a role in the caller’s behavior, but that turned out not to be the case. Table 1 (upper part) shows the distribution of calls in the corpus according to these two categories. To the extent that a determination was possible, only adult callers were included. Information concerning the caller’s age was generally not available, so the caller’s voice and the content of the call were the only clues in this regard. One call placed by an 11-year old who mentioned his age to the call-taker was excluded. With respect to how close they presented themselves as being to the event, callers fell into one of the following three categories: 1) victim of the event; 2) eye witness to the event; 3) the person without whom the police would not have learned of the event, because the victim was unable to lodge a complaint her/himself and no eye witnesses reported the event (see Table 1, lower part). Table 1: The calls analyzed Type of event Event involving a person Event concerning property Total Caller’s relationship to event Victim Eye witness Other Total

Sincere calls 18 2 20

Deceitful calls 15 5 20

8 4 8 20

13 1 6 20

Methodology The data were analyzed from an interactionist and sociopragmatic perspective. Accordingly, discourse is considered here from a perspective that emphasizes its actional dimension (discourse is ‘doing’ with language), and as a place of ongoing construction and negotiation of sense—sense not being a pre-interaction ‘given’ that speech merely transmits—and social roles. More specifically, the analysis draws on Grice’s cooperation principle and conversational maxims (Grice, 1975), and on the Goffmanian concept of frame (Goffman, 1974; Tracy, 1997). Framing corresponds to how the interlocutors conceive of the interaction and the expectations they have as a result. First, the data were transcribed. Then, the calls in each group (from sincere and nonsincere. i.e. deceitful callers) were compared to identify any elements of differentiation, that is, any elements likely to be present much more often in one group than in the other. No account was taken of the suprasegmental level, which often requires a comparison of the same individual in different speech situations. Vocal, but nonverbal, indicators (sobbing, screaming, radical change in speech rate and voice quality, hesitating) were left aside for two reasons. The first is that it is difficult to categorize their intensity. The second (and more -141-

important) is that lay persons and researchers have shown themselves to be hypersensitive to such indicators; according to Vrij (2008: chapter 6), police manuals and the general public alike tend to overvalue nonverbal indicators because it is taken for granted (incorrectly, as all sociolinguists know) that a good liar has a much better ability to control her/his discourse than her/his nonverbal behaviour. Olsson (2004) and Harpster, Adams and Jarvis (2009) regard speaking in a monotone—with an emotionless voice—as an indicator of lying, based on the observation that most callers to emergency services (among whom sincere callers dominate) are stressed and emotionally upset, and that it shows in their voice. But this is so close to what would probably cross the mind of anyone asked to imagine an emergency call that we can reasonably assume a deceitful caller will have an immediate tendency to feign the behavior concerned. For want of skill in detecting feigned emotion in people’s voices, such indicators were left aside and the analysis focussed instead on elements of which speakers are generally much less aware. Results The comparison between the two groups of calls initially revealed 11 differentiating elements. Because emergency calls are unique events that involve different speakers every time and because they unfold, in part, in unpredictable ways, each element is observable in the discourse of both sincere callers and deceitful callers, and, consequently, no element alone can be a determinative indicator of deception. If a correlation is to be established between a caller’s discourse and deception, it can be done only by taking into account combinations of elements3 (Bond and Lee, 2005; Harpster, Adams and Jarvis, 2009; Zhou, Burgoon, Zhang and Nunamaker, 2004). Most of the elements identified are related to what can be thought of as the way in which the caller fulfils her/his role. A speaker can properly fulfil an interactional role only if s/he shares the same frame as her/his interlocutor. An emergency call, like any service encounter, is an asymmetrical interaction, in which the representative of the emergency service is in a position of power with respect to the caller. The latter generally has little or no experience with the interactional genre of the ‘emergency call’ and does not know the organizational frame in which the call-taker operates; consequently, the information s/he spontaneously provides is not necessarily the information the call-taker needs to do her/his job. Ten of the 11 elements of differentiation found in the corpus concern the way in which the caller cooperates (in the sense used by Grice, 1975) and agrees to share the frame defined by the call-taker (the remaining element is the presence of contradictions, much less frequent with sincere callers than with deceitful ones). As a whole, these elements reveal that deceitful callers are less ‘appropriate’ in their behaviour, less cooperative, and less willing than sincere callers to follow the call-taker’s lead (Laforest, Blais and St-Yves, [2007] 2009). From a similar perspective, Olsson (2004) and Harpster, Adams and Jarvis (2009: 72) have spoken of indicators of lying related to attitude, which is defined as ‘the commitment and cooperation displayed by the caller to the emergency operator’. To illustrate, deceitful callers seem more easily thrown off by questions they are not expecting, probably because they are more anxious than sincere callers to stick to the scenario and caller behaviour they have imagined in advance. From a very different perspective, but with similar results, Vrij, Leal, Anders Granhag, Mann, Fisher, Hillman and 3

As Vrij (2008: 108) jokingly affirmed: ‘Looking at individual cues is searching for the verbal equivalence of Pinocchio’s growing nose, which does not exist’.

-142-

Sperry (2009) have demonstrated that asking unanticipated questions is a good strategy for outsmarting liars. Seven of the 11 variables identified proved to be statistically significant when their possible interaction (multivariate statistics) was taken into account. These elements are: 1. Delay in the core of the call’s occurrence I use the term ‘core’ (of the call) to describe the first caller’s turn that states, however minimally, the reason for the call to 911. Generally, the second turn of the call (which is the caller’s first turn, after the opening turn of the call-taker) constitutes the core. But in deceitful calls, the opening sequence is often prolonged (as in example 1, where the core begins at turn 8, line 10) and therefore the core’s position is more variable. (1)4 1 2 3 4 5 6 7 8 9 10 11 12

CT Sûreté du Québec C Oui, bonjour heu: pour la région de St-Oscar s’il-vous-plaît CT Pardon? C Pour la région de St-Oscar CT Oui? C Je suis bien là, oui? CT Heu: on répond pour eux autres, oui C Ok: c’est parce que, moi, mon épouse est partie a travail ce matin mais [...]

1 2 3 4 5 6 7 8 9 10 11 12

CT Sûreté du Québec C Yes, hello hmm: Region of St-Oscar please CT Sorry? C Region of St-Oscar CT Yes? C This is the right number? CT Uh: we take calls for them, yes C Okay: What it’s about, me, my wife went to work this morning but [...]

4. Core elements given out of order The core of the call can vary in length and include the following elements in this order (with the exception of the last one, which can appear anywhere): Caller identification, an explicit request that can be paraphrased ‘I want/Could you’, the reason for the call, some narrative elements, and the mention of the caller’s location. None of the elements is absolutely necessary, but the ‘standard’ order in giving them is less often adhered to by deceitful callers. 4

C = Caller; CT = Call-Taker. The key to transcription symbols is provided in the appendix. In all examples, all names, streets, and telephone numbers have been changed in order to preserve the anonymity of the people involved.

-143-

In example 2 (below), for instance, the caller gives a narrative element (‘there were two of them’, line 3) before giving the reason for the call (‘to take everything’, line 5). 5. Unclear reason for the call Deceitful callers much more frequently (than sincere ones) experience difficulty in clearly expressing why they have called for help. The evaluation of clarity is based first and foremost on the call-taker’s reactions. When the reason for the call is vague or absent from the first main caller’s speaking turn, the call-taker’s questions clearly indicate his/her incomprehension and need for clarification (as in example 2, line 6). (2) 2 C 3 4 5 6 CT 7 C 2 C 3 4 5 6 CT 7 C (call 35)

heu bonsoir, ils étaient deux [sanglote et a du mal à parler] pis: partir avec tout’ Tout:::? L’argent, tout Uh, good evening, there were two of them [sobbing, difficulty speaking] then: to take everything Everything:::? The money, everything

7. Weakly suitable answer to a call-taker’s question (see below) 9. Presence of questions asked by the caller to the call-taker (example 3, line 122) (3) 120 121 122

CT C

120 CT 121 122 C (call 03)

Ok. Bon, écoutez heu: moi monsieur Rondeau, Qu’est-ce que vous me suggérez? Ok. So, listen uh: Mr. Rondeau, I What do you suggest?

10. Absence of further details about the information that answered the call-taker’s questions (see below) 11. Presence of pauses before the caller’s answers to the call-taker’s questions Among these, elements of differentiation Nos. 7 and 10, which specifically concern the way in which the caller answers the questions s/he is asked, constitute the best predictors of deception and I will present them in detail. -144-

There are five phases to an emergency call (Zimmerman, 1984). After a brief opening sequence (Phase 1), the caller more or less explicitly formulates a request for assistance (Phase 2). This request is followed by an interrogative series during which the call-taker seeks to obtain the information needed to organize the help requested (Phase 3), before announcing that the help is being sent, which constitutes her/his response to the caller’s request (Phase 4). A brief closing sequence marks the end of the call (Phase 5). The interrogative series, with questions from the call-taker, is the longest phase of the call. The number of requests for information varies, as does the form of the requests, with both open-ended and close-ended questions being asked. Because time is of the essence in handling emergency calls, the questions follow one another in rapid succession. Consequently, the answers are brief, but the callers nevertheless have the leeway to make answers of variable length, which illustrate their idea of what is expected of them. A comparison of the two series of calls revealed a difference in the degree of coherence of question-answer sequences (cf. Grice’s (1975) cooperation principle). This degree of coherence was evaluated using three (data-driven) categories of answers to questions: 1) Perfectly suitable answer to the question, when the caller provided exactly the information requested and no other (Example 4, lines 30 and 33); (4) 29 30 31 32 33

CT (Mais) il est tu en boisson lui? C Oui, CT Ah bon ok il est tu pas mal euh: en boisson? C Oui

29 CT (But) has he been drinking? 30 C Yes, 31 CT Ah right okay is he pretty uh: 32 far gone? 33 C Yes (Call 09a) 2) Weakly suitable answer, when it is hard to establish a relationship between the information provided and the question asked, which is generally reflected by the calltaker’s reformulation of the same question or the utterance of a new question on the same subject in the next speech turn. Truly incoherent discourse is very rare, and the corpus used in the study contains none. In nearly all cases, one can manage to understand in what way a given answer is relevant; the answers placed into this category are those whose relevance was not immediately apparent (Example 5, lines 38—45); (5) 35 C 36 37 CT 38 C 39

Ça fait un bout’ que j’étais knock-out. Vous dormez dans le fossé là? Ben je suis allé heu: je suis allé faire une job: une estimation pour -145-

40 une job électrique à: heu: 41 boulevard St-Louis pour un gars 42 pis: je suis parti de là il devait 43 être heu: 11 heures et demie minuit 44 pis là tu me dis qu’il est trois 45 heures fait que: 46 CT Ça veut dire que vous avez été 47 sans connaissance monsieur? 35 C 36 37 CT 38 C 39 40 41 42 43 44 45 46 CT 47 (Call 04)

It’s been a while that I was out cold. You’re sleeping in the ditch then? Well I went to do uh: I went to do a job: an estimate for an electrical job at: uh: St. Louis boulevard for a guy and: I left there it must have been uh: eleven thirty midnight and now you tell me it’s three o’clock so: You mean you were unconscious, sir?

3) Suitable and supplementing answer, when unrequested information is added to that requested (Example 6, lines 16—17). (6) 13 CT Le IGA [un supermarché] sur 14 la rue des Pins? 15 C Oui 16 il y a une pharmacie sur le 17 coin là 13 CT The IGA [a supermarket] on 14 Pine Street? 15 C Yes 16 there’s a pharmacy on the 17 corner there (Call 27a) Element of differentiation No. 7 is found in weakly suitable answers, which are much more numerous in calls from deceitful callers than in calls from sincere callers (p <.0032). The importance of this element has emerged in several studies (notably Olsson, 2004; Harpster, Adams and Jarvis, 2009), even though the corpus in each study was different, which lends greater weight to the element. The score for this variable was calculated in relation to the number of requests for information made by the call-taker in each call: it corresponds to the number of weakly suitable answers per 10 questions from the call-taker. -146-

Element No. 10 is related to observing the ‘supplementary information’ in the answers classified into the ‘suitable and supplementing’ category. The supplementary, unrequested information was also classified into three (data-driven) categories: 1) Further details about or clarification of the information that answered the question (as in Example 6, lines 16—17); 2) An element of an answer to a question anticipated by the caller, but not yet asked (i.e. information which is new, but which can be semantically/pragmatically associated with the information that answered the question, as in Example 7, lines 22—27). This kind of supplementary information has been discussed in Atkinson and Drew (1979); 3) Information completely unrelated to the information that answered the question (as in Example 8, lines 23—25). (7) 13 CT ‘ sontaient tu armés? [...] 20 CT Ils sontaient armés de quoi? 21 C Un un un: un vrai revolver, 22 et pis un masque en rubber là euh 23 comme de: de l’Halloween là? 24 CT ok 25 C pis euh: je: sais pas qu’est-ce 26 qu’il avait sur la tête, j’avais, 27 il a un jacket, rouge. 13 CT ‘ were they armed? [...] 20 CT How were they armed? 21 C A a a: a real revolver, 22 and then one of those rubber 23 masks uh like: a Halloween mask? 24 CT okay 25 C and uh: I: don't know what 26 did he have on his head, I had, 27 he’s got a jacket, a red one. (Call 19) (8) 21 CT À quel numéro on vous rejoint? 22 C Euh: 555-55-55, 23 là j’ose pus de: me déshabiller 24 pis lai: laisser traîner quoi que 25 ce soit là, mais: 21 CT At what number can we reach you? 22 C Uh: 555-55-55, 23 So I don’t dare: get undressed -147-

24 anymore and lea: leave anything 25 lying around, but: (Call 34a) Far fewer deceitful callers than sincere ones provide ‘further details about the information that answered the question’; in other words, they are much more reluctant to provide additional information that helps the call-taker better understand their answers to her/his questions (p <.0069) and, as in the case of weakly suitable answers, this can be seen as a manifestation of their less than cooperative behavior. This variable was binarized, the caller obtaining a score of 0 if the call contained at least one occurrence of ‘further details’ and 1 if it contained none. Table 2 shows the average scores of the two variables in each group of calls analyzed. The higher the average, the more the caller can be suspected of deceitful behavior. For ease of reference, the definitions of variables 7 and 10 are restated here: Variable 7: number of weakly suitable answers by caller per 10 questions from call-taker. Variable 10: presence (= 0) or absence (=1) of details about suitable information given in answer to a question. Table 2: Average score for variables 7 and 10 Status of caller sincere deceitful

Variable 7 0.265 1.35

Variable 10 0.05 0.4

Toward the construction of a model of analysis In this study, the dependent variable is the caller’s sincerity. The independent variables are the elements of differentiation identified. The degree of association between the dependent and independent variables was evaluated using a tree-based approach (Breiman, Friedman, Olshen and Stone, 1984). The tree is built by first finding the variable which best splits the data into two groups. This process is applied separately to each sub-group and is repeated until no improvement can be made. The stopping rule is based on a cross-validation technique; a cross-validated estimate of risk is computed for a nested set of subtrees and the subtree with the lowest estimate of risk is selected. The decision tree generated using this approach (see Figure 1) constitutes a rather promising model for determining whether or not a caller is deceitful.

-148-

Light zones = deceitful caller; Dark zones = sincere caller Sensitivity: Prob(predicts lying | really lying) = 17/20 = 85% Specificity: Prob(predicts sincere | really sincere) = 19/20 = 95% Figure 1: The model of analysis of caller behavior

The model predicts that: 1) Regardless of her/his score for Variable 10, a caller with a score greater than 1.05 for Variable 7 is deceitful; 2) A caller who obtains a score < 1.05 for Variable 7 and of 1 for Variable 10 has a strong chance of being deceitful (probability of 83%); 3) All other callers have a strong chance of being sincere (probability of 91%). 4) The probability that a deceitful caller will be identified as such (sensitivity of the model) is 85% and the probability that a sincere caller will be identified as such (specificity of the model) is 95%. Discussion and conclusion This model is no doubt too good to be true, and its limits must be kept in mind. The sample tested is much smaller and more heterogeneous, in terms of the type of crime reported, than that of Harpster, Adams and Jarvis (2009), which consisted of 100 calls for homicide. The calls analyzed are from different regions: procedures for call transfer (to the police or ambulance service) are not the same everywhere, which constitutes another factor of heterogeneity, affecting the beginning of the interaction. A larger sample might reveal that other variables among the 11 variables already identified are significant; it would also provide an opportunity to assess the possible influence of variable interactional styles across individuals, the caller’s gender or origin (which has not yet been addressed in this study), the caller’s relationship to the event (victim or witness, close to or not close to the victim), and -149-

the type of event reported. The latter two variables were looked at closely and were not significant, but this would perhaps not be the case with a broader sample. Finally, and above all, it is only by testing the model with a new sample, independent of the data that served to generate it, that its validity can truly be assessed. Such a sample is currently being established. But already, the results of this research seem to confirm that sincere and deceitful callers behave differently from one another and that much can be learned from a pragmatic/interactional analysis of emergency calls based on the caller/call-taker interaction, all the more given that the two variables on which the model is based have emerged, in a slightly different form, in other studies. Apart from that, the results of the analysis confirm that collaboration between linguists and police services is worthwhile—collaboration that has only just begun in Québec, but promises to be fruitful. Appendix: Transcription symbols Punctuation is used to give an idea of intonation. In the absence of punctuation or another indication, the flow of speech is continuous. Symbol : . , ? ! (xx) (street) [...] [word] C: bla [bla CT: [bla

Interpretation Sound stretch Ending, low falling, intonation Short pause Interrogative intonation Exclamative intonation (‘lively’, animated speech) Inaudible words Transcriber doubt about what those words are Talk omitted from the data segment Transcriber or analyst comment Overlapping passages

References Atkinson, J. M. and Drew, P. (1979) Order in Court: The Organization of Verbal Expression in Judicial Settings. London: MacMillan. Bond, G. D. and Lee, A. Y. (2005) Language on lies in prison: linguistic classification of truthful and deceptive natural language. Applied Cognitive Psychology 19: 313—329. Breiman, L., Friedman, J. H., Olshen, R. A. and Stone, C. J. (1984) Classification and Regression Trees. Belmont, California: Wadsworth International Group. Burgoon, J. K., Blair, J. P., Qin, T. and Nunamaker, J. P. (2003) Detecting deception through linguistic analysis. Lecture Notes in Computer Science 2665/2003: 958. Colwell, K., Hiscock, C. K. and Memon, A. (2002) Interview techniques and the assessment of statement credibility. Applied Cognitive Psychology 16: 287—300. -150-

Drew, P. and Walker, T. (2010) Citizen’s emergency calls. Requesting assistance in calls to the police. In M. Coulthard and A. Johnson (eds) The Routledge Handbook of Forensic Linguistics. London: Routledge, 95—110. Goffman. E. (1974) Frame Analysis. An Essay on the Organization of Experience. New York: Harper Colophon Books. Grice, P. (1975) Logic and communication. In P. Cole and J. L. Morgan (eds) Syntax and Semantics III: Speech Acts. New York: Academic Press, 41—58. Harpster, T., Adams, S. H. and Jarvis, J. P. (2009) Analyzing 911 homicide calls for indicators of guilt or innocence: An exploratory Analysis. Homicide Studies 13(1): 69—93. Laforest, M., Blais, D. and St-Yves, M. ([2007] 2009) 911 Emergency calls: towards a characterization of the spurious caller. In M. St-Yves and M. Tanguay (eds) The Psychology of Criminal Investigations. The Search for the Truth. Toronto: Carswell. 229—244. Mann, S., Vrij, A. and Bull, R. (2002) Suspects, lies, and videotape: an analysis of authentic high-stake liars. Law and Human Behavior 26(3): 365—376. Newman, M. L., Pennebaker, J.W., Berry, D.S. and Richards, J.M. (2003) Lying words: predicting deception from linguistic styles. Personality and Social Psychology Bulletin 29: 665—675. Olsson, J. (2004) Forensic Linguistics. An Introduction to Language, Crime and Law. London: Continuum. Shuy, R. W. (1998) The Language of Confession, Interrogation, and Deception. Thousand Oaks: Sage. Sporer, S. L. (1997) The less traveled road to truth: verbal cues in deception detection in accounts of fabricated and self-experienced events. Applied Cognitive Psychology 11: 373—397. Tracy, K. (1997) Interactional trouble in emergency service requests: a problem of frame. Research on Language and Social Interaction 30: 315—343. Tracy, K. and Anderson, D. L. (1999) Relational positioning strategies in police calls: a dilemma. Discourse Studies 1(2): 201—225. Vrij, A. (2008) Detecting Lies and Deceit. Pitfalls and Opportunities. Chichester: Wiley. Vrij, A., Edward, K., Roberts, K. P. and Bull, R. (2000) Detecting deceit via analysis of verbal and nonverbal behavior. Journal of Nonverbal Behavior 24(4): 239—263.

-151-

Vrij, A., Leal, S., Anders Granhag, P., Mann, S., Fisher, R. P., Hillman, J. and Sperry, K. (2009) Outsmarting the liars: the benefit of asking unanticipated questions. Law and Human Behavior 33(2): 159—166. Whalen, J., R. Whalen, M. R. and Zimmerman, D. H. (1990) Describing trouble: practical epistemology in citizen calls to the police. Language in Society 19: 465—492. Zhou, L, Burgoon, J. K., Zhang, D. and Nunamaker, J.F. (2004) Language dominance in interpersonal deception in computer-mediated communication. Computers in Human Behavior 20: 381—402. Zimmerman, D. H. (1984) Talk and its occasion: the case of calling the police. In D. Schiffrin (ed.) Meaning, Form, and Use in Context: Linguistic Applications. Georgetown University Roundtable on Language and Linguistics. Washington, D.C.: Georgetown University Press, 210—228. Zimmerman, D. H. (1992a) Achieving context: openings in emergency calls. In G. Watson and R. Seiler (eds) Text in Context: Contributions in Ethnomethodology. Newbury Park, NJ: Sage, 35—51. Zimmerman, D. H. (1992b) The interactional organization of calls for emergency assistance. In P. Drew and J. Heritage (eds) Talk at Work. Interaction in Institutional Settings. Cambridge: Cambridge University Press, 418—469.

-152-

The rake’s progress: linguistic strategies for deception Isabel Picornell Centre for Forensic Linguistics, Aston University, UK [email protected] Abstract Traditionally, approaches to deception detection have treated deceptive communication as if they were a one-time event, analysing the overall features and tone of the individual message. However useful this methodology may be, it fails to take account of the fact that deception is a progression of phases, rather than a single occurrence. This study looked at deception in written witness statements as just that. Although statements are written at a single point in time, their storylines are sequences of episodes over an extended timeline. Employing marked sentence structures to code discourse segmentation markers in written narratives, the progression of deception was mapped as it unfolded through the course of the story using the interaction of linguistic cues. In addition to identifying two main deceptive linguistic strategies which deceivers resorted to in writing their statements, results suggest that what may be important is not so much the individual cues, rather the way they are used. Keywords:

DECEPTION; STRATEGIES; WRITTEN TEXT; DISCOURSE SEGMENTATION

Introduction Research into deceptive behaviour to date has found that no single linguistic cue or set of cues occur only during deception and not at any other time (DePaulo et al, 2003; Zuckerman, DePaulo and Rosenthal, 1981). While some combination of cues appear to be more consistently reliable than others (DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper, 2003; Vrij, 2000), findings from different studies contradict each other, and no cues appear valid across topic, time, and contexts. Liars do not always lie in the same way; the Pinocchio’s nose of deception does not exist. Research has been based largely on theoretical assumptions regarding how deception affects individuals’ thoughts, feelings, and cognitive processes, and how this might be reflected in language when people lie compared to when they are telling the truth. Although results in this field have been contradictory, four linguistic features have been the subject of focus by deception researchers: word quantity, pronoun use, emotion words, and markers of cognitive processes. Linguistic deception features Liars may have moral qualms about their deception and fail to embrace their deception as fully as they do their truth telling (Weiner and Mehrabian, 1968). Thus, the guilt and negative emotions they may experience when lying leads deceivers to distance themselves from their lies. Research suggests that deceivers provide shorter responses compared to truth tellers (DePaulo, Lindsay, Malone, Muhlenbruck, Charlton and Cooper, 2003; Hartwig, Granhag, Stromwall and Kronvist, 2006; Vrij, 2008). This possibly arises from deceivers not wanting to provide too much information of an event for which they have no actual experience as saying too much would make it easier for them to be caught out with their lies (Vrij, 2008), particularly if they were to be re-interviewed and end up contradicting themselves (Granhag and Stromwall, 1999). Others suggest that deceivers find it too cognitively demanding to -153-

manage the misinformation (DePaulo et al, 2003), or think that their deception is obvious to others (Gilovich, Medvec and Savitsky, 1998; Savitsky and Gilovich, 2003). Consequently, it is easier to say as little as possible. However, other studies contradict this; liars, in fact, use more words when lying (Anolli and Ciceri, 1997; Anolli, Balconi and Ciceri, 2002; Burgoon, Buller, Floyd and Grandpre, 1996; Burgoon, Blair, Qin, and Nunamaker Jr, 2003; Hancock, Curry, Goorha, and Woodworth, 2004 and 2005; Zhou, Burgoon, Zhang, and Nunamaker, 2004; Zhou and Zhang, 2007). Believing their deception to be more transparent than it actually is, liars may feel the need to provide a complete story filled with the sort of detail they assume is typical of truthful recall (Burgoon et al, 1996), or they try to hide their lies within a mass of truthful albeit irrelevant information (Anolli, Balconi and Ciceri, 2002). Liars’ lack of embracement and their desire to disassociate themselves from their lies may also lead them to refer to themselves less often in communication. This results in a significant other focus (Newman, Pennebaker, Berry and Richards, 2003), with liars’ communications having significantly fewer First Person Singular Pronouns (e.g. I, me, and my) and more Third Person Pronouns than truth tellers (Burgoon, Blair, Qin and Nunamaker, 2003; Hancock et al. 2004 and 2005; Knapp and Comadena, 1979; Knapp, Hart and Dennis, 1974; Vrij, 2000; Weiner and Mehrabian, 1968; Zhou, Burgoon, Nunamaker and Twitchell, 2004). However, the use of Third Person Pronouns during deception remains problematic, with certain studies finding that truth tellers use more of them than deceivers (Bond and Lee, 2005; Newman et al. 2003; Zhou, Burgoon, Twitchell, Quin and Nunamaker, 2004). Guilt and negative emotion may also leak into liars’ language (Newman et al. 2003; Vrij, 2008), resulting in more negative statements (e.g. no, not, never) (Knapp and Comadena, 1979; DePaulo et al. 2003; Zhou et al. 2004) and negative emotion words (e.g. hate, angry, sad) than truth tellers (Newman et al. 2003). However, DePaulo et al. (2003) found only a weak association between negative comments and deception, while Hancock et al. (2004 and 2005) found no support for it. Deceivers’ lack of actual experience in imagined events and the cognitive processes involved in their manufacture may also be reflected in language. Memories derived from actually experienced events result in vivid sensory-derived accounts containing more visual, spatial, and contextual detail, while internally generated (imagined) accounts produce fewer details and more subjective information based on cognition processes (Johnson and Raye, 1981). Memories of imagined events degrade faster than those of experienced events (Suengas and Johnson, 1988), losing their clarity and becoming vague after only 24 hours. This reduction in detail and increased vagueness in imagined memories was earlier identified by Knapp, Hart and Dennis (1974) who found that liars were less specific and more vague than truth tellers (as did Burgoon et al. 2003; DePaulo et al. 2003; Vrij, 2000), used more levelling words suggesting allness (e.g. every, all, none etc), fewer absolute verbs and more modal verb constructions (a feature also identified by Buller and Burgoon, 1994). However, studies into words identifying cognitive processes (e.g. think, know, appear) as a deception marker contradict each other, with some studies reporting increased cognitive information in deceptive accounts, while others identify more cognitive operations in truthful accounts (see review by Masip, Sporer, Garrido and Herrero, 2005). Vrij (2000) initially dismissed the use of cognition words as a lie criterion, arguing that people used such operations to encode their experiences into memory; however, Vrij, Mann, Fisher, Leal, Milne and Bull (2008) later found that deceivers used more cognitive operations in their speech than truth-tellers when recalling events in reverse order.

-154-

Deceptive communication theories Buller, Strzyzewski and Hunsaker (1991) describe the logistics of deceptive conversation as ‘characterised by a series of moves and countermoves’ (p. 28), the success of which rests on deceivers’ skills to adapt to their addressees. In other words, deceptive language should be viewed as strategic adaptations, with deceivers making the most of opportunities presented to them whilst minimising the risks to themselves. They can choose from any number of strategies that involve manipulating the amount, veracity, relevance, and clarity of information as and when the circumstances required it (McCornack, 1992). Buller and Burgoon (1994) also see deception as adaptive, describing deceivers as constantly tweaking linguistic and non-verbal behaviour in response to their addressees’ reactions. The best deception strategy is a flexible one, as this ensures maximum success for deceivers while minimising the risk of being caught lying. Anolli, Balconi and Ciceri (2002) bring together the range of deceptive linguistic patterns identified by other researchers under one ‘family of deceptive acts’ in their Deceptive Miscommunication Theory (DeMiT). They suggest that the selection of a particular linguistic strategy at any one time is dictated by the context in which the deception occurs, and depends on whether it is pre-planned or spontaneous, and how addressees react towards the deceptive message. This ability to strengthen or weaken the message in response to addressees’ behaviour not only provides deceivers with the flexibility to protect themselves from discovery, but also allows them to meet addressees’ expectations as to what might constitute normal truthful communication (Dunbar, Ramirez and Burgoon, 2003). However, this interaction between deceiver and addressee does not exist in all communication formats. Unlike the United Kingdom, where witness statements tend to be a product written by the police with the cooperation of the witness, certain police authorities in the USA require that witnesses write their voluntary statements without providing any direction other than to ‘write what happened’. Such statements provide an opportunity to study deceptive language at work outside the experimental laboratory situation, in a noninteractive real-life serious-crime context. On one hand, lying in writing may appear simpler than face-to-face deception as there is no addressee to adapt to, and deceivers have the time to plan their lie which they would not otherwise have. However, they still have to decide how to convey the false account in a way that will appear balanced and consistent with the known truths; without the luxury of feedback, deceivers have to guess as to how best to structure their deception so as to appear truthful and convincing. Deceivers already hope that they are perceived as being cooperative, given that their witness statements are voluntary. In providing the police with information that may appear relevant to the course of a criminal investigation, they also hope to appear truthful. However, deceivers are faced with the dilemma of how to provide enough information (to appear truthful and cooperative), but not so much information that could otherwise be verified through external evidence. Providing the minimum of information would not be very convincing, but the statement could be padded out with irrelevant truths and/or ambiguous and subjective information which would be difficult to confirm. Although deceivers must include truthful and deceptive information in the one statement, they intend that addressees infer only one truthful message. To achieve this, deceivers must manipulate both the truthful and deceptive content to construct a seamless narrative, adjusting their language accordingly as they progress so that the account appears cohesive and credible; the boundary between truth and lies must be blurred so that the transition between one to the other is softened and the readers’ default assumption of continuity is maintained (Segal, Duchan and Scott, 1991). However, deceivers cannot use the -155-

same language for truthful and deceptive content simply because they cannot provide the same level of detail for both. Thus, deceivers’ linguistic strategies must accommodate their intention to present deceptive information as if it were true and appear committed to their story so that addressees will believe their message. This requires that deceivers adopt a linguistic style which presents deceptive information in such a way that their entire statements meet addressees’ expectations of truthfulness (Wilson and Sperber, 2000; Wilson and Sperber, 2002). At the same time, should addressees become suspicious and discover the deception, then the strategy must also allow deceivers leeway to excuse the misinformation as arising out of confusion, memory failure, or a third party source. Deception as a progression Researchers have tended to explore deception as if it were a separate communication activity with specific identifiable characteristics which distinguish it from normal truthful expressions. Anolli, Balconi and Ciceri (2002) argue that this perception is a myth, that deceptive communication is no different from default truthful communication. What makes deception different is its internal gradient, the series of successive changes to the language which allow deceivers to design a strategy best suited to the particular situation. Baumeister (1993) also saw deception as arising out of a graduation of ‘possibilities, probable truths, working assumptions, leaps to conclusion’ (p. 167); convincing oneself (and others) of the truth of shades of grey was easier than believing black to be white. Previous research has also tended to analyse deceptive communication as if the deception consists of a single act, reflected in a single statement or a single interview. Literature and legislation often refer to ‘an act of deception’, as if deception was achieved by a single behaviour. In fact, ‘deception is a continous process rather than a one-time event’ (Zhou, Burgoon and Twitchell, 2003: 102), a progression of acts over time (White and Burgoon, 2001), where information is managed and manipulated to achieve a final deceptive presentation. When deception is viewed as a single act, it is easy to miss the more subtle behaviours that can be identified when the graduation process itself is examined. If one subscribes to the view that deception is a gradual progression (Anolli, Balconi and Ciceri, 2002; White and Burgoon, 2001), it stands to reason that deceptive language changes over the course of the deception. Indeed, evidence exists in support of this. Zhou, Burgoon and Twitchell (2003) found that, at least with CMC messaging, deceptive cues vary in quantity and type over the time of the deception. In their study involving email exchanges between truthful and deceptive senders over three days, they found that none of the cues were effective differentiators between truth and deception for all the days, and that some cues were stronger at a particular stage of the deception than at others. Zhou et al. were unsure whether the time difference (over several days) or the stage of the deception, or both, influenced the use of cues; but what mattered was the linguistic analysis of a progressive deception strategy. Unlike face-to-face and CMC messaging, written statements have no turn-taking which produces separate linguistic bundles which can be analysed independently of each other. This creates difficulties when trying to analyse linguistically the strategic progression of a deceptive message. Adams and Jarvis (2006) examined differences in language within witness statements by partitioning narratives into prologue, criminal event, and epilogue sections. This is not ideal as partitioning relies on the judgement of analysts as to where the dividing line lies between the relevant sections; furthermore, this results in a maximum of three linguistic bundles (less if the information which defines a particular section is missing) to work with. However, there is an aspect of narrative construction, based on identifying the

-156-

author’s signalling mechanism of discourse discontinuity, that lends itself to statement partition. Episode partition of narratives Studies into how people construct narratives (Bestgen, 1998; Bestgen and Vonk, 1995; Bestgen and Vonk, 2000; Prideaux, 1989; Prideaux and Hogan, 1993; Virtanen, 2008) have found that narrators create thematic bundles (episodes) by using grammatical signals as segmentation markers to manage the flow of information in a story and facilitate readers’ understanding of events. Prideaux’s (1989) work on the role of marked sentence structures in narratives suggests that deviation from the standard sentence construction of main clause + subordinate clause to that of subordinate clause + main clause is not just a stylistic preference, but serves the reader as a kind of linguistic signposting. These structures alert readers to changes in continuity and the introduction of new discourse packages (Bestgen et al., 2009); more importantly, they signal discontinuity in discourse (Bestgen, 1998). Unexpected changes in sentence structures highlight information contained in the sentence and draw readers’ attention to major changes in topic or important incidents (Prideaux and Hogan, 1993) as well as to temporal shifts in the narrative (Bestgen and Vonk, 2000; Prideaux, 2000). Most importantly, the decision where and when to use these structures rests solely with the author, and provides insight into the author’s focus when deciding to introduce a thematic break (Bestgen, 1998). Anolli, Balconi and Ciceri (2002) argue that ‘focus is driven by intentions and guided by the route the communicator formulates to move through the field of thought’ (p. 84). In other words, language is influenced by the author’s focus, and that focus is in turn managed by the author’s intentions and how s/he intends to achieve it. In deception terms, the finished product (the deception) is a compilation of the deceiver’s intention (to tell the truth or to lie) and choice of communication strategy to achieve that intent. The linguistic analysis of individual episodes allows us to map how deceivers work on their deception as they progress through their story. At the very least, examining changes in language across different episodes in a single deceptive statement provides an opportunity to review the author’s linguistic strategy for deception-building, and identify how deceivers juggle linguistic features associated with deception within their own statements. Strategic association of cues: the study To examine the progression of deception within individual statements, linguistic features of 10 deceptive written witness statements were analysed on an episode by episode basis. This small study forms part of a larger research agenda directed at a more systematic approach to the detection of deception, in which 40 written witness statements (22 deceptive and 18 truthful statements) were analysed. The 10 statements were part of an initial group of 15 statements (10 deceptive and 5 truthful statements—labelled Group 1) provided by a police authority, and conclusively identified as deceptive based on external evidence. For the purpose of this analysis, it was not necessary to differentiate between truthful and deceptive episodes within the statements as the study’s intention was to identify the progression of linguistic features throughout individual statements, and not seek to differentiate between truthful and deceptive style. All the statements relate to serious crime (e.g. arson, sexual assault, murder/manslaughter etc), were hand written by American English speakers and voluntarily provided to the police. The linguistic features analysed were selected on the basis of previous deception studies, although certain categories were created or re-defined to remove any ambiguity in -157-

their identification. The cues identified were First Person Singular Pronouns, Third Person Pronouns, Vague Pronouns References (e.g. someone, something, one), Verb Strings (two or more verbs functioning as a single verb e.g. trying to do, started running, proceeded to tell etc), Cognitive Verbs (any verb implying a cognitive process e.g. think, seem, appear etc), and Negation (no, not, never and verbs prefixed by un- which replace not, e.g. unsuccessful). A set of cues (referred to as MoDs—markers of deception) was also created, consisting of Vague Pronoun References, Verb Strings, Cognitive Verbs and Negation, which could be combined together to create non-immediacy through negation or ambiguity, following on from DePaulo et al.’s (2003) argument that cues in combination were more productive deception markers than on their own. For example, the sentence I do not remember seeing someone behind me contains a Negation (not), a Cognitive Verb (remember), a Verb String (remember seeing), and a Vague Pronoun Reference (someone). In addition to being analysed collectively, the Pronouns I, me and my were also analysed individually, as research into these pronouns by Campbell and Pennebaker (2003) suggests that they are not all weighted equally, and an individual’s focus may influence choice of one or the other. Comparing the active I and the passive me results in two very different perceptions of a person’s role in a particular scenario—as agent and as recipient of the action. Individual statements were deconstructed into a sequence of episodes, with marked sentence structures (subordinate clause + main clause) identifying the start of each new episode. Cues were manually identified, tagged, and analysed according to the number of clauses in which they appeared in the whole statement, as well as in individual episodes. Any number of appearances of the same cue in a single clause counted as a single occurrence. For example, multiple use of my (e.g. This guy was landing his punches to my face, my back, and back of my head) or of Third Person Pronouns (e.g. if he got his hands on his pistol) in a single clause was treated as one cue. Similarly, where more than one cue from the MoDs set appeared in the same clause (e.g. I did not seem to think—Negation + Verb String + Cognitive Verb), it was labelled as a single MoDs occurrence. Results On a whole statement level, the highest use of MoDs clauses (≥23%) were associated with statements having the longest clauses (≥7 words per clause) and the highest use of I clauses (≥30%). Statements with the highest use of First Person Singular Pronoun clauses (≥50%) also had the highest use of my (≥15%) and/or me (17%) clauses, with combined use of my and me equalling or exceeding I. Statements with the highest use of my clauses also contained the lowest number of I clauses (≤28%). There also appeared to be an association between the highest use of Third Person Pronoun clauses (≥49%), short clauses (≤5.5 words per clause), and low levels of I clauses (≤28%), while statements with the lowest use of Third Person Pronouns also had the longest clauses (≥7 words per clause). The association between ambiguity, verbosity and deception is consistent with previous literature (Buller and Burgoon, 1994; Dulaney, 1982; Hancock et al. 2005; Knapp, Hart and Dennis, 1974; Kuiken, 1981) as is the association between Third Person Pronouns, short clauses and deception (Anolli, Balconi and Ciceri, 2002; Buller and Burgoon, 1994). However, the association between First Person Singular Pronouns and deception is unusual as previous research supports a negative correlation between self-references and deception (Burgoon, Buller, Floyd and Grandpre, 1996; Burgoon, Buller, Guerrero, Afifi and Feldman, 1996; DePaulo, Rosenthal, Rosenkrantz and Green, 1982; Ebesu and Miller, 1994; Newman et al. 2003). However, in statements with very high First Person Singular Pronouns, numbers were driven up through high use of me and my, which suggest an association between them and deception. If true, then I, me and my do not appear to be weighted equally. -158-

The association between I and MoDs and deception also has no parallel in previous research. DePaulo et al’s (2003) review of studies relating to verbal cues associated with deception identifies a general (albeit weak) trend wherein liars referred to themselves less often than do truth tellers. However, if deceivers prefer to be non-immediate and stand back from their lies (Dulaney, 1982; Knapp, Hart and Dennis, 1974; Mehrabian, 1971; Weiner and Meharbian, 1968) then the association between I and MoDs is not so unexpected as it allows deceivers to be immediate (and appear to embrace their story) yet still be non-committal. The association between cues is better explained by analysing the way they interact with each other as deception progresses, which we are able to do for the first time through the deconstruction of statements into their constituent episodes. Analysing cues on a whole statement level only provides a partial picture as it identifies cues in isolation, missing out on their interaction with each other and how their use waxes and wanes as deceivers linguistically build their deception. This could be compared to the difference between a single snapshot and a video, where detailed analysis of cue behaviour through the episodes may reveal features of deception strategy development not noticeable on the whole statement level. It could also explain some of the contradictory findings of previous studies, whose whole text view may not necessarily explain what is happening in the episodes. Strategy 1 Statement 1 is representative of this strategy found in a number of other statements. Overall, the statement makes very high use of First Person Singular Pronouns (59%) with high use of I (30%), me (17%) and my (14%) clauses. However, these figures do not capture the distribution of the individual pronouns throughout the episodes. In Episode 1, where the scene is set for the deception (which progresses through six episodes), I dominates. Extract 1: Statement 1, Episode 1

I woke up at approx 1800hrs Sunday evening I got ready for work and left my home in Molton at approx 2000hrs. I then went to the Kwik Shop in Ripon to buy a soda as I do every night I work at around 2015 hrs to 2020 hrs. As the deception progresses, the author gradually reduces his use of I in favour of me and my, while Third Person Pronouns (which overall are low) make an incremental appearance from Episode 2 onwards. This focuses attention on what is being done to the author (or to parts of the author), rather than on the author’s own actions. Where I is present, it tends to be associated with MoDs, introducing ambiguity and subjective information that pad out the narrative without contributing much relevant information. Extract 2: Statement 1, Episode 3

at that time I noticed the passenger getting out of the car and walking towards us. he wasn’t running at me or anything but kinda like coming out to check out what was going on with his buddy or something like that. right around the driver pushed me and hit me Extract 3: Statement 1, Episode 6

next thing I remember it the driver on top of me punching me around my chest and I trying to grab arms…He was hitting around my left side and I felt something cold. He went to hit me again and something cut the palm of my left hand and it scared the -159-

hell out of me so I grabbed my side curled up and started yelling and screaming. I think they kicked me a little (but I’m not sure) but he got off me and left me alone for a second. This juggling of cues can be seen in Figure 1 below (very short episodes are omitted), which capture the changing interaction and re-balancing of linguistic cues throughout the narrative’s episodes.

Episode 1

Episode 2 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Episode 3

Episode 6 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Episode 7

Episode 8 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Figure 1: Episode progression of cues in Statement 1

This strategy can be described as personal, verbose, and ambiguous. The deceiver uses many words, longer clauses, and a high number of First Person Singular Pronouns, which ensures that he is present in one form or another throughout the narrative so as to give an appearance of immediacy. However, distancing is achieved through high use of my and/or me which allows the deceiver to step back from fully embracing the information he provides. MoDs are -160-

widely used and heavily linked to I clauses, which gives the impression of information but which in fact is subjective, ambiguous, and irrelevant. Strategy 2 Statement 2 makes use of a second strategy which involves constructing an impersonal context through high use of Third Person Pronouns; this creates an increasingly other focus as the deception progresses through the narrative. I use is low (compared to Strategy 1) as the deceiver takes a largely passive role through me and/or my.

Episode 1

Episode 2 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Episode 4

Episode 6 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Episode 8

Episode 10 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Figure 2: Episode progression of cues in Statement 2

-161-

Extract 4: Statement 2

I started to check on him and he pushed me in my van and then made me drive then he told me to stop crying. He told me to stop the van. He took tape out and put it on my hands and then he cut my lips with his knife and then put tape on my face. There is use some of MoDs, but the cues tend to be isolated and without the clustering found in Strategy 1, possibly because the deceiver has no need for it. Ambiguity is not necessary here as the strategy is not to blur the action but to distance the deceiver from it by ascribing the action to someone else. In Statement 2, episodes 1 to 3 provide a truthful introduction. The fictitious account begins in Episode 4 (see Figure 2) which sees the first appearance of MoDs (all associated with I), as well as the beginning of a shift in focus away from the deceiver as agent of the action (using I) to a more passive role (increasing use of me and my). Third Person Pronouns also made their first appearance in Episode 4 and proceeded to dominate the rest of the narrative. Throughout the episodes, the author juggles me, my and Third Person Pronouns to move from one form of distancing to another. The narrative does not return to the truthful condition and remains other focused to the last episode. Extract 5: Statement 8

a female entered the building and tried to walk past me and not go through the metal detector. She told me she was an employee. I stated that she still had to go through the metal detector. When she was going through the metal detector she didn’t put her bags on the xray machine. She put her bags on the xray machine in a frantic manner. She jumped in front of another person and pushed to the desk. Statement 8 (Figure 3) contains a variant of the impersonal approach. In this completely fictitious statement, Third Person Pronouns predominate; they are found in 50% of clauses compared to only 17% of First Person Singular Pronouns. Thus, the imbalance of pronouns identifies a depersonalised statement. Episode 1 starts impersonal, with high levels of Third Person Pronouns that continue throughout the rest of the episodes, overshadowing any I presence.

Episode 1

Episode 2 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

-162-

Episode 3

Episode 4 I

I

me

me

my

my

3pp

3pp

MoDs

MoDs

Other

Other

Figure 3: Episode Progression Statement 8

A third variant of the strategy is used in Statement 3, which achieves depersonalisation through high use of First Person Plural Pronouns. These pronouns were not included in this study as their interpretation (immediacy or distancing) is heavily dependent on the context in which they are located. We and our can be interpreted as positive indicators of group identity, with the author including himself in a group (Pennebaker and Lay, 2002), or an author could use the pronouns to evade or share responsibility by referring to himself acting in association with others. The pronouns could also be used in a way that excludes the author from the group altogether (Tausczik and Pennebaker, 2010). Extract 6: Statement 3

My wife, myself and my son came home today. We stopped and talked to our friend Alec downstairs. He was telling us …Then me, my wife, my son and our roommate Pete went upstairs. When I unlocked the door we could all smell something funny in the house. In this account, high use of First Person Plural Pronouns combined with high use of my and low use of I has the same effect as Third Person Pronouns, as the collective view dilutes the deceiver’s role as an individual. He still maintains a presence in the narrative, but focus is firmly on the collective actions of the deceiver together with others. Discussion On the basis of the association of cues, two main deceptive strategies are identified. In Strategy 1, deceivers are verbose, immediate, and ambiguous, creating an impression of being Cooperatively Vague. The strategy may be a device for deceivers to construct for themselves the world of the lie (private correspondence, Prideaux, 2011). They populate their version of reality with plenty of information (cooperative) with which they are fully involved (immediate) through high use of self-references, but which is rendered safe as the information is neither relevant nor clear. This distancing strategy, whereby deceivers are verbose but non-committal, has been described in earlier literature in conversational contexts (Anolli, Balconi and Ciceri, 2002; Buller and Burgoon, 1994; Dulaney, 1982; Knapp, Hart and Dennis, 1974; Kuiken, 1981) and in CMC communication (Hancock, Curry, Goorha and Woodworth, 2005). Anolli, Balconi and Ciceri (2002) suggested that this combination of verbosity and ambiguousness was the strategy favoured by deceivers facing silent or acquiescent addressees. Silence as part of an -163-

interaction can be uncomfortable. It can convey negative connotations (Leech, 1983), which may explain why some people feel the need to fill it. It may be that in a non-interactive context, where deceivers are left alone to write and contemplate their statement, some deceivers revert to the same strategy, preparing long statements filled with much subjective and other unverifiable information in order to appear polite and cooperative. In Strategy 2, the deceivers’ approach is an Impersonal one. Clauses are short, there is high use of Third Person Pronouns and low use of self-references, with a preference for me and/or my instead of I. In this strategy, deceivers distance themselves by rejecting responsibility for the deception rather than achieving it through ambiguity; they ascribe the actions to someone else or hide behind a collective association. In their Interpersonal Deception Theory, Buller and Burgoon (1994) call this strategy disassociation; Anolli, Balconi and Ciceri (2002) refer to it as depersonalisation. While these strategies may be identified in some statements by analysing cues on the whole statement level, it may not as easily be identified in other statements except through episode analysis. This is particularly the case if deceivers choose to follow more than one strategy. As the deception progresses, cues subtly interact with each other to strengthen or weaken the deceptive message. However, a sudden change in style when deceivers change linguistic direction, which would be immediately detectable at the episode level, may be missed when analysing cues for the statement as a whole. Why deceivers choose to use a particular strategy is unclear. No one strategy appears to be associated with any particular crime in this group of statements. However, deceivers do appear to favour the Cooperatively Vague strategy for statements relating to entirely fictitious accounts. It may be that where deceivers are unconstrained by other witnesses or a strict timeline, they feel freer to elaborate their lie, resulting in a more immediate and verbose, albeit ambiguous, style. Conclusion The results support the view that deception is a gradual progression of acts and not a single event (Anolli, Balconi and Ciceri, 2002; White and Burgoon, 2001; Zhou, Burgoon and Twitchell, 2003). By deconstructing written witness statements into episode sequences using their authors’ own linguistic signals, we are able to follow liars through their field of thought as they engage in a dynamic pattern of deception. Lacking the interaction with other communicators on which they rely for feedback, deceivers instead opt for certain default communication strategies which they hope will signal cooperation, consistency, and credibility. These strategies approach deception in a subtly incremental manner, as deceivers constantly adjust and re-balance their use of linguistic cues to write as much as they are able to write without saying too much or associating too closely with the lie. The changing cue interactions that takes place as deception progresses through the episodes suggests that what matters is not what cues are used, but how they are used, which may go some way towards explaining the contradictory results of previous studies. This study confirms that deception is much more varied, more subtle, and more strategic than simply defaulting to a specific cue (or set of cues) in a particular way. Success in deception, like evolution, favours the flexible liar (DePaulo et al, 2003). Acknowledgement I would like to acknowledge and thank Sergeant Larry Barksdale of the Lincoln, Nebraska Police Department for his assistance in the collection of data for this study.

-164-

References Adams, S.H. and Jarvis, J.P. (2006) Indicators of veracity and deception: an analysis of written statements made to police. International Journal of Speech, Language and the Law 13 (1): 1—22. Anolli, L., Balconi, M. and Ciceri, R. (2002) Deceptive miscommunication theory (DeMiT): a new model for the analysis of deceptive communication. In L. Anolli, R. Ciceri and G. Riva (eds) Say Not To Say: New Perspectives in Miscommunication. Amsterdam: IOS Press, 73—100. Anolli, L., Balconi, M. and Ciceri, R. (2003) Linguistic styles in deceptive communication: dubitative ambiguity and elliptic eluding in packaged lies. Social Behavior and Personality 31: 687—710. Baumeister, R.F. (1993) Lying to yourself: the enigma of self-deception. In M. Lewis and C. Saarni (eds) Lying and Deception in Everyday Life. New York: The Guildford Press, 166—183. Bestgen, Y. (1998) Segmentation markers as trace and signal of discourse structure. Journal of Pragmatics 29: 753—763. Bestgen, Y. and the Psycholinguistics Group of the Spatial Framing Adverbials Project (2009) The discourse function of sentence-initial adverbials: studies in Comprehension. Linguistic and Psycholinguistic Approaches to Text Structuring Conference, Paris, 21—23 September 2009, 7—14. Bestgen, Y. and Vonk, W. (1995) The role of temporal segmentation markers in discourse processing. Discourse Processes 19: 385—406. Bestgen. Y. and Vonk, W. (2000) Temporal adverbials as segmentation markers in discourse comprehension, Journal of Memory and Language 42: 74—87. Bond, G.D. and Lee, A.Y. (2005) Language of lies in prison: linguistic classification of prisoners’ truthful and deceptive natural language. Applied Cognitive Psychology 19: 313—329. Buller, D.B. and Burgoon, J.K. (1994) Deception: strategic and nonstrategic communications. In J.A. Daly and J.M. Wieman (eds) Strategic Interpersonal Communication. New Jersey: Erlbaum, 191—223. Buller, D.B. and Burgoon, J.K. (1996) Interpersonal deception theory. Communication 6: 203—242. Buller, D.B., Strzyzewski, K.D. and Hunsaker, F.G. (1991) Interpersonal deception 2: the inferiority of conversational participants as deception detectors. Communication Monographs 58: 1—24.

-165-

Burgoon, J.K., Blair, J.P., Qin, T. and Nunamaker Jr, J.F. (2003) Detecting deception through linguistic analysis, Proceedings of the Symposium on Intelligence and Security Informatics. New York: Springer-Verlag, 91—101. Burgoon, J.K., Buller, D.B., Floyd, K. and Grandpre, J. (1996) Deceptive realities: sender, receiver, and observer perspectives in deceptive conversations. Communication Research 23: 724—748. Burgoon, J.K., Buller, D.B., Guerrero, L.K., Afifi, W.A. and Feldman, C.M. (1996) Interpersonal deception: XII. Information management dimensions underlying truthful and deceptive messages. Communication Monographs 63: 50—69. Campbell, R.S. and Pennebaker, J.W. (2003) The secret life of pronouns: flexibility in writing style and physical health. Psychological Science 14 (1): 60—65. DePaulo, B.M., Lindsay, J.J., Malone, B.E., Muhlenbruck, L., Charlton, K. and Cooper, H. (2003) Cues to deception. Psychology Bulletin 129 (1): 74—118. DePaulo, B.M., Rosenthal, R., Rosenkrantz, J. and Green, C.R. (1982) Actual and perceived cues to deception: a closer look at speech. Basic and Applied Social Psychology 3: 291—312. Dulaney Jr, E.F. (1982) Changes in language behavior as a function of veracity. Human Communication Research 9: 75—82. Dunbar, N.E., Ramirez, Jr., A. and Burgoon, J.K. (2003) The effects of participation on the ability to judge deceit. Communication Reports 16 (1): 22—33. Ebesu. A.S. and Miller, M.D. (1994) Verbal and nonverbal behaviors as a function of deception type, Journal of Language and Social Psychology 13 (4): 418—442. Gilovich, T., Medvec, V.H and Savitsky, K. (1998) The illusion of transparency: biased assessments of other’s ability to read one’s emotional states. Journal of Personality and Social Psychology 75 (2): 332—346. Granhag, P.A. and Stromwall, L.A. (1999) Repeated interrogations: stretching the deception detection paradigm. Expert Evidence 7: 163—174. Hancock, J.T., Curry, L.E., Goorha, S. and Woodworth, M.T. (2005) Automated linguistic analysis of deceptive and truthful synchronous computer-mediated communication. HICCS 1: 22c. Hancock, J.T., Curry, L.E., Goorha, S. and Woodworth, M.T. (2004) Lies in conversation: an examination of deception using automated linguistic analysis. Proceedings of the 26th Annual Conference of the Cognitive Science Society 26: 534—540. Hartwig, M., Granhag, P.A., Stromwall, L.A. and Kronvist, O. (2006) Strategic use of evidence during police interviews: when training to detect deception works. Law and Human Behavior 30: 603—619.

-166-

Johnson, M.K. and Raye, C.L. (1981) Reality Monitoring. Psychological Review 88: 67—85. Knapp, M.L. and Comadena, M.E. (1979) Telling it like it isn’t: a review of theory and research on deceptive communications. Human Communication Research, 5 (3): 270—285. Knapp, M.L., Hart, R.P. and Dennis, H.S. (1974) An exploration of deception as a communication construct. Human Communication Research 1: 15—29. Kuiken, D. (1981) Nonimmediate language style and inconsistency between private and expressed evaluations. Journal of Experimental Social Psychology 17: 183—196. Masip, J., Sporer, S.L., Garrido, E. and Herrero, C. (2005) The detection of deception with the Reality Monitoring Approach: a review of the empirical evidence. Psychology, Crime and Law 11 (1): 99—122. Mehrabian, A. (1971) Nonverbal betrayal of feeling. Journal of Experimental Research in Personality 5: 64—73. McCornack, S. A. (1992) Information manipulation theory. Communication Monographs 59: 1—16. Newman, M.L., Pennebaker, J.W., Berry, D.S. and Richards, J.M. (2003) Lying words: predicting deception from linguistic styles. Personality and Social Psychology Bulletin 29: 547—675. Pennebaker, J.W. and Lay, T.C. (2002) Language use and crisis: an analysis of Mayor Rudolph Giuliani’s press conferences. Journal of Research in Personality 36: 272— 281. Prideaux, G.D. (1989) Text data as evidence for language processing principles: the grammar of ordered events. Language Sciences 11 (1): 27—42. Prideaux, G. D. (2000) Discourse analysis and psycholinguistics: breaking down barriers. In A. Lommel and A. Melby (eds) LACUS Forum 26: 1—16. Prideaux, G. D., and Hogan, J.T. (1993) Markedness as a discourse management device: the role of alternative adverbial clause orders. Word 44: 397—411. Savitsky, K. and Gilovitch, T. (2003) The illusion of transparency and the alleviation of speech anxiety. Journal of Experimental Social Psychology 39: 618—625. Segal, E.M., Duchan, J.F. and Scott, P.J. (1991) The role of interclausal connectives in narrative structuring: evidence from adults’ interpretation of simple stories. Discourse Processes 14: 27—54. Suengas, A.G. and Johnson, M.K. (1988) Qualitative effects of rehearsal on memories for perceived and imagined complex events. Journal of Experimental Psychology: General 117 (4): 377—389.

-167-

Tausczik, Y.R. and Pennebaker, J.W. (2010) The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology 29 (1), 24—54. Virtanen, T. (2008) Adverbials of ‘manner’ and ‘manner plus’ in written English: why initial placement? SKY Journal of Linguistics 21: 271—293. Vrij, A. (2000) Detecting Lies and Deceit: The Psychology of Lying and the Implications for Professional Practice. Chichester: John Wiley and Sons. Vrij, A. (2008) Detecting Lies and Deceit: Pitfalls and Opportunities. Chichester: John Wiley and Sons Ltd. Vrij, A., Mann, S., Fisher, R.P., Leal, S., Milne, R. and Bull, R. (2008) Increasing cognitive load to facilitate lie detection: the benefit of recalling an event in reverse order. Law and Human Behavior 32: 253—265. Vrij, A., Mann, S., Kristen, S. and Fisher, R.P. (2007) Cues to deception and ability to detect lies as a function of police interview styles, Law and Human Behavior, 31: 499—518. Weiner, M. and Mehrabian, A. (1968) Language within Language: Immediacy, a Channel in Verbal Communication. New York: Appleton Century Crofts. White, C.H. and Burgoon, J.K. (2001) Adaptation and communicative design: patterns of interaction in truthful and deceptive conversation. Human Communication Research 27: 9—37. Wilson, D. and Sperber, D. (2000) Truthfulness and relevance. UCL Working Papers in Linguistics 12: 215—254. Wilson, D. and Sperber, D. (2002) Relevance theory. UCL Working Papers in Linguistics 14, 249—287. Zhou, L., Burgoon, J.K. and Twitchell, D.P. (2003) A longitudinal analysis of language behavior of deception in email. In H. Chen, R. Moore, D. Zeng and J Leavitt. (eds) Intelligence and Security Informatics 2665: 102—110. Zhou, L., Burgoon, J.K., Twitchell, D.P., Qin, T. and Nunamaker, Jr., J.F. (2004) A comparison of classification methods for predicting deception in computer-mediated communication. Journal of Management Information Systems 20 (4): 139—165. Zhou, L., Burgoon, J.K., Nunamaker Jr, J.F. and Twitchell, D. (2004) Automating linguisticbased cues for detecting deception in text-based asynchronous computer-mediated communication. Group Decisions and Negotiation 13: 81—106 Zuckerman, M., DePaulo, B.M. and Rosenthal, R. (1981) Verbal and nonverbal communication of deception. In I Berkowitz (ed.) Advances in Experimental Social Psychology 14. New York: Academic Press.

-168-

An historical forensic linguistic analysis of contested letters in the Forrest Reid collection Yvonne McGivern Queen's University Belfast, United Kingdom [email protected] Abstract This paper offers an historical forensic linguistic analysis of two disputed letters from the Forrest Reid Collection (MS44) at Queen’s University Belfast. Reid (1875—1947), a novelist and critic, was described in 1937 as ‘one of the three or four most distinguished living writers of English’ (Fitz–Simon, 1998: 5—6). The letters, dated 1926, are recorded in the Collection catalogue as: ‘Short pencil manuscripts by FR purporting to be love letters from Sadie. Melodramatic in tone. Perhaps notes for character FR developing.’ There is, however, no unequivocal external evidence—from within the Collection or in biographical work on Reid—to support the claim that the letters are sketches for creative development or that they were authored by Reid. It has also not been possible to establish the identity of ‘Sadie’ using external sources, nor from the content of the letters themselves. Furthermore, it is unlikely that Reid, in view of his welldocumented homosexuality, was the ‘Beloved’ addressee of the letters, unless of course ‘Sadie’ is the pen name of a male correspondent. This paper therefore investigates the probability of Reid being the author of the Sadie Letters. The possibility that these documents could belong to one of two ‘populations of writing’—creative-developmental work or genuine personal letters—has interesting implications for research design, because in this instance the contested documents can be placed against a substantial body of known writings by the Reid. In the paper, I address these methodological issues and identify key forensic linguistic markers which include grammar and syntax, and sociolinguistic features. Keywords:

AUTHORSHIP ATTRIBUTION; QUESTIONED DOCUMENTS; FORENSIC LINGUISTIC ANALYSIS; EXTERNAL EVIDENCE

1. Introduction Forrest Reid was born in 1875 in Belfast where, apart from time as a student at Christ’s College, Cambridge, he lived until his death in 1947. He is the author of 17 novels, two works of autobiography, Apostate (1926) and Private Road (1940) and critical works on W. B. Yeats and Walter de la Mare. He won the James Tait Black Memorial Prize for the final part of his Tom Barber trilogy, Young Tom (1944). In 2007 Queen’s University Belfast acquired a collection of his letters, manuscripts and notes from his friend, the novelist Stephen Gilbert (1912—2010). The letters which are of interest here were contained in a folder labelled ‘Notes on Pictures’. Besides notes on pictures, this folder contained a list of book collections; travelogue type notes dated 1911; an article on Greek mythology; and some story sketches. The following entry in the archived Collection’s catalogue describes the letters (Brown, 2008:78):

-169-

MS44/13/2 Love Letters from ‘Sadie’ 2 AMS. Short pencil manuscripts by FR purporting to be love letters from Sadie. Melodramatic in tone. Perhaps notes for character FR developing. 1 2

9/5/1926 18/5/1926

n.p. n.p.

ALS, 2p. ALS, 1p.

Letter is torn across.

Figure 1. Photograph of part of Sadie Letter QD1 Copyright of Queen's University Belfast, Forrest Reid Manuscript Collection, MS 44/13/2. Used with permission.

Both letters are written on the same type of cream writing paper, measuring 18.5cm by 23cm (7 x 9 inches); the paper on which Reid was writing letters at this time was similar but larger in size (20cm x 25.75cm). They are written in pencil. Taylor (1980: 155) notes that Reid wrote first drafts of his books and articles with a 2B Koh-i-Noor pencil, a popular brand at the time. Most of Reid’s creative-developmental work is in notebooks, on looseleaf foolscap pages or on scrap paper which tends to be gathered in folders or pinned together. The letters were found loose, without envelopes. Both are folded and folded again, as if to be placed in an envelope; in this they are similar to other letters verifiably Reid’s. As Figure 1 above shows, the letter dated ‘May 9th 26’ is torn into two parts and has a scruffy, smudged appearance. The letter dated ‘18 May’ is in better condition with no smudging and no tears. My interest in the Collection was as a source for both the critical and the creative elements of my PhD thesis. For the creative element, a novel set in Belfast in the 1920s, I planned to use the Collection as a research resource for the life and times of the period, and for language use. For the critical research, I planned to investigate Reid’s working methods. This changed, however, when I read the ‘Sadie’ letters. In the context of both the other correspondence in the Collection and the other creative-developmental work within it, they stand out. Thus I wondered about their attribution to Reid as ‘notes for character ...’. I decided to investigate the attribution further. -170-

As a first step I examined what Love (2002: 51) refers to as external evidence, ‘that from the social world within which the work is created, promulgated and read’. This involved an indepth review of the following: critical and biographical work on Reid; Reid’s published work; other material in the Forrest Reid Collection including letters and creative-developmental work; and references within the disputed letters (a sort of internal external evidence). The critical and biographical studies of Reid (Bryan, 1976; Burlingham, 1953; Cruise, 1997; Goldman and Taylor, 1998; Taylor, 1980) support the view that it is unlikely Reid would have received love letters from a woman. Hughes (1992: 11) notes that in reading Apostate (1926), ‘there is little difficulty in recognising Reid’s homosexuality…’. The theme of most of his fiction is boyhood, youth and the supernatural. The letters appear to concern a complicated adult relationship. Only one of Reid’s post-1926 novels deals with adult relationships, Brian Westby (1934). The Collection catalogue states that ‘Reid himself noted his limitations of scope by pondering that some ‘arrested development’ prevented him from fully realising a world of only adult relationships. There is, however, material that is similar: Notes for ‘Nina Westby’ and Notes for Chapters of Nina Westby. According to Taylor (1980: 97—98), these are work towards an unfinished novel based on the real life case of Madeleine Smith, tried in Glasgow in 1857 for the murder of her former lover. Evidence against her is a series of love letters which she wrote to the lover. Perhaps the ‘Sadie’ letters are work towards Nina Westby? Reid sets Nina Westby in 1857, the year of the Smith case, not 1926, the date on the letters. Further, he left a detailed dramatis personae in which there are no characters called Sadie. There are differences in format, too. The disputed letters are written as letters, on note paper; the Nina Westby material is on scrap paper or in notebooks. I examined other letters to Reid. There are more than 5,000 letters in the Collection from around 135 different people, most of them men. None of the letters from these correspondents contain content similar to that in the ‘Sadie’ letters. Neither do other letters written by Reid. The data within the ‘Sadie’ letters comprises dates, names, locations and a telephone number. The days and the dates correspond with the 1926 calendar: 9th May was a Sunday and 18th May was a Tuesday. The surnames which appear—Bowden, Finley, Hull, Andrews—appear in the 1926 Belfast Street Directory, with some frequency. The place name, Antrim Road, is not a fiction, it is a main road in Belfast. The telephone number, 1246, existed as a Belfast number but I was unable to find a subscriber list more up to date than 1913 when the telephone number was registered to a J.C.C. Payne in premises referred to as a medical hall (a chemist’s shop). The address given for it did not exist in 1926; the number may have been transferred to another address, if not another subscriber. To sum up, on the basis of the external evidence no firm conclusions can be reached. The letters may or may not be creative-developmental work by Reid. Given his preferred subject matter, it is quite likely they are not. The letters may be genuine personal letters. The format and content do not rule this out. It is unlikely, however, that they were written to Reid. There remains the possibility that they are letters written by Reid as ‘Sadie’, for whatever purpose. Finally, they may be letters written by someone called ‘Sadie’ which Reid somehow acquired. Thus further work in the form of a forensic linguistic analysis of the letters, from hereon, the Questioned Documents (QDs), was required. 2. Research question The aim of the forensic linguistic analysis was to determine the probability that Reid is or is not the author of the letters. In conducting the analysis I was setting out to test the hypothesis that Reid is the author of the letters as creative-developmental work. Because of the double-barrelled nature of this hypothesis there are two alternative hypotheses: -171-

1) that Reid is the author and the letters are not creative-developmental work (but are genuine letters) 2) that Reid is not the author. 3. Research design and method It is the theoretical concept of idiolect that underpins the use of linguistics in authorship analysis and the theoretical position from which the researcher approaches the problem of authorship (Coulthard and Johnson 2007: 161). McMenamin (2002: 52—53) defines idiolect as ‘the individual’s unconscious and unique combination of linguistic knowledge, cognitive associations, and extra-linguistic influences.’ That every writer has a unique style is the ‘primary hypothesis’ of attribution studies (Rudman 2000, quoted in Love, 2002: 12). In terms of the logic or structure of the inquiry I followed Wachal (1966: 4 cited in McMenamin 2002: 117—8). According to Wachal, there are three models of authorship analysis: population; consistency; and resemblance. The population model is used if nonlinguistic evidence does not allow a reduction in the number of potential authors to one or a small number—thus there is a ‘population’ of possible authors. This is not the case here where the non-linguistic evidence places the focus on Reid as the suspect author. The consistency model is appropriate when the task is to determine whether two (or more) documents were written by the same author. This is useful here as a first stage since there are two letters in the Questioned Document set. The consistency model is applied to determine the likelihood that both QD1 (the letter dated ‘May 9th’) and QD2 (the letter dated ‘18 May’) were written by the same author. Wachal’s resemblance model is applied as a second stage, comparing the QDs to the known writing. This is appropriate since the non-linguistic evidence, the context in which the QDs were found and the other external evidence, reduces the number of potential authors, in this case to one. As noted above, the QDs could belong to one of two ‘populations of writing’, creativedevelopmental work or genuine personal letters. This has implications for the research design because, in this instance, the QDs can be placed against a substantial body of writing in both ‘populations of writing’ by the suspect author. Thus a key methodological decision is the choice of this known writing against which to compare the QDs. It is important in terms of the validity of the findings to compare like with like. McMenamin (2002: 12) suggests that the known writing should be from the ‘same or similar contexts of writing’, which he defines as ‘purpose, intended reader, topic, medium…, instrument…, time, place, etc.’ Since it is possible that the QDs are genuine personal letters or creative-developmental work the most appropriate sampling solution here is to choose material from these two ‘populations of writing’: to draw a sample from the Collection of Reid’s ‘naturally written’ personal letters to an individual, handwritten on writing paper and written or dated around 1926; and to draw a sample from the Collection of Reid’s creative-developmental work dated around 1926. While this is feasible, use of the Collection as a sample source has two limitations on McMenamin’s ‘same or similar contexts of writing’ criteria. First, it contains no correspondence that is the same as or similar in purpose, intended reader and topic to the QDs: there are no ‘love letters’ from Reid to any of his correspondents (a finding in itself, highlighting the anomalous nature of the QDs in the Collection). Secondly, around 1926 there are only two sets of letters that are verifiably by Reid, those to Andrew Rutherford (MS44/1/115A) and those to Theodore Bartholomew (MS44/1/6B). Despite the limitations, these are the most suitable contenders as they are personal letters verifiably from Reid, written at around the same date as that on the QDs. So, which to choose? The letters to Bartholomew were ruled out because they are mostly about day-to-day life, books and writers, ideas and criticism; there is little in them of a personal nature beyond the superficial. This left the letters to Andrew Rutherford which date from 1899 to -172-

1920. While they cover some of the same areas as those to Bartholomew, a few of them are of a more intimate nature. For example, Reid signs a letter dated 17 August 1904: ‘I am as always your loving friend’. I decided therefore to choose one of the letters to Andrew Rutherford, despite it failing on the time criterion—It is dated ‘28 Nov.’ with ‘1902’ added in pencil. It does meet the other criteria to some extent: it is a personal letter on a topic other than work and writing; it is handwritten in ink; and it is verifiably from Reid. It is 390 words long. As a sample from the population of Reid’s creative-developmental work, I went to the notes for the unfinished novel, Nina Westby (MS44/3/12 and MS44/3/13). This was an appropriate source given the similarity in content with the QDs: the plot involves love letters that reveal secret liaisons and an illicit affair. It also meets the time criterion. It seems likely that Reid was working on Nina Westby some time after finishing the 1922 novel, Pender Among the Residents, and before embarking on Apostate, which was published in 1926 (Taylor 1980: 97—98). The sample chosen came from the notes and chapter outlines a pasted-in typescript. It is 81 words long, labelled in pen, ‘XXViii (cont’d) Last Letter’. It is a reasonable match in terms of format and content for QD2: it is a love letter to a ‘Beloved’ about a meeting. In terms of variables or style markers, I examined a range, informed by McMenamin (2002), Gibbons (2003) and Coulthard and Johnson (2007). These included format, spelling, punctuation, syntax and handwriting. All have been used in other cases (see Coulthard and Johnson, 2007: 164—173; McMenamin, 2002: 216—231). However, none on its own is likely to provide enough information to reach a conclusion: as McMenamin notes, ‘unique markers are extremely rare’ (McMenamin 2002: 172). Thus I sought an ‘aggregate of markers’ (McMenamin 2002: 172). The length of both the QDs and the known writing samples does raise a question in relation to the validity of the research. QD1 is 131 words long and QD2 is 192 words; the known sample of creative-developmental work is 81 words and the known letter sample is 390 words long. Is it possible to derive sufficient information with which to reach a conclusion from such small amounts of writing? This is a situation encountered in many forensic linguistic contexts. Coulthard (2004: 432) notes that in his experience QDs and samples of known writing tend to be under 200 words long and many contain fewer than 100 words. Here this limitation is addressed to some extent by the research approach adopted: the review of external evidence conducted prior to the forensic linguistic analysis enabling a narrowing of the field of potential authors; the comparison between the two documents in the QD set; the selection of known material from two populations of writing; and the use of an ‘aggregate of markers’ (McMenamin 2002:172). Should the findings from the comparison of the QDs and the known writing from either or both populations of writing prove inconclusive a solution, as recommended in the SWGDOC (2000), is to select further samples of material, even to the extent of taking a corpus approach (Turrell, 2010: 239—240). 4. Findings 4.1. QD1 v QD2 In comparing the two QDs I found a substantial number of similarities: punctuation (full stops where question marks are expected, full stops that look like dashes or commas, capitalisations); sentence structure (all simple and compound sentences and some moodless sentences). There were some differences in format, the style of the date and the title of the addressee. I concluded that it is highly probable that the same person wrote both.

-173-

4.2. QDs v Reid’s known creative-developmental work There are few similarities between the sample of known work and the QDs bar an overlap in vocabulary (‘beloved’, ‘ever’, love). The purpose of QD2 is to request a meeting with the ‘Only Beloved’: ‘Will you let me spend another afternoon or evg in your company. Will you grant me this request. ... Will you grant me this wish.’). The purpose of the known work is similar but is executed with the use of the imperative: ‘Come to me sweet one. ... Do come sweet love, my own dear love of a sweetheart. Come beloved ... Come and we shall be happy ...’ ). (In terms of its formal, archaic constructions and lack of contractions, the known creative-developmental work shares features of the known personal writing.) 4.3. QDs v known letter by Reid There are some similarities here in the style of punctuation and in the use of the conjunction ‘but’: in QD1, ‘I am filled with nothing but the thoughts of you…’ and in the known sample, ‘I can say nothing further, but only leave the rest to your generosity.’ The main dissimilarities between the QDs and the known writing are in syntax: sentence structure, style of negation and pronoun use. All of the sentences in the QDs are simple and compound sentences; there are no complex sentences. In the known writing there are simple, compound and complex sentences. The style of negation in the known writing differs from that in the QDs. There are no contractions of the negative (‘…if I cannot do anything…’; ‘…I could not make…’) whereas there is one in QD2, ‘Dont’ (with no apostrophe). There is use of a more archaic formal style in the known writing: ‘I know not what’ and ‘Cannot you understand…’ , which is not in evidence in the QDs. Further, in QD1 there are examples of a particular style of negation: the main verb appears before the negative (adverbial): ‘I will hold out no longer…’ and ‘She shall play her game no longer.’ The ‘no’ in this construction, modifying the ‘longer’, may indicate an Ulster Scots influence. Indeed there are other usages and constructions which suggest Scots influence and this sociolinguistic aspect requires further investigation and analysis. There are features in the QDs that are not seen in the known sample. In QD1 there is a lack of agreement between subject and verb (‘…only you has…’). In QD2, the relative pronoun ‘that’ (‘…always a gentleman that speaks for her…’) is used in place of ‘who’. In QD2 the possessive pronoun ‘their’ is used twice instead of ‘there’ (in both cases in its existential use). 5.

Conclusion

Based on the results of the comparison of the creative-developmental work with the QDs I would suggest that it is probable that Reid is not the author. There are very few similarities. However, the comparison here is a very limited one. The quantity of known writing is small. As to the comparison of the known personal writing with the QDs, there are a great many differences, in particular in syntax, and few similarities. There is a much greater degree of variation between the QDs and the known writing than there is between the two QDs. On the basis of these findings I suggest that it is highly probable that Reid is not the author of the QDs. Nevertheless, this conclusion too is partial and contingent. Again there are limitations, chiefly in terms of absent characters and in the low quantity of writing. 6. Further work The work presented here is in effect a preliminary study. As it stands it has provided an answer to the research question within a very broad range of probability. In other words, there remains some uncertainty about the authorship attribution. To address this, further work must involve a -174-

comparison of the QDs with a more substantial body of known writing, particularly with regard to the creative-developmental work. This would involve creating two corpora: a corpus of Reid’s personal letters; and a corpus of his creative-developmental work. (It might also be useful to create a corpus of material from the wider population c.1926 against which, for example, to test the occurrence of some of the variables and distinctive features identified within the QDs and within the known writing.) The research has also raised issues that have not (yet) been addressed. For example, is this a case of deception? Reid was a highly skilled writer, and the writer is a professional deceiver. Did Reid write these letters to deceive? Further work at the discourse level and work to examine the QDs for possible markers of deception is planned and should help here, as should the corpus-based work. Of course there remains the possibility that the QDs were acquired rather than written by Reid. There are many ‘foreign’ elements (that is, non-Reid authored/generated material) in the Collection and the ‘Sadie’ letters may fall into this category, perhaps procured by Reid to help him write about something with which he was unfamiliar in his own life, an adult romantic relationship. References Belfast Street Directory. (1926) Belfast: Thom, Alexander & Co. Belfast

Telephone Directory. (1913) Accessed 18 November 2008.

http://www.lennonwylie.co.uk/1913PhoneList5.htm

Brown, K. (2008) MS44 Forrest Reid Collection. Belfast: Queen’s University Belfast. Bryan, M. (1976) Forrest Reid. Boston: Twayne. Burlingham, R. (1953) Forrest Reid: A Portrait and a Study. London: Faber & Faber. Coulthard, M. (2004) Author identification, idiolect and linguistic uniqueness. Applied Linguistics. 25: 431—47. Coulthard, M. and Johnson, A. (2007) An Introduction to Forensic Linguistics: Language in Evidence. London: Routledge. Cruise, C. (1997) Error and eros: The fiction of Forrest Reid as a defence of homosexuality. In E. Walshe (ed.) Sex, Nation and Dissent in Irish Writing. Cork: Cork University Press. Fitz-simon, C. (1998) Forrest Reid remembered. In P. Goldman and B. Taylor (eds) Retrospective Adventures: Forrest Reid: Author and Collector 5—6. Aldershot: Scolar Press. Gibbons, J. (2003) Forensic Linguistics: An Introduction to Language in the Justice System. London: Wiley Blackwell. Goldman, P. and Taylor, B. (eds) (1998) Retrospective Adventures: Forrest Reid: Author and Collector. Aldershot: Scolar Press. Hughes, E. (1992) Ulster of the senses. In Fortnight. 11. -175-

Love, H. (2002) Attributing Authorship: An Introduction. Cambridge: Cambridge University Press. Love Letters from ‘Sadie’. MS44/13/2 Queen’s University Library Special Collections (2008). Forrest Reid Collection. Queen’s University Belfast. McMenamin, G. R. (2002) Forensic Stylistics: Advances in Forensic Stylistics. Boca Raton, FLA: CRC Press. Reid, F. Nina Westby. MS44/3/12 and MS44/3/13 Queen’s University Library Special Collections (2008). Forrest Reid Collection. Queen’s University Belfast. Reid, F. Nina Westby. MS44/13/2 Queen’s University Library Special Collections (2008). Forrest Reid Collection. Queen’s University Belfast. Reid, F. (1922) Pender Among the Residents. London: Collins. Reid, F. (1926) Apostate. London: Constable. Reid, F. (1934) Brian Westby. London: Faber & Faber. Reid, F. (1940) Private Road. London: Faber & Faber. Reid, F. (1944) Young Tom; or Very Mixed Company. London: Faber & Faber. Reid, F. To Theodore Barthlomew. Letters. MS44/1/6B Queen’s University Library Special Collections (2008) Forrest Reid Collection. Queen’s University Belfast. Reid, F. To Andrew Rutherford. Letters. MS44/1/115A Queen’s University Library Special Collections (2008) Forrest Reid Collection. Queen’s University Belfast. Rudman, R. (2000) Non-traditional authorship attribution studies: ignis fatuus or Rosetta Stone? BSANZ Bulletin 24: 163—176. SWGDOC: Scientific Working Group for Forensic Document Examination (2000). The Standard Guide for the Examination of Handwritten Items. Washington, D.C.: Federal Bureau of Investigation, Laboratory Division. Taylor, B. (1990) The Green Avenue: The Life and Writings of Forrest Reid, 1875-1947. Cambridge: Cambridge University Press. Turrell, M. T. (2010) The use of textual, grammatical and sociolinguistic evidence in forensic text comparison. The International Journal of Speech, Language and the Law 17: 211— 250. Wachal, R. S. (1966) Linguistic evidence, statistical inference, and disputed authorship. Dissertation. Wisconsin: University of Wisconsin.

-176-

Investigating formulaic language as a marker of authorship Samuel Tomblin Centre for Forensic Linguistics, Aston University, UK [email protected] Abstract This research unites the psycholinguistic theory of formulaic language, that is, prefabricated sequences of words believed to be stored as holistic units, and the practice of forensic authorship attribution with a view to developing a new marker of authorship. Since formulaic sequences are holistically processed, it stands to reason that they are likely to elude a writer’s attempts to disguise their style. It follows that research into formulaic language usage may therefore assist in the development of new tools for authorship attribution. In order to test this hypothesis, a reference list containing 13,412 examples of formulaic sequences was compiled from multiple online sources (e.g., lists of clichés, idioms) which was then used to identify formulaic language in a 20 author corpus containing 100 personal narratives. A series of statistical tests were used to determine whether the proportion of formulaic language compared to novel language was sufficient to differentiate authors and to attribute a Questioned Text to its author. The results are discussed with reference to the reliability and validity of the method. Keywords:

FORMULAIC LANGUAGE; FORMULAIC SEQUENCES; AUTHORSHIP ANALYSIS; AUTHORSHIP ATTRIBUTION; MARKER OF AUTHORSHIP

Introduction: formulaic sequences Language enables us to express our ideas in many different ways and the opportunity for novelty is vast: There is no doubt that essentially all speakers of a language are free to produce sentences they have never heard or produced before. Very few people, on seeing two blue rabbits in a fish-bowl, are going to be poorly equipped, linguistically, to express their experience, even though the sentence they would need to create for the task would undoubtedly be completely novel to them (Fillmore, 1979: 95). However, whilst the potential for novel utterances is limitless, speakers appear ‘to renounce the great freedom that the language offers’. Nattinger and DeCarrico (1992) suggest that ‘just as we are creatures of habit in other aspects of our behaviour, so apparently are we in the ways we come to use language’ (p. 1). Evidence from psycholinguistics (e.g., Wray, 2002), sociolinguistics (e.g., Coulmas, 1979), corpus linguistics (e.g., Moon, 1997; 1998a; 1998b) and both L1 and L2 language acquisition (e.g., Pawley and Syder, 1983; Peters, 1983; Peters, 2009; Peters, 1977; Vihman, 1982) shows that when communicating, we rely on patterns in language and have ‘preferred formulations’ for expressing ideas (Wray, 2006: 591). This results from the fact that much of our everyday activity is routine: ‘As similar speech situations recur, speakers make use of similar and sometimes identical expressions, which have proved to be functionally appropriate’ (Coulmas, 1981: 2). In fact, mastering the balance between novel language and routine language is a key characteristic for sounding like a competent, fluent and native speaker (Ellis, 1996; Fillmore, 1979; Coulmas, 1981; Pawley and Syder, 1983; Howarth, 1998). -177-

Such routine language can in a global sense be termed formulaic which Wray (2002) defines as ‘[w]ords and word strings which appear to be processed without recourse to their lowest level of composition’ (p. 4). Wray provides the example of the breakfast cereal ‘Rice Krispies’. During an advertising campaign for television, people were asked what they thought the product was made of and were surprised to learn that it was rice. According to Wray, people had ‘internalized this household brand name without ever analysing it into its component parts’ (2002: 3). In this way, ‘Rice Krispies’ seems to be stored and produced as a single lexical item, rather than two separate items. The fact that multi-word sequences may be stored as single lexical items is an important feature of formulaic language ( Bannard and Lieven, 2009; Ellis, 1996; Erman and Warren, 2000; Pawley and Syder, 1983; Wray, 2000; 2002; 2008). Formulaic language is an umbrella term and a survey of the literature soon reveals that many terms exist to describe different characteristics of formulaic language. These include Collocations ( Herbst, 1996; Gledhill, 2000; Stubbs, 1995), Idioms ( Grant and Bauer, 2004; Simpson and Mendis, 2003), Fixed Expressions including Idioms (Moon, 1998a) Formulaic Sequences (Wray, 2002; Schmitt and Carter, 2004), Multi-word Items (Moon, 1997), Phrasal Lexemes (Moon, 1998b), Recurrent phrases (Stubbs and Barth, 2003) and Situation Bound Utterances (Kecskés, 2000), to name just a few. In fact, Wray (2002: 9) found 57 different terms each describing what can be characterised as formulaic. Though related, these terms denote slightly different characteristics associated with formulaic language. Some definitions emphasise the importance of context and register ( Cortes, 2004; Kecskés, 2000) whilst others focus on the amount of distance between words (Hoover, 2003) or whether sequences of words should be contiguous ( Hoover, 2002; Stubbs, 2002; Stubbs and Barth, 2003). The definition adopted in this research is that of the formulaic sequence: [A] sequence, continuous or discontinuous, of words or other elements, which is, or appears to be, prefabricated: that is, stored and retrieved whole from memory at the time of use, rather than being subject to generation or analysis by the language grammar (Wray, 2002: 9). Wray’s definition of the formulaic sequence is intended to be as inclusive as possible so that it can be used as a coverall term for any part of language that has been considered formulaic by previous definitions (p. 9). Estimates vary regarding how much of everyday language use is formulaic. Erman and Warren (2000) claim that 55% of spoken and written language may be formulaic whilst Chenoweth found 77% of written answers to essay style exam questions contained formulaic expressions regardless of length (1995: 292). Bannard and Lieven (2009) found that between 86% and 97% of utterances spoken by toddlers were derived from recurring strings and Pawley and Syder (1983) argue that ‘the largest part of the English speaker’s lexicon consists of complex lexical items including several hundred thousand lexicalized sentence stems’ (p. 215) which they define as ‘a unit of clause length or longer whose grammatical form and lexical context is wholly or largely fixed’ (p. 191). A lack of consensus over the exact proportion of formulaic language compared to novel language in everyday usage results from differences in definitions, methods of identification and contexts of use. However the overriding claim is that formulaic language is ubiquitous and prevalent in language (Wray, 2002).

-178-

A function of formulaic sequences: reducing cognitive burden The exact size of the mental lexicon is not known, although de Bot (1992) estimates there to be about 30,000 words in the active lexicon. De Bot calculates that with an average rate of speech of 150 words per minute, peaking at approximately 300 words per minute, the average speaker has 200—400 milliseconds to select the words they wish to use. Expressed another way: 2 to 5 times a second we have to make the right choice from those 30,000 words. And usually we are successful; it is estimated that the probability of making the wrong choice is one in a thousand (p. 11) Producing language is clearly a cumbersome task, albeit one which humans manage with relative ease. However, if sequences of words are stored as single lexical items (as the theory of formulaic language suggests), then accessing that sequence of words from the lexicon, rather than constructing a novel sequence from individual words, plausibly reduces the amount of cognitive processing required: Formulae make the business of speaking (and that of hearing) easier. I assume that when a speaker uses a formula he or she needs only to retrieve it from the dictionary instead of building it up from its constituent parts. In other words, such expressions likely exist as whole or part utterances within the speaker’s dictionary and need not be built up from scratch on every new occasion (Kuiper, 1996: 3). Therefore, there is consensus that reducing cognitive burden is a function of formulaic language ( Kuiper, 2000; Peters, 1983; Wray, 2002; Wray and Perkins, 2000;). By decreasing cognitive load, fluency can be increased (Fillmore, 1979; Kuiper, 1996). Since formulaic sequences are stored in this pre-packaged, holistic form they are likely to escape conscious regulation by authors—in other words, authors will produce sequences of words without necessarily thinking about each individual word. It naturally follows that if authors are unaware that they are using particular sequences of words it will be much harder for them to disguise their style. This point is made by Lancashire (1998): Word, phrase, and collocation frequencies … can be signatures of authorship because of the way the writer’s brain stores and creates speech. Even the author cannot imitate these features, simply because they are normally beyond recognition, unless the author has the same tools and expertise as stylometrists undertaking attribution research. Reliable markers arise from the unique, hidden clusters within the author’s long-term associative memory. (p. 299) It stands to reason that if evidence can be found of formulaic sequence usage varying between authors it should make an excellent candidate for a new marker of authorship. Aims and hypotheses The aim of this paper is to determine whether the overall proportion of formulaic sequences in texts is sufficient to differentiate authors. By concentrating on the proportion of text that is formulaic, it will be possible to make claims about whether the language used by one author is more or less formulaic than that of another. If this is the case, the consistency in levels of formulaic sequences across a series of authors’ texts can be investigated. Finally, it will be possible to determine whether a given text can be successfully attributed to its author, as -179-

would be necessary in a case of forensic authorship analysis. To carry out this investigation, a series of hypotheses must firstly be proposed. Individuals are socialised differently and this affects their repertoires of formulaic sequences (Wray, 2002). Therefore, since each of the authors that contributed data to this research will have a different set of life experiences, they should have a different range of formulaic sequences to draw upon; some with larger repertoires and some with smaller. Taking into account each author’s potential formulaic repertoire and the range of cognitive demands placed on them in producing language, it is hypothesized that authors will use differing proportions of formulaic sequences. Secondly, if the first hypothesis is correct, there should be a significant difference in the proportion of formulaic sequences compared to novel language used by an author and, based on this variable it should be possible to differentiate authors. This is important for the forensic context in demonstrating that the variation between authors is significant. Thirdly, if support is found for the first and second hypotheses, using a corpus of texts which have been carefully controlled for genre and length, which have also been composed in the same time period and on similar topics, it should be possible to attribute a text to its author. In summary, the following hypotheses will be tested: i) ii) iii)

Variation in proportion of formulaic sequences will be greater between authors than within authors; Authors will be potentially differentiable from each other based on the proportion of formulaic sequences in their texts; A randomly selected Questioned Text will be correctly attributed to its author based on the proportion of formulaic sequences in the Questioned Text and in the author’s other texts.

Method The task of identifying formulaic sequences in texts is not an easy one; so difficult in fact that Wray (2008) comments ‘[i]dentifying formulaic sequences in normal language can be rather like trying to find black cats in a dark room: you know they’re there but you just can’t pick them out from everything else’ (p. 101). Formulaic sequences can be identified in several ways, depending on whether the language is spoken or written. In spoken language, formulaicity can be identified through phonological analysis which focuses on stress, articulation, fluency and pausing (e.g., Pawley and Syder, 1983; Peters, 1983; Wray, 2002 for a comprehensive review). Whether formulaic input or output is under investigation is also relevant. Eye-gaze studies, for example, can be used to monitor how participants read formulaic sequences if input is the focus (Underwood, Schmitt and Galpin, 2004). In written language, reference lists such as dictionaries and text books provide a source of established examples of formulaic sequences (Wray, 2008: 109). It is possible, using such sources, to match a given dataset against a reference list and identify those examples which occur. Wray (2008) cautions, however, that if a researcher wishes to use a reference list, it is important for them to think about why that list was produced and what decisions were made about what to include and exclude: An important question for any researcher to consider before using existing lists to identify formulaic sequences is whether the list has gained authority simply by virtue of being published (Wray 2008: 109).

-180-

Whilst a reference list of formulaic sequences may be an excellent resource in practical terms (e.g., analysis on large sets of data can be fast and reliable), caution needs to be expressed over which items are included in the list since without clear justification, there is the possibility for the list to be nothing more than the intuitions of one individual. For the present research, a compromise is proposed: using the internet to build a reference list. By drawing on a multitude of different sources compiled by many members of various speech communities should ensure that the list is as representative of formulaic sequences as possible. Compiling a reference list Terms commonly accepted as examples of formulaic sequences were entered into the online search engine, Google. These included, for example, list of proverbs, list of clichés, list of common phrases, list of similes, and list of popular sayings. The search term list of regular expressions could not be used since regular expression is a specific technical term from the field of computer science and so returned too many irrelevant results. Similarly, the search string list of formulaic language did not provide any useful lists (mainly links to online books and articles related to formulaic language) since no such list has been widely publicised. For each search string, all of the links from the first five pages were explored. There did not appear to be any benefit in exploring beyond the fifth page since these typically included irrelevant links, or links that had already been explored. Every time a link led to a website which contained examples of formulaic sequences, those examples were entered into the database regardless of whether or not they were intuitively pleasing as examples of formulaic sequences. This process was repeated until no new websites were identified. It became clear that several of the websites were sharing examples of formulaic sequences between themselves and so the decision to discontinue adding formulaic sequences was made when it was evident that relatively few new examples were actually being added to the list. The list at this stage contained 17,973 entries. It is difficult to account for the contents of the list in terms of how each individual example of a formulaic sequence can be classified (e.g., idiom, collocation, metaphor etc,.) since formulaic sequences can often be classified into several categories (e.g., Moon, 1998a). However, based on how the websites self-identified themselves, the list appears to be composed of the following proportions: Table 1: Proportion of different types of formulaic language included in the reference list Type of Formulaic Language Clichés Idioms Everyday Expressions and Sayings Proverbs Similes Other (including prepositional phrases, collocations, Latin phrases and phrasal verbs) Totals

Number of Entries 5131 3772 3497 2539 1992 1042

Percentage of Entries 28.6% 21% 19.5% 14.1% 11.1% 5.8 %

17,973

100%

Clichés and idioms account for over half of the entire list of formulaic sequences. The category ‘Everyday Expressions and Sayings’ highlights the problem of relying on self-181-

reports for categorisation purposes: the dividing line between a cliché, idiom and everyday saying is in no way clear cut. Pronouns in the reference list were replaced with an asterisk. The software used to identify matches in the data was capable of cross-referencing to a separate list of pronouns and the asterisk indicated the place where any item from the pronoun list was permissible. The pronoun list contained 86 entries including personal pronouns (e.g., me, you, her, it), possessive pronouns (e.g., mine, yours, hers, its) and possessive determiners (e.g., my, your, her). Through this process, by changing the entry his bark is bigger than his bite to * bark is bigger than * bite enabled matches in the data including her bark is bigger than her bite, its bark is bigger than its bite, my bark is bigger than my bite, your bark is bigger than your bite etc,. A problem with this substitution approach is that there is potential for a nonsense string to be identified e.g., her bark is bigger than his bite, your bark is bigger than its bite etc. (although, of course, some of these might not be nonsense and may be deliberate playing with words). However, since it is unlikely that an author would produce these strings under normal circumstances, the advantages of allowing substitution outweigh the disadvantages of having only fully fixed forms in the list. The only pronouns that remained fixed in the list were those where substitution would affect the meaning e.g., get thee behind me Satan, love that dare not speak its name, one small step for man, cry me a river etc. Many of the entries were obtained from American websites. Since the data to be analysed were produced by native English speakers living in England, UK spelling variants were added to the list alongside the original American spellings. Examples include good fences make good neighbours, horse of a different colour, in honour of, and in self-defence. Finally, there were many duplicates in the list, as noted above, which were removed. The final reference list contained 13,412 entries. Not every entry in the list will be acceptable to everyone as an example of a formulaic sequence. Some people will find some entries more problematic and less prototypical than others (e.g., Jiminy Christmas, date rape). The aim of the list is not really to reach universal agreement about what constitutes a formulaic sequence; rather, the aim is to collate as many potentially formulaic sequences as possible in order to investigate whether evidence can be found that some individual authors use formulaic sequences more than others. Just as the list cannot claim to be representative of formulaic sequences for each individual, questions must also be asked about its authority. That is to say that the entries of formulaic sequences have not been verified by independent means, other than by their inclusion on public websites as opposed to being included, for example, on the basis of corpus frequency counts. The result is that a broader, more inclusive list has been created. However, the trade off has been a lack of authority in as much as entries are those that other people have decided are special in some way (be it as a cliché, idiom, common expression or collocation etc,.) which in turn can be considered to be ‘formulaic’ rather than being independently identified in corpora. Whilst the authority of the list may be called into question, the counter argument is that it is in fact representative of the language community—that is, people identified and recognised these examples as holding special status. Therefore, whilst the data collection method differs significantly, the end product equates to asking members of the same speech community to identify formulaic language in texts (e.g., Foster, 2001; Van Lancker-Sidtis and Rallon, 2004) and therefore a level of resilience and authority can be claimed through consensus. In conclusion, there are limitations to the list, both in terms of what it contains and how well it can match examples of formulaic sequences in real text. However, it does hold certain advantages which are particularly favourable for the forensic context. By using an automated approach, large volumes of data can be analysed almost instantaneously. It offers reliability; formulaic sequences included in the list will be matched in any data on any occasion. However, the list cannot claim to identify every single instance of formulaic -182-

sequences in text, nor will it identify variants of items contained in the list (with the exception of pronoun substitution). It cannot even guarantee that every instance it identifies will be formulaic. However, the list is large and varied so the crucial point is that it contains items which have the potential to be formulaic. It is this potential that makes the list a satisfactory initial exploration into the relationship between formulaic language and authorship. With a full understanding of the benefits and limitations of the list, it is now possible to apply it to the authorship data in order to begin our investigation into whether formulaic language has potential as a marker of authorship. Data The data comprise 100 texts written by 20 authors, each author producing five texts. Authors were provided over a five day period with a daily structured writing task. Authors were sent two essay-style questions each morning and were required to answer whichever one they felt most comfortable writing about. Open-ended questions which elicited personal narratives were asked. By asking emotionally-charged questions, it is hoped that the likelihood of participants focussing on their language use was reduced (Labov, 1970; Labov, 1972; Labov and Waletsky, 1997). The 100 texts contained 65,113 words with each author producing an average of 3,325 words across their five texts. The average text length was 651 words with the shortest being 485 words and the longest being 822 words. The software compared the data with the reference list and highlighted all instances of exact matches. Results A total of 604 formulaic sequence tokens were identified in the data, of which there were 300 types. Table 2 shows the ten most frequently occurring formulaic sequences whilst Table 3 shows a selection of ten formulaic sequences that were used only once across the whole data set. Table 2: Most frequently occurring formulaic sequences across the data Formulaic Sequence In the end At least Go back At the end In front of In fact On the phone At home At the same time As if

Frequency of Occurrence Across All Data 20 17 14 12 12 11 11 9 9 8

-183-

Table 3: Least frequently occurring formulaic sequences across the data Formulaic Sequence Under the influence Under the weather Vice versa What on earth What will be will be Wide awake With flying colours With the exception of Worst nightmare X Factor

Frequency of Occurrence Across All Data 1 1 1 1 1 1 1 1 1 1

Table 4 shows how many words each author produced over the total of their fives texts and how many of those words were identified as being formulaic, that is, part of a formulaic sequence (e.g., in fact counts as two formulaic words, in the end counts as three formulaic words whilst at the same time counts as four and so on). The authors are listed in numerical order from the author using the lowest frequency of formulaic sequences to the one using the highest. To facilitate comparison between the authors, a normalised frequency of formulaic language per 100 words is also provided. Table 4: Proportion of formulaic language across the data Author

Total Words

Total Formulaic Words

MELANIE SARAH ROSE JOHN CARLA JUNE MARK DAVID NICOLA GREG ALAN MICHAEL SUE RICK JENNY JUDY HANNAH KEITH ELAINE THOMAS

2879 2957 3820 3119 3217 3151 2844 3058 3021 2980 3916 2516 3716 3583 3518 3427 3559 3067 2941 3824

34 46 66 55 59 59 56 63 62 70 92 61 94 93 103 104 111 95 94 130

Normalised Frequency of Formulaic Language per 100 words 5.97 7.57 8.63 8.81 8.99 9.28 9.92 10.05 10.24 11.62 11.67 12.12 12.63 12.90 14.82 15.26 15.54 15.74 16.03 17.02

Establishing variation between authors Frequency of use of formulaic sequences was tested between 20 authors providing five texts each. A Kruskal-Wallis test showed significantly more variation between authors than within texts by the same author (χ2 = 35, df = 19, p = 0.013)—in other words, the five texts produced by a single author are more alike in the proportion of formulaic sequences contained therein, compared to the texts produced by other authors. The first hypothesis, that variation between authors will be greater than within authors, is therefore supported. -184-

A log linear analysis was carried out to determine any interactions between the factors gender (male/female), age (below 25/above 25) and education (Pre-university/Undergraduate/ Postgraduate). Analysis showed that no significant interactions could be separated out from the saturated model indicating that there were no significant patterns in the proportion of formulaic sequence usage for gender, age or education. With regard to the second hypothesis, that authors will be differentiated from each other based on the normalised frequency of formulaic language usage, it is evident from Table 4 that these authors do in fact use different normalised frequencies of formulaic sequences in their texts. However, the statistical significance of these differences is not clear. In what follows, the second hypothesis is statistically tested. Differentiating authors: a test case As a test case the highest and lowest mean ranked authors were compared (Thomas and Melanie respectively). With just five texts each (equivalent to a total of 3,824 words and 2,879 words) it was possible to differentiate these two authors based on the normalised frequencies of occurrences of formulaic language (Mann-Whitney U = 1, N = 10, p = 0.016). This provides evidence that using the normalised frequency of formulaic language as a marker of authorship works. The question that remains is how well it works for authors whose normalised frequency of formulaic language is more similar. Differentiating authors: a harder case Taking the highest and lowest mean ranked authors somewhat improves the likelihood of reaching significance, since these authors were at the extreme ends of normalised frequency of formulaic language usage. Therefore, to further test the method, the two authors with the closest normalised frequency of formulaic language were compared (Greg and Alan, with 11.62 and 11.67 formulaic words per 100 respectively). With just five texts each (equivalent to a total of 2,980 words and 3,916 words) it was not possible to differentiate these two authors (Mann-Whitney U = 11, N = 10, p = 0.841). This result clearly invites the question of how effective the method is when two authors whose normalised frequency of formulaic language is neither very similar nor very different. Differentiating authors: exploring the limits Two sets of authors were selected to explore the limits of the method. The 5th ranked author, Carla, and the 16th ranked author, Judy, were selected for the analysis (8.99 and 15.26 formulaic words per 100 respectively). With just five texts each (equivalent to a total of 3,217 words and 3,427 words) it was possible to differentiate these two authors (Mann-Whitney U = 23, N = 10, p = 0.032). Secondly, the 7th ranked author, Rick, and the 14th ranked author, Mark, with 12.90 and 9.92 formulaic words per 100 respectively were compared. With five texts each (equivalent to a total of 3,583 words and 2,844 words) it was not possible to differentiate these two authors (Mann-Whitney U = 6, N = 10, p = 0.222). It can be seen that the normalised frequency of formulaic language was too close for Rick and Mark, whereas the texts produced by Carla and Judy enabled differentiation. Taking these results into account, only partial support can be claimed for the second hypothesis, since the method only appears to work when the difference in normalised frequency of formulaic language between the authors is larger (although this is a relative term and future statistical testing would be required in order to accurately establish the boundaries of this distance). Therefore, we can more safely say that based on the twenty authors -185-

investigated in this study, some authors exhibit different proportions of formulaic sequences in their texts from some other authors. The analysis carried out here relies on pairwise distinctions (e.g., Grant, 2010) as opposed to population wide distinctions (e.g., Chaski, 2001). Therefore, the variable, normalised frequency of formulaic language, holds potential to differentiate some pairs of authors but not all pairs of authors. In this regard, it is analogous to using the visual description of height as a variable on which to differentiate people. Some people will be taller, some will be shorter, and some will be the same height and it would not be possible to establish a threshold at which differentiation between people becomes possible. The same is true of using the quantity of formulaic sequences in a text as a marker of authorship. In a closed sample, some authors can be differentiated whilst others cannot. Therefore, it is not possible to claim the method described here as a universal method that will be applicable in all cases. Assessing forensic potential In an attempt to replicate a realistic forensic case, five texts by each of two authors were randomly selected for analysis: Nicola and Greg. The two groups of texts were tested to see if they were normally distributed. Both groups showed no significant difference from normal (Nicola: KSZ = 0.913, N = 5, p = 0.376; Greg: KSZ = 0.445, N = 5, p = 0.989). The second text by Nicola was randomly selected by SPSS to act as the Questioned Text. A two-tailed one-sample t-test showed no significant difference between the normalised frequency of formulaic language in the four texts by Nicola compared to the Questioned Text, also by Nicola (t(3) = 0.601, p = 0.590). As the prediction from the means was that Nicola's scores would be lower than those of Greg, a uni-directional hypothesis was tested. A one-tailed one-sample t-test showed a significantly higher normalised frequency of formulaic language in the five texts by Greg compared to the Questioned Text (t(4) = 2.157, p = 0.0485). In real terms, we can say that there is a 95% chance that Nicola wrote the Questioned Text which is arguably an acceptable level of confidence for forensic linguistics evidence and which we know to be a correct attribution. In terms of the final hypothesis, that a randomly selected Questioned Text will be correctly assigned to its author based on the normalised frequency of formulaic language in that author’s other four texts, the results demonstrate that when a Questioned Text is compared to nine Known Texts produced by a closed set of two authors, it is possible to correctly attribute the Questioned Text to its correct author. The final hypothesis is therefore supported. Discussion These results provide evidence that taking the normalised frequency of formulaic language usage as a marker of authorship does have the potential to differentiate authors and, more importantly, to attribute a Questioned Text correctly to its author. In line with the aims of this research, the focus has only been on the proportion of formulaic language usage compared to novel language, in other words, whether authors or more or less ‘formulaic’ than others. However, it is important to acknowledge that no individual formulaic sequences emerged as being characteristic of authorship, that is, no individual formulaic sequence appears to be related to idiolect. In order to fully contextualise the success and effectiveness of the method, it is necessary to discuss the validity and reliability of the method.

-186-

Is the method valid? To assess the validity of the method, it is necessary to critically examine whether formulaic language has actually been identified through this process. There are two considerations in this regard: i) The entries that were included in the reference list, and ii) The entries that were actually identified in the data. Dealing firstly with the reference list, as has been argued in this paper, it is highly unlikely that everybody will agree that what is included in the reference list is a formulaic sequence and therefore is an example of formulaic language. The key point, as has also been emphasised, is not that any one individual agrees with every item on the list; rather, that each item on the list holds an equal opportunity to be formulaic for any author. Furthermore, the list cannot claim to be exhaustive and there are undoubtedly other entries that could have been included. However, to compensate for these unavoidable shortfalls, the list is as deliberately large and inclusive as possible, covering a multitude of different types of formulaic sequence. As long as researchers accept collocations, idioms, similes, everyday sayings and so on to be formulaic, the list is valid. Next to consider is whether those formulaic sequences identified in the data are valid in terms of being evidence of authorship, or whether they are indicative of something else. The theoretical basis for this paper has been that different authors will have different cognitive abilities which will be evidenced through their reliance on formulaic sequences. The reality is that several other factors may have had an impact on an author’s use of formulaic sequences. Such factors may include how well rehearsed or edited their particular narrative was and whether they were concentrating fully and solely on the task (or whether they were concurrently preparing a meal, chatting on a social networking website, watching television etc.). However, it is hoped that by collecting five texts from each author over a series of five days, such additional cognitive pressures may have been mitigated by texts produced on days when there were perhaps fewer cognitive pressures to give a representative account of each individual author’s average cognitive load when producing language. (Although, clearly, producing a threat letter, suicide note, or ransom demand will carry additional cognitive pressures that go far beyond the scope of this research.) Is the method reliable? In order for the method to be reliable, it would need to be proven that the same examples would be identified each time the analysis is replicated. This is true since the method is automated and so is unaffected by factors which commonly affect reliability (e.g., tiredness of the researcher, unprincipled analysis of large quantities of data, etc,.). However, establishing that the method is reliable each time the analysis is carried out is only useful if the method can be applied to any type of data. The research described in this paper has focussed only on formulaic sequences occurring in a very restricted type of data—short personal narratives. The reliability of the method may be criticised on the basis that the data used are in some way special. Did the questions asked to elicit the narrative data encourage a higher normalised frequency of formulaic language in the responses? To assess this, all of the narrative eliciting questions were matched against the reference list. No incidences of formulaic sequences were identified. It is therefore unlikely that the authors were primed in their use of formulaic sequences and the data can be argued to have occurred naturally. However, a potential criticism may be that the narratives themselves are not representative of normal, everyday language. After all, the narratives were deliberately intended to encapsulate the authors. As entertaining personal narratives, it is conceivable, perhaps even probable, that the authors will have told these narratives in various ways on various occasions, and they -187-

may therefore be rehearsed, revised and may contain hyperbole. As such, it may be hard to argue them to be naturally-occurring (in the same way that traditional oral stories contain higher occurrences of formulaic language to aid memory during public performances, cf. Rubin (1998)). Finally the range of speech communities represented by the list should also be considered. A wide variety of UK and USA variants have been included. In principle, the reference list can therefore be used to identify formulaic language in texts produced by speakers of British or American variants of English. However, it could only be applied to texts which follow the standard conventions of English and may be less applicable to nonstandard varieties of English (such as text message language, computer-mediated communication etc.). It is therefore unreliable as a universally-applicable method for authorship analysis. Conclusions This paper has outlined a method of authorship attribution which takes the normalised frequency of formulaic language compared to novel language as a marker of authorship. Using just five texts totalling approximately 3,000 words from each of 20 authors, it was established that there is more variation between authors than within, that some pairs of authors with different normalised frequencies of formulaic language usage can be statistically differentiated and that when two authors are randomly selected, a Questioned Text can be correctly attributed to its author. The method has also been argued to be valid, although far more testing than is possible in this initial exploration is required in order to demonstrate reliability. Despite the positive conclusions that can be drawn from this pilot investigation method, it is important to stress that although the two authors with the highest and lowest normalised frequencies of formulaic language in their texts could be differentiated, it was not possible to differentiate the two authors with the most similar normalised frequencies of formulaic language. However, there is currently no unified method for authorship analysis, and instead the linguist must select the most appropriate methods from a rich toolkit. With further testing, the method described here could conceivably be added to that toolkit as another variable on which some authors have been demonstrated to vary from others and may add further evidence in some cases of authorship attribution. References Bannard, C. and Lieven, E. (2009) Repetition and reuse in child language learning. In R. Corrigan, E. Moravcsick, H. Ouali and K. Wheatley (eds) Formulaic Language: Acquisition, Loss, Psychological Reality, and Functional Explanations Vol. 2. Amsterdam: John Benjamins Publishing Co., 299—321. Chaski, C. (2001) Empirical evaluations of language-based author identification. Forensic Linguistics: The International Journal of Speech, Language and the Law 8(1): 1—65. Chenoweth, N. A. (1995) Formulaicity in essay exam answers. Language Sciences 17(3): 283—97. Cortes, V. (2004) Lexical bundles in published and student disciplinary writing: examples from history and biology. English for Specific Purposes 23: 397—423.

-188-

Coulmas, F. (1979) On the sociolinguistic relevance of routine formulae. Journal of Pragmatics 3: 239—66. Coulmas, F. (1981) Introduction: conversational routine. In F. Coulmas (ed.) Conversational Routine: Explorations in Standardized Communication Situations and Prepatterned Speech. The Hague, Netherlands: Mouton Publishers, 1—17. de Bot, K. (1992) A bilingual production model: Levelt's 'Speaking' model adapted. Applied Linguistics 13(1): 1—24. Ellis, N. (1996) Sequencing in SLA: phonological memory, chunking, and points of order. Studies in Second Language Acquisition 18: 91—126. Erman, B. and Warren, B. (2000) The idiom principle and the open choice principle. Text 20(1): 29—62. Fillmore, C. (1979) On fluency. In C. Fillmore, D. Kempler and W.S.-Y. Wang (eds) Individual Differences in Language Ability and Language Behavior. London: Academic Press, 85—101. Foster, P. (2001) Rules and routines: a consideration of their role in the task-based production of native and non-native speakers. In M. Bygate, P. Skehan and M. Swain (eds) Researching Pedagogic Tasks: Second Language Learning, Teaching and Testing. London: Longman, 75—94. Gledhill, C. (2000) The discourse function of collocation in research article introductions. English for Specific Purposes 19: 115—135. Grant, L. and Bauer, L. (2004) Criteria for re-defining idioms: are we barking up the wrong tree? Applied Linguistics 25(1): 38—61. Grant, T. (2010) Text messaging forensics: txt 4n6: idiolect free authorship analysis? In M. Coutlhard and A. Johnson (eds) The Routledge Handbook of Forensic Linguistics, Abingdon, Oxford: Routledge, 508—522. Herbst, T. (1996) What are collocations: sandy beaches or false teeth? English Studies 4: 379—393. Hoover, D. L. (2002) Frequent word sequences and statistical stylistics. Literary and Linguistic Computing 17(2): 157—80. Hoover, D. L. (2003) Frequent collocations and authorial style. Literary and Linguistic Computing 18(3): 261—86. Howarth, P. (1998) Phraseology and second language proficiency. Applied Linguistics 19(1): 24—44. Kecskés, I. (2000) A cognitive-pragmatic approach to situation-bound utterances. Journal of Pragmatics 32: 605—625.

-189-

Kuiper, K. (1996) Smooth Talkers: The Linguistic Performance of Auctioneers and Sportscasters. New Jersey: Lawrence Erlbaum. Kuiper, K. (2000) On the linguistic properties of formulaic speech. Oral Tradition 15(2): 279—305. Labov, W. (1970) The study of language in its social context. In J.B. Pride and J. Holmes (eds) Sociolinguistics: Selected Readings. Harmondsworth: Penguin, 180—202. Labov, W. (1972) Language in the inner city: studies in the Black English vernacular. Oxford: Basil Blackwell. Labov, W. and Waletsky, J. (1997) Narrative Analysis: oral versions of personal experience. Journal of Narrative and Life History 7(1—4): 3—38. Lancashire, I. (1998) Paradigms of authorship. Shakespeare Studies 26: 296—301. Moon, R. (1997) Vocabulary connections: multi-word items in English. In N. Schmitt and M. McCarthy (eds) Vocabulary: Description, Acquisition and Pedagogy. Cambridge: Cambridge University Press, 40—63. Moon, R. (1998a) Fixed Expressions and Idioms in English. Oxford: Clarendon Press. Moon, R. (1998b) Frequencies and forms of phrasal lexemes in English. In A.P. Cowie (ed.) Phraseology: Theory, Analysis and Applications. Oxford: Oxford University Press, 79—100. Nattinger, J. R. and DeCarrico, J. S. (1992) Lexical Phrases and Language Teaching. Oxford: Oxford University Press. Pawley, A. and Syder, F. (1983) Two puzzles for linguistic theory: nativelike selection and nativelike fluency. In J. Richards and R. Schmidt (eds) Language and Communication. New York: Longman, 191—226. Peters, A. (1977) Language learning strategies: does the whole equal the sum of the parts? Language 53(3): 560—73. Peters, A. (1983) The Units of Language Acquisition. Cambridge: Cambridge University Press. Peters, A. (2009) Connecting the dots to unpack the language. In R. Corrigan, E. Moravcsick, H. Ouali and K. Wheatley (eds) Formulaic Language: Acquisition, Loss, Psychological Reality, and Functional Explanations Vol. 2. Amsterdam: John Benjamins Publishing Co., 387—404. Rubin, D. C. (1998) Memory in Oral Traditions: the Cognitive Psychology of Epic, Ballads, and Counting-out Rhymes. Oxford: Oxford University Press.

-190-

Schmitt, N. and Carter, R. (2004) Formulaic sequences in action: an introduction. In N. Schmitt (ed.) Formulaic Sequences: Acquisition, Processing and Use. Amsterdam: John Benjamins Publishing Company, 1—22. Simpson, R. and Mendis, D. (2003) A corpus-based study of idioms in academic speech. TESOL Quarterly 37(3): 419—441. Stubbs, M. (1995) Collocations and semantic profiles: on the cause of the trouble with quantitative studies. Functions of Language 2(1): 23—55. Stubbs, M. (2002) Two quantitative methods of studying phraseology in English. International Journal of Corpus Linguistics, 7(2): 215—44. Stubbs, M. and Barth, I. (2003) Using recurrent phrases as text-type discriminators: a quantitative method and some findings. Functions of Language 10(1): 61—104. Underwood, G., Schmitt, N. and Galpin, A. (2004) The eyes have it: an eye-movement study into the processing of formulaic sequences. In N. Schmitt (ed.) Formulaic Sequences Amsterdam: John Benjamins Publishing Co., 153—172. van Lancker-Sidtis, D. and Rallon, G. (2004) Tracking the incidence of formulaic expressions in everyday speech: methods for classification and verification. Language and Communication 24: 207—240. Vihman, M. (1982) Formulas in first and second language acquisition. In L. Obler and L. Menn (eds) Exceptional Language and Linguistics. London: Academic Press Ltd., 261—284. Wray, A. (2000) Formulaic sequences in second language teaching: principle and practice. Applied Linguistics 21(4): 463—89. Wray, A. (2002) Formulaic Language and the Lexicon. Cambridge: Cambridge University Press. Wray, A. (2006) Formulaic language. In E.K. Brown (ed.) The Encyclopedia of Language and Linguistics. Oxford: Elsevier, 590—7. Wray, A. (2008) Formulaic Language: Pushing the Boundaries. Oxford: Oxford University Press. Wray, A. and Perkins, M. (2000) The functions of formulaic language: An integrated model. Language and Communication 20: 1—28.

-191-

The use of sequences of linguistic categories in forensic written text comparison revisited Núria Bel1, Sheila Queralt Estevez2, Maria S. Spassova3 and M. Teresa Turell4 1 IULA (Universitat Pompeu Fabra), Spain 234 ForensicLab, IULA (Universitat Pompeu Fabra), Spain 3 New Bulgarian University 1 [email protected] 2 [email protected] 3 [email protected] 4 [email protected] Abstract In recent years, the possibility of studying syntax use through computer-aided queries of annotated corpora has led researchers working in the field of forensic written text comparison to explore a new possible marker of authorship, namely, tag sequences as representation of combinations of linguistic categories. A series of studies carried out during the first research stage at ForensicLab using Spanish language data have shown that tag sequences exhibit a significant discriminatory capacity and can be applied to authorship attribution tasks more effectively. In the second research stage reported in this paper, the analysis aims to identify specific traits of each linguistic category implemented in those tags within the exploited tag set which play a major role in the correct classification of texts and those which do not, without losing sight of the fact that either their exclusion or inclusion in tag composition can help to improve this forensic linguistic comparison method. This paper reports on the findings from the statistical testing of several variants of the Institut Universitari de Lingüística Aplicada’s (IULA) tag set system and their evaluation in the context of authorship analysis. For testing purposes, a corpus of two types of written texts (novel fragments and newspaper articles), from six contemporary Spanish speaking novelists, was compiled. Furthermore, a subcorpus was used of texts written by one of the writers included in this study, whose authorship was anonymised. Preliminary studies show that, in both types of written texts, the use of trigrams produces more statistically significant results than the use of bigrams, especially trigrams consisting of prepositional phrases and, to a lesser extent, verbal and compound adjective phrases. Keywords:

MORPHO-SYNTACTICALLY ANNOTATED TAG SEQUENCES; N-GRAMS; DISCRIMINANT FUNCTION ANALYSIS; FORENSIC WRITTEN TEXT COMPARISON; AUTHORSHIP ATTRIBUTION; SPANISH

1. Introduction Modern Forensic Linguistics, particularly in the area of language as evidence, faces the fundamental challenge of undertaking forensic voice and text comparison analyses by using data from real world texts, reference corpora material and real case documents, which could ultimately lead to more reliable and robust reports in forensic speaker identification and authorship attribution contexts. This paper takes the view that forensic voice and written text comparison methodology should be based upon both qualitative approaches, derived from the forensic linguist’s knowledge (auditive methods in forensic phonetics and acoustics; diverse qualitative techniques in written authorship attribution), and semi-automatic and quantitative -192-

approaches to real-world and real-case texts, based on multidimensional voice analysis methods, Linear (or Function) Discriminant Analysis, LR Bayesian methods applied to both oral and written texts, use of reference corpora, and analysis of lexical richness, amongst others). The analysis described in this paper presents experimental research in progress conducted at our forensic linguistics laboratory on the establishment of sequences of linguistic categories (Morpho-syntactic Annotated Tag Sequences (MATS), or n-grams) as potential discriminatory markers of authorship. Table 1 illustrates the variable under consideration. Table 1: Morpho-syntactically Annotated Tag Sequences (MATS) Source: Turell (2010:233) El comercial de la empresa vendía los mejores hoteles … AMS N5-MS P AFS N5-FS VDA6S AMP ADJ N5-MP …

bigram

trigram

In this table it can be observed that once tagged, the sentence: ‘el comercial de la empresa vendía los mejores hoteles’ (‘the firm's salesperson was selling the best hotels’) is projected and represented. Two examples of MATS are marked, namely, ‘el comercial’ (AMS N5-MS), which is a bigram and ‘el comercial de’, (AMSN5-MS P), which is a trigram, and where A stands for article, M for masculine, S for singular, N5-MS for singular masculine common noun, and P for preposition. This approach is not new in non-forensic contexts. Recent (Feiguina and Hirst 2007; Nazar and Sánchez Pol, 2007; Spassova, 2007; Spassova and Turell, 2007) and less recent (Baayen, van Halteren et al., 1996; Statamatos, 2000) studies have shown the discriminatory potential quality of these markers across genres and authors, establishing that the discriminatory efficiency of these sequences is higher with long text samples and with a large number of reference control texts (Grant, 2007) and that bigrams (combination of two categories, also known as tag pairs) and trigrams (combination of three categories, or triplets) are more discriminatory than longer sequences of linguistic categories. Our laboratory has been one of the first to apply this method to real forensic cases. 2. Method The method used to explore the discriminatory potential of this marker of authorship, namely, tag sequences as representation of combinations of linguistic categories, is structured around the following activities: 1.

2.

A pre-processing phase, in which texts are segmented into their basic components: title, paragraphs, sentences, and paragraph beginnings and ends are marked (< =s >< =p >). A morpho-syntactic tagging phase, during which the text is converted into a row of token types and tags. -193-

3. 4.

5.

A disambiguation stage, through which texts are disambiguated and errors are corrected. A tag extraction phase—making use of LEGOLAS 2.01—during which the information obtained refers to the number of MATS types and tokens and on the MATS frequency values to be used in the subsequent statistical analysis. Once the tags have been extracted, a last stage involves the application of Linear Discriminant Analysis (LDA), in order to classify the different text sets, and the projection of results onto graphs.

During the pre-processing and processing phases several processing and disambiguation tools from the IULA’s technical corpus were used2. 3. Research stages So far, this on-going research work has involved two stages, whose main tasks are summarized in Figure 1.

Figure 1: Research stages

Stage 1 During Stage 1, by drawing from results derived from Spassova (2009), the discriminatory potential of n-grams was evaluated vis-à-vis a real-world corpus of Spanish Novels (N)—17 (8 women; 9 men) and Newspaper Articles (NA)—10 (4 women; 6 men) and a corpus of real forensic cases. Table 2 shows the distribution of writers considered by sub-corpus, gender and linguistic variety (P stands for Peninsular Spanish; A stands for American Spanish) during this stage.

1

LEGOLAS 2.0 is a tool that has been developed by ForensicLab for the automatic extraction of linguistic tags. A new version of this extraction tool, LEGOLAS 2.1, is being programmed and will soon be commercialized. 2 http://brangaene.upf.es/plncorpus/index2.htm ; http://brangaene.upf.es/plncorpus/faq.html

-194-

Table 2: Distribution of writers by sub-corpus, gender and linguistic variety in Stage 1

Subcorpora

Total N writers

N

17

NA

10

Gender

Variety

W

M

W

8

P

4

6

M

9

A

4

3

W

4

P

1

2

M

6

A

3

4

This evaluation also involved the consideration of the effect of several factors linked with the characteristics of the text such as length, textual genre and time of text production, with samples in one measurement time (MT1) and a second measurement time (MT2, 6-10 years later). The series of experiments carried out during this first research stage using Spanish written data confirmed that tag sequences of linguistic categories show a significant discriminatory capacity and that this variable can be applied to authorship attribution tasks reliably and effectively. Stage 2 Objectives The motivation behind the implementation of Stage 2 is that the analyses would be optimitized and the discriminatory potential of MATS further confirmed. The experimental studies involved in Stage 2 aim to identify the specific traits of each linguistic category implemented in those tags (within the exploited tag set), which play a major role in the correct classification of texts and those which do not, without losing sight of the fact that either their exclusion or inclusion in tag composition can help to improve this forensic linguistic comparison method. The variable In Stage 2, two types of MATS (n-grams) were used: bigrams (sequences of two grammatical categories) and trigrams (sequences of three grammatical categories). Bigrams and trigrams were chosen because preliminary studies showed that they were the most significant and discriminatory n-grams (see Spassova, 2007; 2009 and Turell, 2007). In this stage, 19 different tag sets from the IULA’s tag set were extracted on the basis of research experimentation. These tag sets were designed with the combination of the different linguistic characteristics of a grammatical category. After conducting many experiments, the most robust results during this stage were obtained with tag set No. 13, illustrated in Figure 2, which was chosen as the tag set to apply in Stage 2.

-195-

Figure 2: Linguistic characterisation of the 19 tag sets

Table 3 shows a more specific and detailed linguistic characterization of tag set No. 13. Table 3: Linguistic characterisation of tag set No. 13 IULA’s tag set

Tag set No.13

AFP

AP



Article-Plural

EP12MS

ES



Specifier-Singular

JQ—FP

JP



Adjective-Plural

N5-MS

NS



Noun-Singular

REO-1MP

RP



Pronoun-Plural

VDP2S-

V2S 

Key

Verb 2nd PersonSingular

Hypotheses In order to show the discriminatory potential quality of MATS (n-grams), three of the most common hypotheses used in forensic and non-forensic authorship attribution studies were tested, namely, a) that there seems to be more inter than intra-writer variation; b) that the written idiolectal style of an individual does seem to vary substantially throughout time, and c) that this style seems to remain quite stable across genres. The corpus used in Stage 2 to test these hypotheses is presented in Table 4.

-196-

Table 4: Corpus distribution by writer’s gender, genre and production time (Stage 2) Text information

Writers

Biological gender

MT2 NA

MT1 N

MT2 N

MT1 NA

800 words

800 words

300 words

300 words3

AP

M

10

10

9

7

JM

M

10

10

7

--

XM

M

10

10

7

6

EL

F

10

10

7

--

RM

F

10

10

3

8

LE

F

10

10

8

--

Anonymised

--

1

1

1

1

Table 4 presents the overall distribution of the two types of written texts—Novel fragments (N) and Newspaper Articles (NA)—analysed from the point of view of the MATS behaviour and produced in two differents times (MT1 and MT2 (6—10 years later)—from six contemporary Peninsular Spanish novelists. This table illustrates that in the case of the NA sample it was only possible to collect data in MT1 for only three (2 men and 1 woman) out of these six writers, something which explains the different nature of the experiments conducted and reported in the next sections of the paper. For control purposes, we also used a small corpus of four texts written by one of the novelists included in this study, whose authorship was anonymised. The statistical framework used involved Linear Discriminant Analysis (LDA)—with classification and cross-validation methods—and ANOVA (analysis of variance). 4. Results and discussion Inter-writer variation Firstly, results on inter-writer variation for both bigrams and trigrams, in both genres (N and NA) and two times of measurement (MT1 and MT2) are presented. Figures 3, 4, 5 and 6 present results for bigrams and trigrams in Novels written in MT1 and MT2. These figures illustrate inter-writer variation between the 6 novelists considered and show that both bigrams

3

Linguistic variables can sometimes show significant variation between authors by text length; however, in the case of MATS, Spassova’s (2009) study proved that a difference in text length between Novels (800 words) and Newspaper Articles (300 words) did not have an effect on the significance and reliability of results.

-197-

and trigrams discriminate well between authors, although trigrams appear to be more discriminatory than bigrams, both in MT1 and MT2.

Figure 3: Inter-writer variation—Novels—MT1—Bigrams

Figure 4: Inter-writer variation—Novel —MT2—Bigrams

-198-

Figure 5: Inter-writer variation—Novels—MT1—Trigrams

Figure 6: Inter-writer variation—Novels—MT2—Trigrams

-199-

Figures 7, 8, 9 and 10 present results for bigrams and trigrams in Newspaper Articles (NA) written in MT1 and MT2 by 3 writers. These figures illustrate inter-writer variation and also show that both bigrams and trigrams discriminate well between authors, both in MT1 and MT2.

Figure 7: Inter-writer variation—Newspaper Articles—MT1—Bigrams

Figure 8: Inter-writer variation—Newspaper Articles—MT2—Bigrams

-200-

Figure 9: Inter-writer variation—Newspaper Articles—MT1—Trigrams

Figure 10: Inter-writer variation—Newspaper Articles—MT2—Trigrams

-201-

The just mentioned patterns are confirmed when the LDA methods of classification and cross-validation are compared in Figures 11 and 12. In this case, the data considered involve only Novels (N) and Newspaper Articles (NA) from 3 Peninsular Spanish authors in MT1 and MT2, since it was only possible to obtain NA in MT2 from 3 authors.

Figure 11: Correct classification of texts in N and NA (bigrams and trigrams) in MT1

Figure 12: Correct classification of texts in N and NA (bigrams and trigrams) in MT2

In order to illustrate experimental control, results on one of the control experiments carried out using an anonymised text from one of the sample authors are presented in Figures 13 and 14, where the projection in MT1 of bigrams and trigrams, respectively, in Newspaper Articles (NA) from six writers can be observed. These figures show that the anonymised text is correctly attributed to writer AP.

-202-

Figure 13: Bigrams—MT1—6 NA writers + anonymised text

Figure 14: Trigrams—MT1—6 NA writers + anonymised text

-203-

Intra-writer variation Secondly, results on intra-writer variation for both bigrams and trigrams are presented, taking into consideration the effect of time (MT1 and MT2). Results obtained for the Novel corpus are shown in Figure 15, which illustrates that the behaviour of bigrams and trigrams does not vary very much thoughout time, thus confirming hypotheses 2, and also hypotheses 1, since evidence is provided that there is much more inter-writer than intra-writer variation. In any case, for most writers, trigrams exhibit a much more constant behaviour along time, showing lower intra-writer values, than bigrams.

Figure 15: Intra-writer variation in Novels according to time (bigrams and trigrams)

Another dimension of intra-writer variation has to do with textual genre. As Figure 16 illustrates intra-author variation according to textual genre in MT1 is smaller when trigrams are considered in the analysis, but the values are higher than in the case of the variable of time. Thus, hypotheses 3, on the effect on genre, is only partially confirmed, something that was expected since Novel and Newspaper Article are genres substantially and structurally divergent4.

4

Experiments have also been conducted about the effect of gender on the specific behavior of MATS. Preliminary and so still inconclusive results show similar patterns to those found by Spassova and Grant (2008).

-204-

Figure 16: Intra-author variation according to genre in MT1 (bigrams and trigrams

Most frequently used MATS Having shown that Morpho-syntactically Annotated Tag Sequences (MATS), in particular bigrams and trigrams, exhibit a discriminatory potential in Spanish written literary texts, it is considered that in order to evaluate this potential further, one relevant issue would be to specify the nature of the most frequently used MATS, results which could involve direct applications to real forensic cases involving written Spanish texts and also contribute to the long-needed base rate knowledge distribution of morpho-syntactic markers. Figures 17, 18, 19 and 20 illustrate this distribution for novels written by six writers in Peninsular Spanish. Results show that: a. The distribution by writer of the most discriminant bigrams and trigrams in Peninsular Spanish for the six writers analyzed in MT1 and MT2 includes Noun Phrases (NP) of the type (Specifier + Noun; Article + Noun), Adjective Phrases (AdjP), Prepositional Noun Phrases (PNP) and, to a lesser extent, Verb Phrases (VP). b. There is more inter-writer than intra-writer variation. c. For those bigrams and trigrams that are equally used in MT1 and MT2 by the same speaker (for example, trigram AS-NS-P), their discriminatory quality potential remains quite stable throughout time.

-205-

Figure 17: Most frequently used MATS—Novels—Bigrams—MT1

Figure 18: Most frequently used MATS—Novels—Bigrams—MT2

-206-

Figure 19: Most frequently used MATS—Novels—Trigrams—MT1

Figure 20: Most frequently used MATS—Novels—Trigrams—MT2

5. Conclusions It is hoped that this paper has shown that: a. The new tag system adopted in Stage 2 has optimitized the categorization of Morphosyntactically Annotated Tag Sequences (in particular, bigrams and trigrams) and their measurement as potential markers of authorship.

-207-

b. Most frequently produced bigrams and trigrams can be used to establish that there is more inter-writer than intra-writer variation as to the specific behaviour of these markers. c. Most frequently used bigrams and trigrams can characterize an individual writer’s idiolectal style, which does not seem to vary substantially throughout time but can vary slightly according to textual genre. 6. Final remarks The results presented in this paper also illustrate that there is a need to conduct further research along several directions. On the one hand, more experiments need to be carried out with bigger samples of speakers and more extensive corpora; on the other, the methods used to establish the discriminatory potential of Morpho-syntactically Annoated Tag Sequences (MATS) need to be extended to real forensic texts, in order to establish this measurement applicabilty as a more conclusive method for the purposes of forensic linguistic profiling. References Baayen, R.H., van Halteren, H. and Tweedie, F. J. (1996) Outside the cave of shadows: Using syntactic annotation to enhance authorship attribution. Literary and Linguistic Computing 11(3): 121—131. Feiguina, O. and Hirst, G. (2007) Authorship attribution for small texts: Literary and forensic experiments. Paper presented to the International Workshop on Plagiarism Analysis, Authorship Identification and Near-Duplicate Detection. 30th Annual International ACM SIGIR Conference (SIGIR '07). Grant, T. (2007) Quantifying evidence in forensic authorship analysis. The International Journal of Speech, Language and the Law 14(1): 1—25. Nazar, R. and Sanchez Pol, M. (2007) An extremely simple authorship attribution system. In M. T. Turell, M. S. Spassova and J. Cicres (eds) Proceedings of the Second European IAFL Conference on Forensic Linguistics/Language and the Law 197— 203. Barcelona: Publicacions de l’IULA. Spassova, M. S. (2007) The relevance of inter and intra authorial variation in authorship attribution. Some findings on syntactic identification markers. Paper presented at the 8th Biennial Conference on Forensic Linguistics/Language and the Law. University of Washington, Seattle. Spassova, M. S. (2009) El potencial discriminatorio de las secuencias de categorías gramaticales en la atribución forense de autoría de textos en español. Unpublished PhD thesis, Universitat Pompeu Fabra. Spassova, M. S. and Grant, T. (2008) Categorizing Spanish written texts by author gender and origin by means of Morpho-Syntactic Trigrams: some observations on method’s feasibility of application for linguistic profiling. Paper presented at the Curriculum, language and the law Inter-University Centre, University of Zagreb, Dubrovnik, Croatia. -208-

Spassova, M. S and Turell, M. Teresa. (2007) The use of morpho-syntactically annotated tag sequences as forensic markers of authorship attribution. In M. T. Turell, M. S. Spassova and J. Cicres (eds) Proceedings of the Second European IAFL Conference on Forensic 456 Linguistics/Language and the Law. 229—237. Barcelona: Publicacions de l’IULA. Stamatatos, E., Fakotakis, N. and Kokkinakis, G. (2000) Automatic text categorization in terms of genre and author. Computational Linguistics 26 (4): 471—495. Turell, M. Teresa. (2010) The use of textual, grammatical and sociolinguistic evidence in forensic text comparison. The International Journal of Speech, Language and the Law 17 (2): 211—250.

-209-

Whose Tweet? Authorship analysis of micro-blogs and other short-form messages Nicci MacLeod1 and Tim Grant2 Centre for Forensic Linguistics, Aston University, UK 1 [email protected] 2 [email protected] Abstract Approaches to authorship attribution have traditionally been constrained by the size of the message to which they can be successfully applied, making them unsuitable for analysing shorter messages such as SMS Text Messages, micro-blogs (e.g. Twitter) or Instant Messaging. Having many potential authors of a number of texts (as in, for example, an online context) has also proved problematic for traditional descriptive methods, which have tended to be successfully applied in cases where there is a small and closed set of possible authors. This paper reports the findings of a project which aimed to develop and automate techniques from forensic linguistics that have been successfully applied to the analysis of short message content in criminal cases. Using data drawn from UK-focused online groups within Twitter, the research extends the applicability of Grant’s (2007; 2010) stylistic and statistical techniques for the analysis of authorship of short texts into the online environment. Initial identification of distinctive textual features commonly found within short messages allows for the development of a taxonomy which can then be used when calculating the ‘distance’ between messages containing instances of these feature types. The end result is an automated process with a high level of success in assigning tweets to the correct author. The research has the potential to extend the scope of reliable and valid authorship analysis into hitherto unexplored contexts. Given the relative anonymity of the internet and the availability of cloaking technology, linguistic research of this nature represents a crucial contribution to the investigative toolkit. Keywords:

AUTHORSHIP ANALYSIS; STYLISTIC METHODS; STATISTICAL METHODS; ONLINE MESSAGING

1. Introduction It has been widely noted that there is increasing use of online communications for the organisation and dissemination of a wide range of criminal activities and material. The fundamental anonymity offered by the internet and the ease with which multiple identities can be created enables individuals to share such information in relative security. State-of-theart work in authorship analysis has had considerable success for cases where there is a small and known set of authors, and sufficient quantity of text of known authorship. These methods do not easily translate into computer-mediated communication where there may be a large and unknown number of authors all contributing an unknown number of short messages. This paper reports on a project that extended existing work in forensic linguistics that had been successfully applied (at evidential standard) to criminal investigations involving SMS text messages, by developing an automated process that can be applied to online environments by non-specialist users. -210-

Recent attempts to develop methods for attributing authorship have emerged from two broad disciplines—linguistics (e.g. Chaski, 2001; Grant & Baker, 2001; McMenamin 1993; 2002) and computing (e.g. Argamon, 2008; Hoover, 2003; Koppel et al. 2006; 2011). Traditionally concerned with literary, biblical and political texts, interest has shifted in recent times to the identification of authors of shorter texts such as blogs (Koppel et al. 2011) and SMS texts (Grant, 2010). As Zheng et al. (2005) point out, the misuse of online messages for inappropriate and/or illegal purposes has become a serious concern in recent times. Aside from the ease of anonymity for online authors and the brevity of the texts, difficulties in establishing robust methods have been compounded by the large and open ended nature of the set of potential authors in this context. Online texts are ‘shorter, noisier and they have a greater number of candidate authors’ (Abbasi & Chen, 2005: 67). Features such as relative frequencies of function words and word frequency distributions have traditionally been brought together in multivariate models for attributing authorship, and indeed the individual’s variation in their use of function words remains a popular method to this day (Grant & Baker, 2001). Other researchers in the area (e.g. Miranda-García & Calle-Martín, 2005; Smith & Kelly, 2002) have had some success with lexical richness (the frequency of rare words, e.g. hapax legomena and hapax dislegomena) and repetition (the frequency of common words). Furthermore, average word, sentence, clause and paragraph lengths, word type frequencies and distributions, collocation and content analysis have all been utilised for the task, although it has been noted that these are often used in combination for maximal discriminatory power—identification is achieved through an aggregate of markers (Grant & Baker, 2001; McMenamin, 2001). Chaski’s (2001) approach, although not without its critics on account of some significant methodological weaknesses (e.g. Grant & Baker, 2001; McMenamin, 2001) tested a number of features for authorship analysis, including syntactic analysis, syntactically classified punctuation, sentential complexity, vocabulary richness, readability, content analysis, spelling errors, punctuation errors, word form errors, and grammatical errors, and found that only syntactic analysis and syntactically classified punctuation successfully discriminated and clustered documents. Koppel et al. (2011) note that almost all existing research in the field of authorship attribution ‘considers only the simplest version of the problem’ (p. 84), that is to say, those instances where a relatively long anonymous text is attributed to one of a small, closed set of candidates. As they point out, this version of the authorship attribution is rare in the real world—conversely, we are often faced with the potential of thousands of candidate authors; the possibility that none of the known candidates authored the text; and the likelihood that either the known texts and/or anonymous text may be limited. Addressing these limitations, Koppel et al. (2006) report on their own technique for solving authorship attribution even when the candidate set numbers in the many thousands. With a test candidate set of 10,000 bloggers, they aim to determine which individual authored a given 500 word snippet. Their approach involves determining whether a given snippet includes a set of linguistic features unique to a given author. Their results showed that this rather crude approach worked to a certain extent, but that only when a response of Don’t Know was permissible was the method able to achieve reasonably reliable attribution of snippets in the case of thousands of authors. Koppel et al. (2011) describe existing methods for automated authorship attribution as falling into two paradigms—the similarity based paradigm, where the distance between two documents and an anonymous document is measured, and attribution is based on the author whose known writing has more in common with the questioned text; and the machine learning paradigm, where the known writings of each candidate author are used to construct a -211-

classifier, which is then used to classify anonymous documents. The authors point out that similarity based methods are more appropriate when considering a large volume of candidate authors (Koppel et al., 2006), and that using these methods allows for a document to be verified as having been written by a given author ‘if the similarity between the document and the author’s known writing exceeds some threshold’ (2011: 85). They take 4-grams (strings of characters of length four that include no spaces, or strings of four or fewer characters surrounded by spaces) as the basis for their analysis. Character n-grams have been shown to be effective for authorship attribution, and Koppel et al. point out that one advantage is their measurability in any language without the need for specialist background knowledge. However, from a linguistic perspective they lack salience, much like the features focussed on by the early stylometrists: ‘in forensic analysis there are obvious dangers in computationally pursuing an algorithm which distinguishes authors and yet has no linguistic explanation or validity’ (Smith et al, 2009). Koppel et al’s method was shown to be successful in 46% of cases, which rose to 93.2% precision after the introduction of a ‘Don’t Know’ option. The authors conclude that their method represents an effective means of handling large candidate sets for which traditional categorization methods were ineffective, but acknowledge that the case of small open candidate sets and limited anonymous text has, as yet, no satisfactory solution. Burrows (2002), noting that existing methods in computational stylistics are ‘better fitted for ‘closed’ games than open ones’ (p.267), offers a method for authorship attribution which is suited to those cases where there is little or no outside evidence to identify the most likely candidate. Burrows points out that most methods currently employed in the area rely on multivariate statistical comparison between certain features of a given example, and an appropriate set of norms. These comprise the frequencies of relatively simple phenomena, and can include alphabetical characters, whole words, or common grammatical forms. As Burrows points out, the advantage of working with whole words lies in their ‘accessibility and meaningfulness’ (2002: 268), while it has become customary to allow particular variables to “declare themselves’, thus obviating...the danger of a pre-determined outcome’ (2002: 268). He goes on to explain that a large set of variables that are weak discriminators is likely to offer better results than a small set of strong ones, given that strong discriminators are susceptible to being recognised and manipulated by users. As he succinctly puts it, ‘a distinctive ‘stylistic signature’ is usually made up of many tiny strokes’ (2002: 268). The procedure he develops is, he claims, successful in distinguishing the most likely author of texts exceeding 1500 words—but, more relevant to our own purposes, of even greater value in reducing the pool of likely candidates for texts as short as 100 words. Moving on to authorship attribution methods more obviously rooted in linguistic theory, McMenamin (2010) outlines his approach to forensic texts, which is grounded firmly in stylistics—‘the scientific interpretation of style-markers as observed, described and analysed in the language of groups and individuals’ (McMenamin, 2010: 488). Conceptualising style markers as ‘the observable result of the habitual and usually unconscious choices an author makes in the process of writing’ (2010: 488), he goes on to distinguish between a) the choice between optional forms and b) deviations from the norm. Deviations from the norm may often be associated with particular classes of people, as in the case of mixing up homonyms such as ‘your’ and ‘you’re’ or ‘their’ and ‘there’—deviations that could be ‘common to careless or undereducated writers’ (2010: 489), or the use of ‘then’ for ‘than’, which could be indicative of a particular linguistic variety in which these forms are homonymous. These features, then, are unlikely to be individuating, although their coselection could be. -212-

McMenamin distinguishes between the consistency model, used to determine if particular texts were written by the same author, and the population model which must be used when the pool of candidates is large, i.e. not limited to one or two suspect writers: ‘in this instance, the resemblance model is used repeatedly on one author after another until all are excluded’ (2010: 490). McMenamin’s approach is largely qualitative, as reflected in his assertion that ‘linguistic assessments of style precede their expression as numerical values and are often a more realistic representation of the facts’ (2010: 491), and this focus has been maintained by others, such as Coulthard (reported in Grant, 2010), although other research such as Grant & Baker (2001) and Grant (2010) has sought to quantify the selection and significance of style markers. Grant (2010), in discussing authorship attribution of SMS text messages, explains how linguistic distinctiveness and linguistic consistency are matters of degree, and that questions of both can be explored using statistical methods. He calls for descriptive methods to be developed further, particularly in terms of enhancing them to enable the quantifiable comparison of consistency and distinctiveness. To this end, Grant utilises Jaccard’s coefficient, a statistical tool for establishing degrees of similarity between cases. The presence or absence of each stylistic feature identified above a certain frequency within the corpus is coded, as 1 or 0 respectively. These codings then allow for statistical comparison for similarity or dissimilarity. Jaccard’s coefficient can be used to compare pairs of messages each of which is coded as a series of zeros and ones relating to the absence or presence of specific linguistic features. Jaccard is essentially a correlation coefficient applied to these binary strings and results in a (dis)similarity metric which resolves to a decimal figure between zero and one where one indicates the two text messages contain identical linguistic features and zero indicates no linguistic features in common. An important advantage of Jaccard is that a match of two absence scores across two texts has no effect on the overall similarity score (Smith et al., 2009). As short-form messages are indeed short the absence of a given feature from a text carries no meaning and does not affect the calculation of similarity in either direction. Building on the work of Grant (2010) the current project uses an extension of Jaccard called Delta-S (Δs). Delta-S was developed in marine biology and forensic psychology (Woodhams, Grant and Price, 2007) to allow the weighting of variables within a Jaccard calculation as being related to one another. In short-form messaging this requires a taxonomic description to be developed which declares, for example, substitution of different digits in a text to represent more similar stylistic choices than an accent stylisation. The taxonomy developed in this project is described and discussed below. The power of Δs is that it allows the recognition of similar but not identical stylistic choices to be represented in the final similarity metric. 2.

Methodological approach

2.1. The data The dataset analysed in the current study is a corpus of microblogs sourced from the social networking site Twitter (see Figure 1). Microblogging is a form of communication in which users can describe their current status in short posts distributed by instant messages, mobile phones, email or the Web. Twitter is a relatively new method of mass communication, operating in real-time and designed for mobility (Chang, 2010). -213-

Figure 1: Twitter Screenshot

Because users do not require knowledge of any standardised interaction technique, they are able to customise Twitter to suit their needs, resulting in ‘a diverse user base using the service for heterogeneous ends’ (Efron & Winget, 2010). There are a number of terms that have sprung up from the Twitter community to aid in organisation and readability. The prefacing of a tweet with ‘RT’ (‘Retweet’) indicates that it is a reposting of another user’s tweet, while the use of the hashtag—prefacing a word with the symbol ‘#’—is a convention allowing the filter of tweets by topic (Crystal, 2011; Eliot, 2009), and thus serves as a ‘bottom-up userproposed tagging convention’ (Chang, 2010: 1). Users’ guides such as twittonary.com offer definitions for words purported to be specific to the Twitter context, but the extent to which these are actually drawn on by users remains unclear. 2.2. Feature selection While there has been a dramatic increase in the use of microblogging services over the last four years, research into the linguistic features of the texts and the habits and motivations of its users remains minimal (Efron & Winget, 2010). One contribution comes from Crystal (2011), who notes that tweets display a two-part structure, the first being the user’s name and the message itself, and the second containing metadata, including its temporal source and Internet origin. Narrowing the focus to the internal grammatical structure of the message, he notes that the use of nonstandard punctuation often makes it difficult to assign tweets unambiguously to a particular syntactic category. Many tweets take a rather fragmented form, and words are sometimes ‘juxtaposed in a way which makes an immediate interpretation impossible’ (2011: 45). Crystal notes that the average number of words per tweet in his corpus was 14.7, observing that this is higher than is the case for Instant Messages (IM). He also shows ellipsis of the subject and auxiliary verb to be a frequent occurrence in tweets. Based on his corpus, he argues that within tweets there is not the same range of texting abbreviations as in SMS. This brief discussion is concluded with the observation that Twitter is a ‘variety in evolution’, the norms of which are still in the relatively early stages of development. The table below demonstrates the features extracted in previous work by Grant and colleagues (Smith et al., 2009) in the area of authorship attribution of SMS texts, for calculation of the Delta-S metric—a more robust version of Jaccard’s co-efficient (Smith et al., 2009).

-214-

Table 1: Original SMS features list (from Smith et al, 2009) Feature Mispellings Lower case ‘I’

Description Any word not found in an English dictionary Non-captialisation of the word “I”

Example “I saw it on the news this mroing” “i don’t think so”

Acronyms

Use of acronyms

“Who are you, the CIA?”

‘G’ clipping Accent stylisation

Dropping the final ‘g’ of words Using phonetic spelling to convey a specific accent Using onomatopoeia to convey an exclamation Conveying specific pronunciation through spelling

“I’m only askin” “Dey don’t fink dat it could happen to dem ” “Boom, you’re dead”

Whole word letter homophone substitution

Replacing entire words with a single letter

“R U still coming out tonight?”

Syllable homophone substitution Whole word number homophone substitution

Replacing syllables within words with a single letter Replacing entire words with a number

“It doesn’t matter ne way”

Syllable number homophone substitution

Replacing syllables within words with a number

“wait until 2moro”

Whole word typographic homophone substitution

Replacing entire words with a character

“Meet you @ the bus stop”

Syllable typographic homophone substitution

Replacing syllables within words with a character

“I don’t know anything about th@”

Shortenings

Common words shortened to a few initial letters Series of characters used to represent faces Commonly used phrases reduced to their initial letters Use of a single exclamation mark

“I need to do this by Sep 10th” “:-)”

Use of a multiple exclamation mark

“No way!!!!!!!!!!!”

Use of a mixed characters to convey an exclamation

“What the hell?!?!?!?”

Exclamatory onomatopoeia Prosodic emphasisers

Emoticons Initialisms Singular typographic exclamation Multiple typographic exclamation Mixed typographic exclamation

“Booooooring”

“What are you waiting 4?”

“ASAP” “No way!”

Further to this list, a detailed reading of existing computer-mediated communication (CMC) literature (Crystal, 2008, 2011; Ling & Baron, 2007; Thurlow & Brown, 2003) contributed to the initial set of the type of stylistic features we could expect from our data. Since the linguistic analysis of micro-blogging communication is a relatively new field, the initial list included features of a number of other CMC genres including SMS and Instant Messaging. The second step in the feature extraction was wholly data driven. Drawing on a development set of around 18500 tweets, a qualitative analysis was performed with the assistance of Wordsmith Tools (Scott, 2008) to identify occurrences of some of the features initially provided by the literature review. Lexicons containing every example of a given feature as it appeared in the corpus were then created by manually extracting items from the -215-

Wordlist tool and creating plaintext files for the programmers. Figure 2 shows a section of the exclamatory onomatopoeia lexicon.

Figure 2: Exclamatory Onomatopoeia Lexicon (Extract)

Lexicons were developed in this way for features such as exclamatory onomatopoeia, initialisms and acronyms, for example, but were less suitable for features such as whole word numeral substitution. For these, a rule needed to be developed to allow the computer to distinguish between 2 the number and 2 representing the words to (preposition), to (infinitive marker) and too (adverb). Further to this, it was considered entirely feasible that some authors might regularly substitute the infinitive marker with the numeral (for example), but rarely do so for the prepositional use, and vice versa. Thus, a series of grammatical rules were developed by the first author, which attempted to distinguish the different uses on the basis of the context in which they occurred. There were further distinctions to be made on the basis of spacing—evidence was found in the development corpus of some users inserting a space before and after the 2, while others inserted one before but not after, others vice versa, and some none at all, relying on the height difference of the character to visually break up the individual words. Some examples appear below. a) New article on darkpolitricks: GORDON DUFF: ROGUE AMERICA2 (number, no pre-space) b) @Skepta 2girls 1cup :| (number, pre-space, no trailing space) -216-

c) @chilemad having a one 2 one (preposition, pre-space, trailing space) d) RT @thekatvond: Honking ur horn lk a crazyperson in stopped traffic is a gd example of not bein able 2 accept the uncontrollable-it's al ... (infinitive, pre-space, trailing space). What resulted from this process was the refining of the original feature category into thirtytwo separate features based on all the possible combinations of a) the numeral used, b) which item was replaced, and c) the use of spacing. These distinctions were made at the bottom of the feature categorisation system—the top four layers are illustrated in Figure 3 overleaf. As the illustration shows, the top-most level at which features were classified was on the basis of lexis, grammar and punctuation, as well as by features peculiar to the mode of production, including hashtags and re-tweets. The features classified under the grammar heading related mainly to omission of particular classes of word such as verbs and determiners—patterns fairly typical of ‘telegraphic’ speech, which, as expected, are not particularly useful in assessing matters of authorship. Thus, the analysis focussed in the main on the features classified under lexis and punctuation. The Delta-S (Δs) distance metric was used to determine and measure distance between two documents by using the presence, and position in the hierarchy, of the stylistic features. 3. Evaluation After developing the feature set the next step was to test the method’s effectiveness at a number of tasks, the end task being the identification of the likely author for a single or small number of short messages, an ‘open’ problem, where: · There are many unknown potential authors in the candidate set; and · The author of the unknown message may not be present in the candidate set. The aim was for the system to be able to provide one of the following responses: · Author Identified: · Potential Author: · Undetermined: · Not Present:

the results exceed a specific confidence level; the results approach the confidence level; the message contains too few stylistic features to make a judgement above a determined confidence threshold; the author is unlikely to be present in the candidate set.

Where only a small set of messages are available from an unknown author, the decision was made to aggregate these messages in an attempt to improve the chances of attribution. However, it must be borne in mind that, particularly in an online context, there is no guarantee that all the texts in a suspect set were authored by one individual, since a number of authors may have access to a particular micro-blogging account.

-217-

Figure 3: Top four levels of the feature categorisation system.

3.1. Attribution confidence We reflect here only on the ability of the methodology to discriminate authors of messages: issues concerning scalability to very large datasets are to be the focus of a future project. The Delta-S (Δs) metric is calculated between a single test message of Author X and a set of comparison messages from authors (A1 , A2 , A3 ,... An) ; (B1 , B2 , B3 ,... Bn), (C1 , C2 , C3 ,... Cn), etc. This gives a series of samples: A{ Δs(X→A1), Δs(X→A2), Δs(X→A3) … Δs(X→AN) }; B{ Δs(X→B1), Δs(X→B2), Δs(X→B3) … Δs(X→BN) }; C{ Δs(X→C1), Δs(X→C2), Δs(X→C3) … Δs(X→CN) } etc. -218-

These can then be compared with a non-parametric statistical significance test (in this case Mann-Whitney U-test) to determine the concordance probabilities: P( A > B ), P( A > C) , P( B> C ) etc. These represent the probability that both samples can be drawn from the same set, with a low probability indicating more significant differences. 3.2. Performance for single messages The initial test took 10 single, random tweets from a known Author A (Xa), and generated the Δs distance measures to 100 other tweets from Author A, and 100 tweets from author B. Three experiments were carried out each with different authors and test messages. The results are presented in Table 2 , which shows the number of messages identified correctly (out of 10 for each trial), the number of these identified correctly with high statistical significance, and the number of messages that could not be assigned. In no case were messages incorrectly linked. Table 2: Performance for Single Messages

The results show reasonable accuracy and discrimination for a single message, with the correct author identified in the majority of cases, and many of these assigned with a significant level of certainty. Furthermore, no messages were incorrectly assigned to an author. One of the reasons a number of messages could not be assigned is the frequent sparsity of features within such short messages. In this dataset a tweet is typically 12 words long and on average contains fewer than 3 stylistic features. Thus, an approach needed to be developed which could allow for the fact that some will contain many identifying features, whilst others will contain few or none. 3.3. Performance for aggregated messages The effect of feature sparsity can be reduced by aggregating messages before the Δs calculation. This second test aggregated random tweets from a known author A into 10 batches of 1, 2, 5 or 10 messages each. The Δs distance measures were then calculated for each of these aggregations to 100 batches of other messages from author A and 100 batches of messages from author B. The authors used in this trial were that same as in Experiment 2, the worst performing from the single text test.

-219-

Table 3: Performance for Aggregated Messages

The results show an improvement in performance after aggregation. The number of words in the aggregated messages averages around 90 for the 10 message case; still well below the lower limits of stylometric techniques. This author is fairly typical, using an average of 9 words per message, with each message containing an average 2.5 stylistic features. The increase in performance is striking even for modest levels of aggregation. Again, no messages were incorrectly assigned to an author. 3.4. Performance for multiple authors The next scenario that was considered was one in which multiple authors are present in the candidate set. The test took 10 single, random tweets from a known Author A (Xa), and generated the Δs distance measures between these and 100 other tweets from Author A , 100 tweets from author B, 100 from author C, D etc. Candidate sets of 2, 5, 10 and 20 authors were considered. The results are shown in Table 4. For each message, the rank order (with 1 being the most similar and 20 being the greatest distance) shows the ranking of the ‘correct’ author as the likely author of the message in question. The table also shows the level of significance with which messages were incorrectly assigned when they were not ranked first.

-220-

Table 4: Performance for Multiple Authors

These results show that the methodology has reasonable success identifying authors from a set of 20 authors, a relatively large candidate author set. Of particular interest is that in those cases where the correct author is not top ranked, the significance of the result is not definitive. In practical terms this minimises the risk of false positive results. 4. Concluding remarks This paper has demonstrated that positive results are possible for typical short message content (SMS text and Twitter), and the approach reported on here advances the state of the art in terms of the size of message to which authorship analysis can be applied. The implementation of feature identification has proved effective in terms of the accuracy and coverage of the feature instances identified and annotated per message. However, some improvements could be made to increase overall performance. Time did not allow, for example, for rigorous part-of-speech tagging of the corpus, which would have allowed for greater use of the rule-based feature categorisation as reported on in section 2.2. Furthermore, although a number of detailed lexicons were developed for feature categorisation tasks (such as onomatopoeic expressions and various sub-categories of initialisms), there are a number of other feature types for which this remains to be completed. Further improvement to the identification process may be possible by weighting particular features according to how common or rare they are. This would mean that the presence of a very common phrase initialism such as ‘LOL’ (‘Laugh Out Loud’) in both the questioned text and a candidate set of texts would receive a lower weighting than the presence of a rarer one, such as BBIAB (Be Back In A Bit). All these improvements would be likely to contribute to a more refined system with even higher success rates. A further issue for future research is the scalability of the process. In light of the practical reality of online messaging, any operationally useful system would need to generate valid results on the very large data sets typical of the context. -221-

Acknowledgements Special thanks to Darrell Smith & colleagues at Lexegesys Ltd, with whom we collaborated on this project. References Abbasi, A, & Chen, H. (2005) Applying authorship analysis to extremist-group web forum messages. IEEE Intelligent Systems 20 (5), 67—75. Argamon, S. (2008) Interpreting Burrows’s Delta: geometric and probabilistic foundations Literary and Linguistic Computing 23 (2), 131—147. Burrows, J. (2002) ‘Delta’: a measure of stylistic difference and a guide to likely authorship. Literary and Linguistic Computing 17 (3), 267—287. Chang, H.-C. (2010) A new perspective on Twitter hashtag use: diffusion of innovation theory. Proceedings of the American Society for Information Science and Technology 47: 1—4. Chaski, C. (2001) Empirical evaluations of language-based authorship identification techniques. International Journal of Speech, Language & the Law 8 (1), 1—65. Crystal, D. (2008) txtng: the gr8 db8. Oxford: OUP. Crystal, D. (2011) Internet Linguistics: A Student Guide. Abingdon: Routledge. Efron, M. and Winget, M. (2010) Questions are content: a taxonomy of questions in a microblogging environment. Proceedings of the American Society for Information Science and Technology 47: 1—10. Eliot, G. (2009) Common Twitter terms: using and understanding the language of Twitter. http://www.suite101.com/content/common-twittertermsa101868#ixzz1Ega53YeR Grant, T. (2007) Calculating TXTual distance in forensic authorship analysis. Paper presented at the International Association of Forensic Linguists 8th Biennial Conference, University of Washington, Seattle, USA, July 12—15, 2007. Grant, T. (2010) Text messaging forensics: txt 4n6: idiolect free authorship analysis? in M.Coulthard & A. Johnson (eds) The Routledge Handbook of Forensic Linguistics, 508—522. Grant, T. & Baker, K. (2001) Identifying reliable, valid markers of authorship: a response to Chaski. International Journal of Speech, Language & the Law 8 (1), 66—79. -222-

Grant, T., MacLeod, N., Exell, A., Smith, D., Spencer, S. & Webb, A. (2011) Authorship analysis for short form messages. Unpublished Research Report, Aston University/Lexegesys. Hoover, D.L. (2003) Multivariate analysis and the study of style variation. Literary and Linguistic Computing, 18 (4), 341—359. Java, A., Song, X., Finin, T. & Tseng, B. (2007) Why we twitter: understanding microblogging usage and communities Proceedings of the 9th WebKDD and 1st SNA-KDD 2007 workshop on Web mining and social network analysis, http://portal.acm.org/citation.cfm?id=1348556 Koppel, M., Schler, J. & Argamon, S. (2011) Authorship attribution in the Wild. Language Resources & Evaluation 45, 83—94. Koppel, M., Schler, J., Argamon, S. & Messeri, E. (2006) Authorship attribution with thousands of candidate authors. Proceedings of the 29th ACM SIGIR Conference on Research and Development on Information Retrieval Seattle, Washington. Ling, R. & Baron, N.S. (2007) Text messaging and IM: Linguistic comparison of American college data. Journal of Language & Social Psychology 26 (3), 291— 298. Mcmenamin, G.R. (2001) Style markers in authorship studies. International Journal of Speech, Language & the Law 8(2), 93—97. McMenamin, G.R. (2010) Theory and practice of forensic stylistics. in M. Coulthard & A. Johnson (eds) The Routledge Handbook of Forensic Linguistics 487—507, London: Routledge. Miranda-García, A. & Calle-Martín, J. (2005) The validity of lemma-based lexical richness in authorship attribution: a proposal for the Old English Gospels. ICAME 29, 115— 130 Scott, M. (2008) WordSmith Tools Version 5, Liverpool: Lexical Analysis Software. Smith, J. A. & Kelly, C. (2002) Stylistic constancy and change across literary corpora: Using measures of lexical richness to date works. Computers and the Humanities 36: 411— 430. Smith, D.J., Spencer, S. & Grant, T. (2009) Authorship analysis for counter terrorism Unpublished Research Report, QinetiQ/Aston University. Stamatos, E. (2008) A survey of modern authorship attribution methods. Journal of the American Society for Information Science and Technology 60 (3), 538—556.

-223-

Thurlow, C. & Brown, A. (2003) Generation Txt? The sociolinguistics of young people’s text messaging. Discourse Analysis Online 1 http://extra.shu.ac.uk/daol/articles/v1/n1/a3/thurlow2002003-paper.html Woodhams, J., Grant, T. D., & Price, A. R. G. (2007) From marine ecology to crime analysis: improving the detection of serial sexual offences using a taxonomic similarity measure. The Journal of Investigative Psychology and Offender Profiling 4, 17—27. Zheng, R., Li, J., Chen, H. & Huang, Z. (2005) A framework for authorship identification of online messages: writing style features and classification techniques. Journal of the American Society for Information Science & Technology 57(3), 378—393.

-224-

‘There are letters for you all on the sideboard’: what can linguists learn from multiple suicide-note writers? Jess J. Shapero 1 and Susan A. Blackwell2 University of Birmingham, UK 1 [email protected] 2 [email protected] Abstract Suicide notes have long fascinated forensic linguists, especially in the context of disputed authenticity or authorship, and it has long been debated how to distinguish genuine from fabricated suicide notes (e.g. Shneidman and Farberow, 1957). For this study, a corpus of 286 suicide notes covering the period 1995—1999 was obtained from the Birmingham Coroner (Shapero, 2011). A number of these were produced by writers who left more than one note. It might be expected that these would be versions of the same text adapted for different readers, but closer inspection revealed that this was far from the case. The notes were often produced for distinct purposes as well as for multiple recipients, and they tended to exhibit a surprising lack of common characteristics. This study compares the variation exhibited within writers with the variation between writers, using a semantic tagger to investigate the ‘aboutness’ of the notes. We believe that our study can assist linguists in forming judgements about the authenticity of a disputed suicide note, and demonstrates that linguists should choose their criteria carefully before concluding that dissimilarity rules out common authorship. Keywords:

SUICIDE NOTES; SEMANTIC TAGGING; AUTHORSHIP; WMATRIX; CORPUS LINGUISTICS; FORENSIC LINGUISTICS

1. Introduction Suicide notes have long fascinated forensic linguists, especially in the context of disputed authenticity or authorship. One notorious case in England was that of Eddie Gilfoyle, who was convicted in Liverpool in 1993 of murdering his pregnant wife Paula, after (the prosecution claimed) inducing her to write her own suicide note. Linguistic analysis of the disputed note in comparison with known samples by both husband and wife did not yield a conclusive result (see Goutsos, 1995). Many studies still make reference to Shneidman and Farberow (1957), who compiled a small comparison corpus of genuine and simulated suicide notes from writers in Los Angeles, USA. Their corpus comprised only 33 of each category, however, and this clearly imposes limitations on the scope of forensic linguistic research. This study makes use of a much larger body of British English data, compiled by one of the present authors in the course of her doctoral research (Shapero, 2011). The material was collected from the Birmingham (U.K.) Coroner’s Office and consists of 286 suicide notes from deceased persons whose Coroner’s inquests were held during the period 1995—1999. 74 of the notes were written by females (22 writers) and 212 by males (92 writers). The corpus thus includes a number of authors who each wrote a single text and others who produced more than one, sometimes several, texts. In all, 63 notes were written by single-note writers and 223 notes by multiple-note writers, as shown in Table 1. In the whole corpus, approximately 82% of the notes by females and 76% of the notes by males are by authors who wrote more than one note.

-225-

Table 1: Number of authors by number of notes written

No. of authors 1 1 1 3 2 3 2 3 8 9 18 63

No. of notes written 16 11 10 9 8 7 6 5 4 3 2 1

This paper focuses on the 51 multiple-note writers. These individuals each produced between 2 and 16 texts, within a short time interval, in the same circumstances and for what can be described as the same reason (although not necessarily for the same purpose; see below). They can thus be considered somewhat comparable to subjects acting as their own ‘controls’ in experimental conditions. This is probably as good as it gets in the real world of observationally obtained naturally-occurring data. The authenticity of the notes in our corpus is never in doubt, having been accepted by the Coroner. Therefore, while differences between writers may be due to authorship, any differences within writers must be explicable in terms of factors such as the writers’ intentions and target readers. This has implications for two distinct types of linguistic enquiry, at least. Firstly, a study of the similarities and differences between genuine multiple notes produced by the same writer should throw some light on questions relating to authorship attribution in disputed cases. If two notes exhibit striking similarities, is it safe to conclude that they have a common author? Conversely, if two notes purportedly by the same writer exhibit certain kinds of linguistic differences, does this cast doubt on their authenticity? Secondly, it may be possible to distinguish sub-genres of suicide notes by studying a corpus of this size. We wanted to establish whether our multiple note-writers were writing essentially the same message for several different people (in which case we would expect to find a low level of intra-author variation); or whether their notes could be described as exhibiting sufficient differences in their social context and linguistic function that they can be classed as belonging to different (sub-) genres. This paper cannot claim to provide groundbreaking answers to questions of either authorship or genre; but it explores a methodology based on semantic tagging which may offer a consistent means of analysis suitable for application to both types of problem. 2. Methodology The issue of what to count as a suicide note (for example, whether to include portions of diaries or greetings cards containing apparently suicidal messages) is dealt with in Shapero (2011). However, there were also many decisions to be made about what to include in the sub-corpus of multiple-note writer texts for this study. For example, texts on envelopes that had no contents were included, and it consequently seemed reasonable to include some texts written on envelopes that also contained letters. However, this was largely a matter of judgement that depended on the nature of the text on the envelope. Typically, a very short text on an envelope coupled with a lengthy letter inside it was counted as one note. Another issue was whether the -226-

Coroner’s records were entirely reliable when they stated, for instance, that two suicide notes had been found when there were actually a different number of notes present in the Coroner’s files. For purposes of practicality, the sub-corpus was built with only the notes in our possession which were whole and legible. Where only one text was found (or was whole and legible) it was deemed acceptable to exclude it from the sub-corpus regardless of what the Coroner’s files said (and regardless of how many other partial or barely readable notes an author had written). Finally, there was the matter of whether a note’s contents were entirely reliable. For example, an author who in one note mentioned having written other notes may not necessarily have done so. We were left with 223 notes by multiple-note writers which had been produced by 9 female and 42 male authors (which represent 40.91% and 45.65% of the female and male authors, respectively, in the whole corpus). We then decided to apply a cut-off point of a minimum of 8 notes per author, in order to obtain a reasonably comparable set of text groupings. This yielded a sub-corpus of 8 authors and 80 texts. We hypothesised that similarities and differences between suicide notes would be most usefully analysed in terms of their semantic content, rather than lexis per se or syntax. We use the term ‘aboutness’ to reference this somewhat elusive quality (see Fairthorne 1969; Phillips 1989; Rayson 2007; Scott 2007; Van Dijk 1977). The software we used to gain access to the ‘aboutness’ of our notes was Wmatrix2 (Rayson, 2007) which is capable of performing automatic semantic tagging. It is claimed to be 91-92% accurate (Rayson et al., 2004). According to Archer et al. (2002) the system recognises 232 categories. However, the categories are quite fine-grained, many having 1 to 3 ‘+’ and/or ‘-’ signs suffixed to them indicating increasing, or decreasing, degrees of the category. A frequency list of the semantic tags for all the suicide texts (in the full corpus) combined gives no fewer than 347 categories. As an example, Table 2 shows the great variety of tags for the ‘Emotion’ categories labelled ‘E’, and Table 3 shows the many variants for the category ‘E2’. When using WMatrix, the linguist has the option of comparing the tags from his/her own corpus against the frequencies generated from a comparison corpus. This could be the BNC ‘sampler’ which is built into the software (Rayson, 2007); or the user could supply their own reference corpus. Table 2: WMatrix emotion tags Semantic tag Frequency E1

11

Relative frequency 0.03

Semantic tag Frequency E4.1+

81

Relative frequency 0.23

E2

4

0.01

E4.1++

1

0.00

E2+

447

1.25

E4.1+++

1

0.00

E2++

3

0.01

E4.1-

96

0.27

E2+++

8

0.02

E4.2+

30

0.08

E2-

36

0.10

E4.2-

5

0.01

E3+

29

0.08

E5+

3

0.01

E3-

40

0.11

E5-

39

0.11

E6+

7

0.02

E6-

62

0.17

-227-

Table 3: Emotion tag E2 and its variants Tag

Word/phrase

putting_up_with tolerate put_up_with E2+ affection appreciated / appreciative / appreciation beloved caring darling dear enjoy / enjoyed / enjoying fancy fond / fonder / fondly going_for like / likes / liked live_with / liveing_with / lived_with love / loved / loves / loving / loveing precious went_in_for E2

Tag

Word/phrase

E2++

prefer / preferably i_’d_rather

E2+++ dearest favourite treasured animosity Ecan_not_stand disgust drudge fault hate / hatred / hates / hated n’t_for_me resented were_against

3. Semantic tagging results Shapero (2011) found that 10 of the Wmatrix semantic categories described the distinctive ‘aboutness’ of suicide notes in general. We now wished to see whether any of the WMatrix categories were useful for describing the ‘aboutness’ of multiple as compared with single notes. In the following discussion, we loosely refer to notes by authors who wrote more than one note, and sometimes to the authors themselves, as ‘Multiples’. Similarly, when referring to authors who wrote only one note we may use the term ‘Singles’ to describe them or their notes. Our 8 authors in the sub-corpus are labelled f1—f3 and m1—m5 to distinguish the 3 female and 5 male writers. First of all, we looked at all 223 ‘multiple’ notes. Table 4 shows the top 20 semantic categories in the Multiples when compared with the Singles (as a comparison corpus, see above). They are presented in Wmatrix’s default order which is descending Log Likelihood (the ‘LL’ column in the table), a measure of observed relative to expected frequency: for a detailed explanation, with statistical formulae, see Rayson et al. (2002). The definitions in the ‘category’ column are cited verbatim from Wmatrix, and are not always entirely transparent. For instance, ‘Generally kinds, groups, examples’ for A4.1 can be glossed as ‘General/abstract terms denoting types, groups, examples’. For a fuller description of the categories used, see Archer et al. (2002). It appears from this that writers of multiple notes are more likely than writers of single notes to address topics of money and pay, houses and buildings, and vehicles and transport. A good example of this male writer 1 (m1), as illustrated in Extract 1 (names have been changed). Extract 1

Half of the money in these properties (after adjustments about your own money which is in them - don’t forget the original £4000) will belong to Jane’s side.

-228-

Table 4: Multiples compared with singles: top semantic categories Tag I1.1 A4.1 H1 S8+ M3 G2.1 X7+ L3 I2.1 O4.3 K3 H5 E5B4 I2.2 T1.3 G3

M 50 57 67 145 83 68 258 19 19 18 17 40 37 14 51 212 13

M% 0.17 0.19 0.23 0.50 0.28 0.23 0.88 0.06 0.06 0.06 0.06 0.14 0.13 0.05 0.17 0.72 0.04

S 1 2 4 16 7 5 36 0 0 0 0 2 2 0 4 31 0

S% 0.02 0.03 0.06 0.25 0.11 0.08 0.55 0 0 0 0 0.03 0.03 0 0.06 0.47 0

LL 13.70 12.29 9.82 8.61 8.05 7.94 7.82 7.65 7.65 7.25 6.85 6.83 5.93 5.64 5.48 5.35 5.24

A5.2S5+ Q2.1

41 61 167

0.14 0.21 0.57

3 6 24

0.05 0.09 0.37

4.82 4.59 4.52

Category Money and pay Generally kinds, groups, examples Architecture, houses and buildings Helping Vehicles and transport on land Law and order Wanted Plants Business: Generally Colour and colour patterns Recorded sound Furniture and household fittings Fear/shock Cleaning and personal care Business: Selling Time: Period Warfare, defence and the army; weapons Evaluation: FALSE Belonging to a group Speech: Communicative

Table 5 shows the top 20 semantic categories across the 223 Multiple notes, in order of raw frequency. There were 338 semantic categories used in all by the Multiple authors. For the subcorpus of 80 texts used in this study, 285 categories in all were found. Here again, some of the category labels are opaque: Z5 ‘Grammatical bin’ is glossed by Archer et al. (2002: 36) as ‘prepositions/adverbs/conjunctions, etc.’ and Z4 ‘Discourse bin’ consists of ‘discourse markers, emphatic communication terms’ (ibid.). Table 5: Multiples: top semantic categories Tag

Raw Freq. Category

Tag

Raw Freq. Category

Z5

6834

Grammatical bin

A1.1.1

353

Z8 Z99 A3+ Z6 A7+

5367 1589 969 640 500

Pronouns Unmatched Existing Negative Likely

N5.1+ Z1 N1 A13.3 M1

349 341 324 313 299

A9+ Z4 M6 E2+

431 413 360 354

Getting and possession S4 Discourse bin T1.1.3 Location and direction X7+ Like Q2.2

294 275 258 236

General actions / making Entire; maximum Personal names Numbers Degree: Boosters Moving, coming and going Kin Time: Future Wanted Speech acts

Table 6 shows the total number of WMatrix categories used by each of the eight authors who wrote eight or more notes. (None of the figures is mutually exclusive.)

-229-

Table 6: Category types used by 8 authors Author f1 f2 f3 m1 m2 m3 m4 m5

Texts 11 16 9 8 9 10 8 9

Cat. Types 73 196 114 160 189 132 145 16

Table 7 clearly shows that our 8 ‘Multiple’ authors made very different use of their semantic resources, as reflected in widely varying numbers of WMatrix categories. (In this Table, as in the previous one, none of the figures is mutually exclusive.) Table 7: Total category types across notes (Letters A—P are arbitrary labels for individual notes) Wr.

A

B

C

D

E

F

G

H

I

J

K

f1

7

28

13

21

12

28

24

24

28

9

9

f2

41

107

109

12

89

43

31

36

54

26

35

f3

7

15

28

34

22

24

48

38

74

m1

78

73

71

76

57

5

13

17

m2

71

166

16

19

14

8

20

9

34

m3

25

40

36

37

41

5

54

55

43

m4

71

47

75

55

52

60

76

23

m5

1

5

3

3

2

3

6

2

L

M

N

O

P

54

27

35

15

14

61

2

However, this in itself does not tell us to what extent the authors are writing about different topics. For this, we need to examine the range of categories rather than simply the total numbers of categories. Table 8 shows those categories that were used 55 or more times when data from all 8 authors are combined. (Figures are raw frequencies; the Wmatrix Tag codes have been omitted for reasons of space. The figures are in descending order of row totals (not shown).)

-230-

Table 8: Tags with token frequency >= 55 across all authors

Grammatical bin Pronouns Unmatched Existing Negative Likely Getting and possession General actions / making Discourse Bin Location and direction Entire; maximum Time: Future Degree: Boosters Moving, coming and going Like Numbers Wanted Knowledgeable Thought, belief Speech acts Anatomy and physiology Time: Period Speech: Communicative Strong obligation/necessity Kin Objects generally Personal names Putting, pulling, pushing, transporting Quantities Exclusivizers/particularizers Religion and the supernatural

f1 81 93 42 26 15 10 5 9 6 2 1 10 3 0 3 1 10 2 4 3 0 0 0 2 0 6 4

f2 672 349 64 65 39 40 31 31 22 27 30 44 17 22 32 21 19 28 24 7 17 21 6 13 8 8 18

f3 214 217 33 31 19 16 16 16 14 10 17 5 11 12 15 5 5 2 3 8 5 9 5 5 15 1 3

m1 400 246 52 55 34 30 22 22 24 26 16 27 15 19 8 15 10 21 3 17 5 4 9 16 9 25 6

m2 596 414 79 60 22 34 28 21 26 17 13 13 14 28 14 22 8 7 19 30 24 22 26 15 13 9 27

m3 338 247 122 40 65 26 25 27 16 19 19 5 12 16 13 21 3 1 8 7 9 9 13 9 7 15 5

m4 323 359 93 47 77 45 32 23 23 24 20 8 38 12 19 3 31 21 19 7 15 8 14 13 16 3 2

m5 2 7 9 1 0 1 1 0 1 1 1 0 0 1 4 0 0 0 0 0 0 0 0 0 1 0 1

1 1 4 0

14 9 9 22

3 2 5 2

8 14 5 1

11 23 4 14

15 4 10 12

10 6 19 4

0 0 0 0

From Table 8 it is apparent that there is considerable variation between writers in the ‘aboutness’ of their notes. For instance, writers F2, M2 and M3 seem to have a good deal to say about religion and the supernatural, while writers F1, M1 and M5 have little or nothing to say on this subject. 4. Inter-author variation The categories which distinguish between writers appear to include those listed in Tables 9— 13 which show raw frequencies for various WMatrix categories across writers, and Figures 1—7 which show examples of concordances for particular writers illustrating some of their favoured themes. It should be borne in mind, of course, that the total token counts vary widely between authors; statistical tests would be needed to establish whether these apparent differences are actually significant.

-231-

Table 9: Between-author comparison figures for some of the ‘A’ categories Tag

Description

f1

f2

f3

m1

m2

m3

m4

m5

Total

A1.4

Chance, luck

0

0

0

1

7

2

4

0

14

A13.2

Degree: Maximizers

1

8

1

1

4

2

1

0

18

A2.1+

Change

1

6

0

8

8

0

5

0

28

A2.2

Cause&Effect/Connection

1

9

1

8

7

3

6

0

35

A3+

Existing

26

65

31

55

60

40

47

1

325

A4.1

Generally kinds, groups, examples

0

12

3

3

7

0

1

0

26

A5.1+

Evaluation: Good

8

9

2

3

6

0

13

0

41

A5.2

Evaluation: FALSE

0

2

16

0

1

6

0

0

25

A6.1-

Comparing: Different

0

23

0

4

5

3

1

0

36

A6.2+

Comparing: Usual

0

9

0

2

3

0

0

0

14

A7

Probability

0

7

0

2

2

0

0

0

11

1 2 3 4 5 6 7 8 9

ssible . [name] Aug 10th 1996 a know way . And if perhaps it lving destiny of the universe . he withdrawal of Strength , and Friends , I have written out my ct or love . If my death is the will please put a copy of these may be interested or may wonder re like as well as our rascally

Reasons leads to In regard to consequent reasons result reasons why relations

Figure 1: F2’s use of A2.2: cause & effect/connection

-232-

for Taking My Own Life I have ta an everlasting dreamless sleep Man , He has given us each a Min loss of independence through the for what I have done . I have wr of suicide , may I ask if you wi in their envelopes , and also gi I have taken this course . I lea . I cant help wondering if they

Table 10: Between-author comparison figures for some of the ‘B’ to ‘I’ categories

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Tag

Description

f1

f2

f3

m1

m2

m3

m4

m5

Total

B1

Anatomy and physiology

0

17

5

5

24

9

15

0

75

B3

Medicines and medical treatment

1

10

2

2

6

4

0

0

25

E2+

Like

3

32

15

8

14

13

19

4

108

E3-

Violent/Angry

0

4

0

0

8

2

0

0

14

E4.1-

Sad

10

9

4

1

1

0

9

0

34

H1

Architecture, houses and buildings

1

1

2

18

0

2

4

0

28

H4

Residence

0

7

1

11

3

3

3

0

28

H5

Furniture and household fittings

1

1

2

12

0

1

0

0

17

I1

Money generally

0

0

1

14

0

3

4

0

22

I2.2

Business: Selling

0

0

2

13

5

1

2

0

23

our help the last two years I have enjoyed , Sorry Sarah Please for give me . Dear s I have let you down badley but I love , he was my life and I through he love t said all I should to you but , I loved part . now he can Diana Roger Jill I love a liar I never told him to leave I loved I loved that liar cheat &; coward Dear but I am sorry . Tell the kids I loved they had to be in this Lesley I enjoyed sss I enjoyed our holiday together love ; ? ] Now you can do what you like You will be able to do what you like used to think how lucky I was you loved ie all our marred life . you never loved

Figure 2: F3’s use of E2+: Like

-233-

my holidays with you as well , Sorr James &; Lesley I have let you down you both , and am sorry , you do n’ me , but I can see he never did , I you both , do n’t grieve for me , I you all , and am so proud of you . that liar cheat &; coward Dear Jame James Lesley Sarah I am sorry but I them all very much and I am sorry , our holiday together love mom No 99 mom No 99 I could have come to term you will not have to wait till I go as you said you can , let your dogs me , and I throught we had a good l me I stood by you all the time ,

Table 11: Between-author comparison figures for some of the ‘K’ to ‘Q’ categories Tag

Description

f1

f2

f3

m1

m2

m3

m4

m5

K1

Entertainment generally

0

2

3

0

5

0

2

0

12

L1-

Dead

0

12

3

0

25

0

8

0

48

L1+

Alive

4

15

3

5

7

2

7

0

43

M1

Moving, coming and going

0

22

12

19

28

16

12

1

110

M2

Putting, pulling, pushing, transporting

1

14

3

8

11

15

10

0

62

M3

Vehicles and transport on land

1

3

4

6

12

3

3

0

32

M6

Location and direction

2

27

10

26

17

19

24

1

126

N1

Numbers

1

21

5

15

22

21

3

0

88

N4

Linear order

1

6

3

2

14

9

5

0

40

N5

Quantities

1

9

2

14

23

4

6

0

59

N5+

Quantities: many/much

4

19

1

9

9

2

9

0

53

N6+

Frequent

0

8

3

0

12

2

3

0

28

O1

Substances and materials generally

0

0

12

0

2

0

0

0

14

O1.1

Substances and materials: Solid

0

0

0

1

9

1

0

0

11

O2

Objects generally

6

8

1

25

9

15

3

0

67

O4.6-

Temperature: Cold

0

0

0

7

1

0

0

0

8

Q1.2

Paper documents and writing

0

10

0

10

6

1

3

0

30

Q2.1

Speech: Communicative

0

6

5

9

26

13

14

0

73

Q2.2

Speech acts

3

7

8

17

30

7

7

0

79

1 2 3 4 5 6 7 8 9 10 11 12

MOMENT THAT YOU WILL FIND RRRR LL FOUR OF US WILL BE TOGETHER agine and I have been with her it together and not have spent being turned from side to side en 5 STONE when she died . She ere on a Motorway for a period ery moderate traffic she would I can tell her when I see her ain . OMMISSION ON LAST PAGE X les . SO : PLEASE BE GOOD GUYS L DATE - HOPEFULLY BE TOGETHER

Total

AGAIN AND THAT ALL FOUR OF US WILL BE AGAIN IF YOU THINK THAT I AM JUST TEL every day for the past 18 months Since sh every minute of our lives enjoying each othe every two hours) I requested an interview ** often wished that she was dead , and of an hour with very moderate traffic she often complain of the fumes and ask m gain . OMMISSION ON LAST PAGE X Many Many times she has asked me why God would AGAIN AND HELP ME TO EXPEDITE MATTERS GAIN WITH GOD MY SON &; HIS WIFE GO

Figure 3: M2’s use of N6+: Frequent

-234-

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

talk to you more about the whole forgetting it happened and doing things for you and Jack to keep day know it was not a deliberate Charles in your own way . Terrible say that because it is the only deboard . Use my flat - you have ot be with you . I tried to keep k . Bunduki Safes near bed &; in tool cupboard General financial ancial things ( bank , ‘Access ‘ , and No. 5 also vacant . ( All going concern or not . The metal asket on the typewriter contains going on immediately , as do the wo small houses are also in this ng cabinet in the labelled ‘car’ ancelling ] The reason your doorbell does n’t work is that a wire is disconnected behind the tton at the front door , and the now . ( Expiry dates on freezer nnn - or anyone - Jack’s Flat 2 can talk to him by pressing the king up the phone - or here is a

thing things things thing things thing keys things tool things credit card keys basket things files basket file doorbell wire bell bell tems doorbell bell key

. It was like a black nightmare . for you and Jack to keep things g going , or . something I wanted t - if I had ever wanted to hurt so have happened , as you will soon I have ever had a real talent for - there is more food in the fridg normal for you as long as I could cupboard General financial things ( bank, ‘Access ‘ credit card ac account , etc.) are in top drawe in cabinet on wall near my filing on the typewriter contains things which are going on immediately , underneath it . The rent sheets f . ( with phone numbers of tenants . It belongs to us both equally . does n’t work is that a wire is d is disconnected behind the bell b button at the front door , and th is missing . ( I took it , rememb do n’t matter , now that they are does not work . You can talk to h push low down near this door , an to his flat . ----

Figure 4: M1’s use of O2: Objects generally

We may note in passing that WMatrix is not entirely accurate in its semantic tagging: some of the ‘things’ identified in Figure 4 above are metaphorical rather than physical, and so should not have been placed in the O2 category. Nonetheless the system is sufficiently robust to generate output which serves to identify the ‘aboutness’ of a text: most of the ‘things’ referred to by M1 are in fact physical objects such as keys, tools and credit cards.

-235-

Table 12: Between-author comparison figures for some of the ‘S’ to ‘T’ categories Tag

Description

f1

f2

f3

m1

m2

m3

m4

m5

Total

S1.2

Personality traits

0

10

0

0

1

0

1

0

12

S3.1

Personal relationship: General

5

16

3

1

3

0

5

0

33

S3.2

Relationship: Intimacy and sex

0

8

1

1

5

5

9

0

29

S4

Kin

0

8

15

9

13

7

16

1

69

S5-

Not part of a group

0

7

0

0

3

1

0

0

11

S5+

Belonging to a group

0

7

5

0

8

1

2

0

23

S7.1+

In power

0

1

0

12

3

1

1

0

18

S7.4+

Allowed

1

5

1

4

4

0

8

0

23

S8+

Helping

4

4

3

12

13

5

7

0

48

S9

Religion and the supernatural

0

22

2

1

14

12

4

0

55

T1

Time

2

6

8

2

6

4

19

0

47

T1.1.1

Time: Past

0

5

2

2

13

7

1

0

30

T1.1.2

Time: Present; simultaneous

0

3

5

3

10

11

14

0

46

T1.1.3

Time: Future

10

44

5

27

13

5

8

0

112

T1.3

Time: Period

0

21

9

4

22

9

8

0

73

T2-

Time: Ending

0

12

0

4

2

5

5

0

28

T2++

Time: Beginning

0

10

10

4

5

11

0

0

40

1 2 3 4 5 6 7 8 9

pportive and showed me so much t be sad youve give me so much uld have spent the rest of our ME PLESE AND BE HAPPY SO SORRY essage and when I said wher ‘s e them to me . I ‘ll allway ‘s or letting you all down All my all think when I told Julie to . I tried to hate her but the

love ca n’t take this feeling so lon love I just wa n’t to be happy and t lives together but I suppose it was n’t meant LOVE YOU So sorry Julie did n’t reali love you said you did n’t mean it li love you even after this I could nev love Steve Tim I wa n’t you to have fuck of that was the last time I spo love for her is so great . I know pe

Figure 5: M4’s use of S3.2: Relationship: intimacy & sex 1 2 3 4 5 6 7 8 9 10

ends . My only concern is that you will special which I hope neither of us will rget . My only concern is that you will focate it . Do n’t be sad Sam . I ‘ll ends . My only concern is that you will . Do n’t be sad Kim . At least I ‘ll s my friend . The prospects of the future Please do n’t be sad . At least I will de . Ellie. Graham , do n’t be sad . I’ll ou . Ellie. George , do n’t be sad . I ‘ll

Figure 6: F1’s use of T1.1.3: Time: future

-236-

blame yourself for not supporting m ever forget . My only concern is th blame yourself for not supporting m always be with you . Ellie. Kim , g blame yourself for not supporting m be at peace in what I have chosen t seem too daunting for me . Please d be at peace in what I have chosen t always be with you. Ellie. George, always be with you Ellie.

Table 13: Between-author comparison figures for some of the ‘X’ to ‘Z’ categories Tag

Description

f1

f2

f3

m1

m2

m3

m4

m5

Total

X2.1

Thought, belief

4

24

3

3

19

8

19

0

80

X2.2+

Knowledgeable

2

28

2

21

7

1

21

0

82

X3.4

Sensory: Sight

0

0

3

3

5

2

5

0

18

X4.1

Mental object: Conceptual object

0

8

0

3

5

1

7

0

24

X7+

Wanted

10

19

5

10

8

3

31

0

86

X8+

Trying hard

0

1

4

10

1

10

6

0

32

X9.2-

Failure

0

1

0

0

2

6

4

0

13

Z1

Personal names

4

18

3

6

27

5

2

1

66

Z2

Geographical names

1

6

1

3

4

3

8

0

26

Z3

Other proper names

0

2

3

0

8

5

3

0

21

1 2 3 4 5 6 8 9 10 11

Jim Thank you for coming round and e to do this one thing rite i have ns because thay are no good I have keep on lying all of the time and ke P.T , O Please ains of an eight year old and hard ay it is going at the mowment , in of view lost everything even if I le and i whont to put that rite by hic I am lerning But I have got to

trying tryed tryed trying TRY trying trying try trying try

to help me. but you have been waist to do it befor but faild but this t to use them but thay do nt work I m to hide that I am all rite HA HA HA and understand and can you Do one l to put things in order which one to to tell the truth about my failings to say that you have maid all of th hard to think stright First of all you say to me that you are not a vi

Figure 7: M3’s use of X8+: Trying hard

5. Variation between notes As well as the variation between authors noted above, there is also considerable variation within authors. It might be assumed that when a suicide leaves multiple notes, they are merely variants of the same text, intended for different recipients. Our research, however, reveals that this practice is not particularly common. Figures 8 and 9, for instance, are notes by the same woman. While note A is addressed ‘To Whom it may Concern’ and states her desire to be left to die, note G is addressed to a named individual and expresses her personal feelings for her friends. Not only are the notes intended for different readers, but their purposes are entirely different.

-237-

To Whom it may Concern . P.T.O Voluntary Euthanasia -----------------------------Do not resuscitate me or treat me in any way except for pain killers if necessary. Do not remove my body from my home before my death. Aug 10th 1996 Dr Jones Tel 999 9999 A.B. Smith P.T.O Nov 10th 1996 A.B.S. If anyone, doctor, neighbour, well-meaning friend, police etc, should treat or resuscitate me without my clearly stated wish, I understand I have the right to prosicute for assault, and will not hesitate to do so A.B.S. ---------------------------------------------There is no need to make any arrangement to “look after” or “care” for me If I am helpless, please put me on the bed and leave me to die. I shall take nothing to eat or drink so that I shall die as soon as possible. A.B.Smith Aug 10th 1996 Figure 8: F2, note A: full text

My dear Kim, When you get this letter I shall no longer be here. May I thank you most sincerely for your friendship and all the many things you have done for me over the years for which I thank you. I am indeed truly grateful. I sincerely hope the future will hold better times for you, and much happiness for Chris and Jane and their families. All my kindest thoughts to you and to your loved ones. Annie X X X xx Figure 9: F2, note G: full text

Table 14 shows the frequencies for various WMatrix categories across all the texts of writer F2. Her notes B and C are of similar token lengths (431 and 422 running words, respectively), and so it is not unreasonable to compare frequencies of WMatrix categories between them. There are several quite striking differences here, including the figures for Z6, ‘negative’, of which there are almost twice as many in note C as in note B.

-238-

Table 14: Within-author variation: F2 A

B

C

E

F

I

L

Totals

Z5

Grammatical bin

25

131

148

100

25

53

36

672

Z8

Pronouns

18

47

42

52

23

37

29

349

A3+

Existing

2

19

10

7

5

5

3

65

Z99

Unmatched

7

2

6

5

3

7

2

64

T1.1.3

Time: Future

4

4

0

6

5

6

4

44

A7+

Likely

2

8

9

9

0

3

3

40

Z6

Negative

5

9

17

5

0

0

0

39

E2+

Like

0

1

2

3

4

4

2

32

A1.1.1

General actions / making

2

9

6

5

0

2

1

31

A9+

Getting and possession

2

5

4

6

3

3

0

31

N5.1+

Entire; maximum

2

1

4

4

2

5

3

30

X2.2+

Knowledgeable

0

4

1

3

4

4

2

28

M6

Location and direction

0

4

3

4

1

2

4

27

X2.1

Thought, belief

0

4

8

7

0

0

1

24

A6.1-

Comparing: Different

0

7

7

6

0

1

2

23

M1

Moving, coming and going

1

7

3

4

1

2

1

22

S9

Religion and the supernatural

0

3

10

4

0

1

0

22

Z4

Discourse Bin

2

3

2

3

2

1

2

22

N1

Numbers

5

0

4

0

0

0

2

21

T1.3

Time: Period

3

2

4

2

2

1

2

21

124

431

442

352

115

184

148

We can examine the texts which have generated these discrepant frequencies by producing concordances of tag Z6 for both notes, as shown in Figures 10 and 11.

-239-

1 2 3 4 5 6 7 8 9

deteriorating, and it will natural strength, but as that is at time is reached, in case I can ot do it without assistance. I am .e. senilely insane, so that I do t know what I am doing, and would would not do it if I did realise. ythm of Nature. Life is eternal; me believe they know, but that is

not not not not n’t not Nor nothing not

be very long before it is insuffici possible , I must go before that ti do it without assistance . I am not " confused " i.e. senilely insane , know what I am doing , and would no do it if I did realise . Nor am I u am I unhappy . On the contrary I am in Creation can cease to exist . It the same . But does it really matte

not not nor nor no nor no not not not not nor no nor not nor not

be able to live alone with a normal want to be resuscitated , nor , wi , with the exception of pain killer do I want anything to eat or drink relatives nor close friends to whom close friends to whom my death wou torture could be greater than livin even in my own home with constant v lingering like a rotten fruit , ove a christian , believing human life believe it is the law of Nature , n of God , that when my life can have more meaning nor useful purpose , i useful purpose , its basic physical acting in accordance with the Laws of the Hypocratic Oath ( whose mod that given to it by the Ancient Gre

Figure 10: F2, note B: negatives (Z6) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

even the remote possibility I may h a normal independent life , I do I do not want to be resuscitated , I want any treatment whatsoever , e as quickly as possible . I have as possible . I have no relatives ke the slightest difference , and asting company in an institution , depart in the rhythm of nature , , over-ripening on the tree . I am we have been destined to do . I do believe it is the law of Nature , God , that when my life can have n my life can have no more meaning the individual or loved ones , are accordance with the Laws of God , h ( whose modern interpretation is

Figure 11: F2, note C: negatives (Z6)

Text B is in fact headed ‘Reasons for Taking My Own Life’ and is somewhat philosophical in nature, whereas text C is headed ‘To Whom it May Concern’ and focuses on the writer’s desire not to be resuscitated. It is her list of things which she does not wish to be done which gives rise to the higher frequency of negatives in this latter note.

-240-

Table 15: Within-author variation: M4 A

B

C

D

E

F

G

Totals

Z8

Pronouns

59

36

62

46

30

52

61

359

Z5

Grammatical bin

42

27

62

42

29

40

68

323

Z99

Unmatched

15

5

15

9

10

12

26

93

Z6

Negative

15

6

11

10

13

7

14

77

A3+

Existing

4

4

12

4

8

6

9

47

A7+

Likely

7

6

6

9

7

2

6

45

A13.3

Degree: Boosters

4

4

10

3

9

4

4

38

A9+

Getting and possession

4

2

6

3

5

3

8

32

X7+

Wanted

3

7

5

5

4

0

7

31

M6

Location and direction

1

5

6

1

1

3

7

24

A1.1.1

General actions / making

1

1

5

1

4

3

7

23

Z4

Discourse Bin

6

3

4

2

3

1

4

23

X2.2+

Knowledgeable

7

0

1

4

0

5

4

21

N5.1+

Entire; maximum

1

1

8

0

5

3

2

20

A14

Exclusivizers/particularizers

4

1

3

4

2

1

4

19

E2+

Like

3

1

4

3

4

2

1

19

T1

Time

2

3

3

2

0

2

6

19

X2.1

Thought, belief

5

2

6

0

2

2

2

19

S4

Kin

2

0

4

0

2

2

3

16

262

155

317

199

189

209

329

Writer M4’s notes can be described in a similar way. Here, notes D, E and F are similar in word length, so comparisons are valid. Let us examine the details for the occurrence of tag A7+, ‘likely’, in note D, which contains 9 occurrences, and in note F which contains only 2 despite being the longest of the three notes in question. Figures 12 and 14 show the concordance lines, while Figures 13 and 15 show the corresponding full texts.

-241-

1 2 3 4 5 6 8 9 10

ry did n’t realise I got so boring stand why.Loved you so much . How just switch me off . If only you uld of given me a chance I know I used I ‘ll never know now wish you ed to me.I needed answers and you way ‘s love you even after this I s been like the last two month ‘s sleep feel like Ive let you down

could could would could could could could ca ca

n’t understand why.Loved you so muc you just switch me off. If only you of given me a chance I know I could of maid you happy just got side tr have talked to me.I needed answers n’t give them to me . I ‘ll allway never hate you . You should have to n’t sleep feel like Ive let you do n’t take it Do n’t Forget I ‘ll all

Figure 12: M4, note D: Likely (A7+)

So sorry Mary didn’t realise I got so boring couldn’t understand why.Loved you so much. How could you just switch me off. If only you would of given me a chance I know I could of maid you happy just got side tracked with trying to make the house more comftable for us. tried my best. should of known you stopped loving me when I had sun stroke and left me the message and when I said wher’s love you said you didn’t mean it like that how long was you planning this did you ever love me or was I just there to be used I’ll never know now wish you could have talked to me.I needed answers and you couldn’t give them to me. I’ll allway’s love you even after this I could never hate you. You should have told me Mary [?] Don’t want to die but I don’t want to be without You . You don’t know what its been like the last two month’s can’t sleep feel like Ive let you down can’t take it Don’t Forget I’ll allway love you Hope you fined happiness Bye Joe Figure 13: M4, note D in full 1 2

st got nothing left inside me . I seemed like she had never knew me

would ca

love to hold my head up and walk t n’t get my head round all of this

Figure 14: M4, note F: Likely (A7+)

Steve I wan’t you to have my guitars at least I know you will look after them . Bin good friends you and me . Do me a favor youre really good with words Shouldn’t really ask you to do this I know its a bit of a cheeck help my mom & Dad through this Ive just got nothing left inside me. I would love to hold my head up and walk through this . People don’t understand how much this has hurt me should have seen what was happening. I know I need glasses but I didn’t think I was this blind . You all think when I told Mary to fuck of that was the last time I spoke to her it wasn’t I sent her a tape and phoned her she seemed like she had never knew me can’t get my head round all of this . I tried to hate her but the love for her is so great. I know people say times a great healler but I dont wan’t any one else to weak to win her back . Ive lost it Steve its allmoste as if she has died. thought I was so lucky. Never forget you. I love you Steve youve been a good friend Bye Joe Figure 15: M4, note F in full

An inspection of the actual notes which yielded these figures reveals that WMatrix has done quite a good job of identifying the ‘aboutness’ of these texts. While Note D is full of wistful reflections on what might have happened in the writer’s relationship with the female recipient, -242-

note F is more practically oriented, and like many suicide notes leaves instructions to the reader who is described as ‘a good friend’. The computer-generated count of tags classified as A7+, ‘likely’, has picked up the heavy use of modality in the former of these two texts. In the examples above we have discussed the relative frequencies of a single semantic tag in texts by the same author. In future research it might prove fruitful to examine combinations of tags as the key to identifying sub-genres within the somewhat crude category of ‘suicide note’. We might find, for instance, that a low frequency of ‘negative’ and ‘modality’ tags combined with a high frequency of ‘objects’ is indicative of an ‘instructions’ sub-genre. In order to investigate this possibility it will probably be necessary to merge sets of Wmatrix categories into superordinate groupings in order to enable tests for statistical significance to be conducted with meaningful results. 6. Conclusion A corpus of genuine suicide notes in computer-readable form is a valuable resource: in particular, a collection of multiple suicide notes by the same writer is a fruitful source of data for the forensic linguist, and for linguists in general who are interested in issues of genre and authorship. This study has shown that suicide notes exhibit considerable variation both between and within authors. Some of this variation appears to be attributable to the ‘aboutness’ of the texts rather than to individual stylistic peculiarities of the author. The semantic tagging system provided with the WMatrix software has proved invaluable in identifying ‘aboutness’ in the texts of our corpus and in flagging up particular areas of semantics which are likely to reward further research. The surprisingly high levels of intra-author variation which we discovered in these undisputed notes indicates that it would be most unwise to rely on such variation alone as a diagnostic criterion of divergent authorship. However, it may be that collocations of particular semantic tags can be found to indicate sub-genres of suicide note. If we can exclude this ‘aboutness-related’ variation from the sum total of the differences between two texts, any remaining semantic variation may yet be indicative of different writers. Until further light has been shed on the predictability of the kinds of patterns described here, forensic linguists would be well advised to rely instead on syntactic and lexical patterning when seeking to make judgements concerning likely authorship. References Archer, D. Wilson, A. & Rayson, P. (2002) Introduction to the USAS Category System. Computing Department, Lancaster University. http://ucrel.lancs.ac.uk/usas/usas%20guide.pdf Fairthorne, R. A. (1969) Content analysis, specification and control. Annual Review of Information Science and Technology 4: 73—109. Goutsos, D. (1995) Review article: forensic stylistics. Forensic Linguistics: The International Journal of Speech, Language and the Law 2(1): 99—113. Phillips, M. (1989) Lexical Structure of Text. Discourse Analysis Monograph No.12, English Language Research. Birmingham U.K.: The University of Birmingham.

-243-

Rayson, P. (2007) Wmatrix: A Web-based Corpus Processing Environment. Computing Department, Lancaster University. http://www.comp.lancs.ac.uk/~paul/publications/icame01.pdf Rayson, P., Wilson, A., and Leech, G. (2002) Grammatical word class variation within the British National Corpus Sampler. In P. Peters, P. Collins and A. Smith (eds) New Frontiers of Corpus Research: Papers from the Twenty-first International Conference on English Language Research on Computerized Corpora, Sydney 2000. Amsterdam: Rodopi, 295—306. Rayson, P., Archer, D., Piao, S. L., and McEnery, T. (2004) The UCREL semantic analysis system. Proceedings of the Workshop on Beyond Named Entity Recognition Semantic Labelling for NLP Tasks in Association with 4th International Conference on Language Resources and Evaluation (LREC 2004), 25th May 2004, Lisbon, Portugal: 7—12. Scott, M. (2007) Wordsmith Tools Manual. Version 4.0. Oxford: Oxford University Press. Shapero, J.J. (2011) The Language of Suicide Notes. Unpublished Ph.D. thesis, University of Birmingham. Shneidman, E.S. & Farberow, N.L. (1957) Genuine and simulated suicide notes. In E.S. Shneidman and N.L. Farberow (eds) Clues to Suicide. New York: McGraw-Hill Book Company. Van Dijk, T.A. (1977) Text and Context: Explorations in the Semantics and Pragmatics of Discourse. London: Longman.

-244-

‘Did he have an accent?’ Forensic speaker descriptions of unknown voices Mark Griffiths Trinity College London, UK [email protected] Abstract This paper addresses the issue of forensic speaker description: cases where the voice of the suspect may form a part of a police investigation, but where no voice recording exists. There is currently no established reliable procedure for eliciting earwitness descriptions of voices. When dealing with earwitnesses, evidence gathering is restricted to the police asking for descriptions of the voice(s) encountered, which are then written down. There is no synthetic reconstruction of the voice. Much doubt has existed over non-linguists’ abilities to provide voice descriptions, yet every day non-linguist police officers engage non-linguist earwitnesses in the process of eliciting linguistic data. This paper offers evidence that whilst non-linguist ideologies of language are often at odds with those of the linguist, there are nevertheless consistencies and patterns in the non-linguist’s perceptions and descriptions of accents and voices. A case is made that it is possible, with the right framework, to elicit data that can be meaningful and make a contribution to the investigative process, and that there is the potential to develop an audiofit. Keywords:

FORENSIC SPEAKER DESCRIPTION; AUDIOFIT; NONLINGUISTS; ACCENT; VOICE; POLICE

Context The aim of this paper is to focus on an area of research that has yet to be identified as a single body of study within applied or forensic linguistics, which I entitle forensic speaker description (FSD): the elicitation for forensic purposes of descriptions of known or unknown voices by members of the public by police officers. Nolan and Grabe (1996) paint an illustrative picture of the type of linguistic evidence where FSD may be relevant. Criminals, contrary to the ideal for children in Victorian England, are often heard but not seen. Victims may not have seen attackers, but may have heard them speak; witnesses may have seen little of masked armed robbers, but have overheard their interactions; and recipients of obscene telephone calls have been exposed only to the voice of the perpetrator. (Nolan and Grabe, 1996:74) This is distinct from the far more widely-researched forensic speaker identification (FSI) in both the source of the evidence—recorded (FSI) vs. unrecorded (FSD)—and the personnel handling the evidence—recordings are passed on to professional linguists for analysis (FSI) vs. recordings are elicited and processed primarily by police officers (FSD). Employing forensic speaker description as a method for handling voice evidence is always likely to present problems not encountered in other lines of investigation. A simple table comparison of FSD with other forms of audiovisual evidence—visual identification via closed circuit television (CCTV), eyewitness description (EYW) and forensic speaker identification (FSI)—illustrates some of the major difficulties facing forensic speaker description work:

-245-

Table 1:

Comparison of visual and linguistic evidence sources

CCTV

EYW

FSI

FSD

A recorded sample of evidence is available









The evidence has a technical source









The evidence can be technically assisted/enhanced









As Table 1 illustrates, FSD is what we might call the most impoverished source of evidence. FSD is what researchers and legal professionals are likely to be reduced to when no video or audio recording exist and there is no data from an electronic source that can be technically analysed or enhanced. In these scenarios, when there is simply no recorded voice evidence available, it is likely that the only distinctive evidential feature of the perpetrator—the voice—may be manifest solely in the memory of the earwitness, that is, the direct victims of or audially present bystanders to a crime. The problem of accessing descriptions from earwitnesses is further exacerbated by the fact that few will have had any linguistic training. There is a weak diffusion of linguistic expertise (referring to the analysis of language rather than the ability to speak other languages) throughout a given population, and professional linguists are few and far between. Linguistics is not frequently studied to any depth by students in schools or colleges, and the proportion of the adult general public in the UK that has undergone any linguistic training is extremely small, being as it is, a relatively unknown field. The lack of awareness or experience of linguistic issues is of course not uniformly low. Whilst there will be a few people with a general interest in language at least at an anecdotal level, it is clear that most of the general public are ‘non-linguists’. Ergo, most earwitnesses are likely to be members of the general public untrained in the science of linguistics, people who, according to Shuy, do not ‘have the metalanguage to describe his/her own language.’ (Shuy, 1993:14). Interestingly, the term ‘general public’ is used in the police context to contrast those who work in law enforcement with those who do not. From the linguist’s perspective, the contrast here is between those with formal linguistic training and those without, regardless of a person’s relationship to the criminal justice system. It would seem clear that despite being tasked with eliciting evidence and solving crimes, the training received by police officers offers no linguistic advantage. In preparation for this paper, interviews were conducted with police officers (a Police Constable and a Detective Inspector) from South Wales Police. The officers identified that aside from being given training in general questioning techniques such as the cognitive interview (Fisher & Geiselman, 1992), few if any police officers have any specific linguistic training, and receive no training courses or guidelines on the elicitation or provision of voice descriptions aside from the general guidelines for taking and providing statements. Given that the police offers assumed this would be the pattern countrywide, the UK police are, from the linguist’s point of view, as likely to be members of the non-linguist general public as the suspects, victims and witnesses they deal with. This assumption can also be extended to those staffing the courts. Philippon et al (2007) also observes that police officers are no more efficient at dealing with, organising and collecting linguistic evidence than ordinary members of the public, indicating that even work experience does not make up for a lack of specific linguistic training.

-246-

The resultant situation is that suspects, victims, witnesses, arresting officers and legal professionals are therefore almost exclusively non-linguists, accessing an identical linguistic resource for the provision, elicitation and interpretation of voice descriptions. Moreover, there may be little difference in terms of technical linguistic sophistication between a description of a voice in a pub, a police station or a law court, being that all the venues are populated almost exclusively by non-linguists. We can easily surmise that every hour of every day, up and down the country and around the world, non-linguist members of the general public are appointed to elicit the best possible linguistic evidence, from other nonlinguist members of the general public, which other non-linguists then represent in law courts. Yet so long as there are earwitnesses to crimes that have no aural recordings, eliciting forensic speaker descriptions will continue to be a necessary part of the legal investigative process. Research aims Given the picture painted above, the aims of the research described in this paper are to address five research questions: 1. Do non-linguists have the ability to perceive regional and social accents? 2. Is accent perception a function of respondent origin? 3. Is there any other way of eliciting accent perceptions from non-linguists? 4. What do non-linguists notice in voices? 5. Is there the potential to build an audiofit? Method The research was designed around eliciting respondents’ impressions and descriptions of unknown voices. A direct approach was employed, whereby recordings of real voices were played to respondents. The respondents were told that the voices they heard were recordings of interviews that took place with speakers who were being held in custody in relation to the same crime. Voice donors were recruited to play the role of the interviewees recorded in the police station. As one of the first studies of its kind, we have little guidance in the literature as to the ideal number of voice tokens to present in forensic speaker description research. Good practice used in forensic voice identification line-ups suggests a maximum of nine voices, as recommended by the McFarlane guidelines (Nolan, 2003). However, the nine-voice recommendation pertains to the field of forensic speaker identification, where the respondent is passive and has only to recognise/identify the speaker. Providing forensic speaker descriptions is active by nature, with the earwitness providing the evidence in their descriptions, rather than responding to recorded or live speakers as one might expect in, for example, a voice identification line-up. With this in mind, the potential for listener fatigue/cognitive overload must be acknowledged. It was therefore decided to reduce the number of voice samples in order to reduce the cognitive load on the respondents, and work with a bank of five exemplar voices. It was essential that the recordings of the voices did not rouse any suspicions in the respondents in terms of them not being ‘authentic’ police recordings. To this end, three considerations were designed into the voice recording process. First is the consideration of community authenticity—the voice donors need to sound like native speakers of their presented accents but should not sound like emblematic stereotypes of the accent; second is the consideration of style authenticity—if the respondents are to believe that the recordings

-247-

are taken from police interviews as planned, it is crucial that speakers’ statements are spontaneous and unrehearsed, rather than read from a printed text. If respondents suspect at any point that the speakers have donated their voices or have falsified their accents, or have been coached for use in a social experiment, the validity of the research would be compromised. These considerations underlie the selection of the speakers and the content and delivery of the voice samples set out below. Background to the voice donors To ensure style authenticity, this research observed Broeders’ (1996:9) emphatic stance not to use actors or ‘professional speakers’ in forensic speaker research. Broeders argues that linguists and phoneticians will be fully aware that features of the speech of actors and other professional speakers can be manifestly different from the speech of other language users. Only ordinary members of the public were approached to be voice donors and all voice donors were recruited on the basis of their accents. There is a solid body of research that suggests that accent and voice features of a speaker could be largely responsible for the varied impressions that non-linguists may have of the same voice (e.g. Dixon and Mahoney, 2004; Seggie, 1983; van Bezooijen and Ytsma, 1999) and in experimenting with possible design of an audiofit we require stimulus material that will be meaningfully different for the non-linguists. The first three voice donors were selected on the basis of their native regional accents, which were likely to be salient to the respondents. They were not, however, assumed to be absolute tokens of the dialect of their origin, and this was specifically left to the respondents to evaluate. There needed to be a strong likelihood of the voice donors’ accents being recognisably and authentically associated with five regional or social groups found in the south of the UK. Those groups were: the regional accents of South Wales, South West of England and South East England (often associated with Estuary English); the accent of public-school educated Received Pronunciation (RP)—like speakers—an accent that has been transplanted to private schools around the world, and is more of a social than a regional accent; and an accent which is commonly found in the south of the UK but which is neither particular marked for regionality or social background. This accent might, in non-linguists’ terms, be called ‘neutral’. A map indicating the geographical provenance of the voice donors is given in Figure 1. Speaker 1 26 year-old male from Cardiff, Wales. Cardiff is the largest city in Wales with its own recognised distinct accent. More recently, the accent has been popularised in the television sitcom ‘Gavin and Stacey’. In preparatory interviews for this paper, the speaker exhibited a great many features associated with the Cardiff urban accent. Speaker 2 A 27 year-old male from Bristol, South West England. Bristol is the principle city of South West England and is known for a distinct regional accent. A stereotypical urban Bristol accent has also been popularised in the television sketch comedy ‘Little Britain’ as the accent of the character Vickie Pollard. In preparatory interviews, the speaker showed many of the stereotypical features associated with this accent.

-248-

Speaker 3 A 30 year-old male from Hastings, South East England, some 60 miles south of London. Whilst not being a particularly large urban settlement (and as such unlikely to have a distinct accent recognisable to the wider country), the town lies in the dialect area heavily influenced by the spread of Estuary English and a linguist would expect respondents to make a connection between an urban speaker from Hastings and the wider London-influenced speech. On this basis and by listening to examples of the speaker’s speech, it was felt that there was a strong likelihood of the speaker’s accent being associated with the South East of England. Speaker 4 A 21 year-old male from Eastbourne, 15 miles from Hastings in the South East of England. Speaker 4 was educated in a private school and speaks with an accent that he himself identified as ‘public school’. A linguist might label the accent used by Speaker 4 as being general RP, the form of RP used most commonly in the UK, rather than conservative or advanced forms of RP (Gimson 1980:91). It was felt that any potential differences in respondents’ identification of the provenance of Speakers 3 and 4 are likely to lead to an interesting insight into the nonlinguist’s typology of southern UK speech. Speaker 5 A 28 year-old male from Birmingham in the Midlands of England. Common to many studies is the dichotomous juxtapositioning of regional vs. ‘high status’ accents, and a great many studies have investigated the contrast between the impression created by ‘local’ or ‘posh’. Speaker 5’s current accent has been influenced by social, professional and regional life choices that he has made. The speaker reports that he left Birmingham at 18 to attend university and due to his work as a lawyer, he has made a conscious effort to ‘lose’ the much socially stigmatised Birmingham accent in order to attract less derision in a profession that is famed for having a dominance of Oxford/Cambridge graduates with middle class, RP accents (Darbyshire 2005). Speaker 5 was therefore included for his potential to appear ‘neutral’ in accent.

-249-

Speaker 5: Birmingham London

Speaker 1: Cardiff

Speaker 3: Hastings Speaker 2: Bristol

Speaker 4: Eastbourne

Figure 1: Geographical provenance of voice donors

The content of the voice donors’ speech A verbal guise technique (Gallois and Callan, 1981) was employed, whereby the five recruited voice donors were asked to produce carefully-controlled, content-neutral texts: all speakers were required to discuss the same series of events as if they had been co-witnesses. Each speaker gave an account of a series of events on a ‘night out’, visiting bars and socialising in a group, telling the same set of facts, produced as responses to questions (the questions were edited out in order to focus attention only on the exemplar voices.) Respondents were presented with the five recordings, in any one of five randomised sequences to offset any order effects and asked to record their responses on a questionnaire. The participants consisted of two cohorts: Group 1 n=144 (116 female/24 male/6 no information given) and Group 2 n=122 (91 female/31 male). Following evidence from Braun (1996) and Hollien and Schwartz (2000), suggesting no difference in respondent performance in speaker identification tasks as a function of respondent sex, no distinction was drawn on the basis of sex. All respondents were undergraduate students, living in South Wales, between the ages of 18 and 24. In Group 1 (G1), 98.6% of the participants were aged between 18 and 21; in Group 2 (G2) the figure was 95.5%. Findings from Braun (1996) and Bull and Clifford (1984) suggest that for speaker identification research, respondents over 16 and under 40 may yield the most reliable results. Given the lack of research in the field of forensic speaker description, this guidance was applied to the current research, and all participants fell within an age range of predicted competence for the tasks. Respondents were not formally selected but were a convenience sample, based on availability and willingness to participate. Whilst G1 and G2 were exposed to the same voices under the same conditions, they were given different questionnaires. The differences can be summarised in the four points below: -250-

1. G1 was asked to provide labels for the accents, but in open questions; G2 was provided with a map and a tabulated list of the regions and urban centres as a prompt and reference tool, a process based on Preston (1996) and Niedzielski and Preston (1999) perceptual mapping tasks. 2. G2 was also asked to rate the speakers’ ‘strength of accent’ on a Likert-type scale, a composite devised to enable respondents to plot their intuitions regarding regionality, ‘accentless’ speech and ‘well-spoken’ or RP-like speech. The scale inspired by van Bezooijen and van Hout’s (1985) ‘degree of accentedness scale’, was based on the hypothesis that in the minds of non-linguist respondents in the UK, regional accent and RP are mutually exclusive, but are both separated by the nonlinguist concept of ‘no accent’. On the scale itself, ‘1’=‘regional accent’ and ‘7’=‘well spoken’, and the scale was intersected at ‘4’ by ‘no accent’. Point 4 on the scale was not left to imply absence of regional or well-spoken accent, or no strong opinion, but was denoted as specifically ‘no accent’. The scale therefore was an attempt to address three concepts in one. 3. G1 was asked to provide free qualitative descriptions of voice characteristics for each speaker. In total, the G1 non-linguists provided a corpus of 1068 voice descriptions, in the form of single word comments or phrases. The contents of these descriptions were analysed for themes, and from this analysis emerged a distinct set of qualitative subcategories invoked by non-linguists to articulate their perceptions of the five voices encountered. The qualitative subcategories that emerged from the non-linguist data are remarkably similar to those described by Hollien (1990) when performing forensic speaker analysis. The categories that overlapped with those of the professional forensic linguist were transformed into bi-polar semantic differential scales and presented to G2 to be used for articulating respondents’ impressions of voice qualities. The features were: speaker speed (fast/slow), speaker hesitancy/confidence, speaker pitch (high/low), speaker monotone/melodious, speaker loudness and speaker steadiness (steady/shaky). It is the results of G2’s evaluations that are reported in this paper. 4. G2 was also asked to consider an alternative way of giving voice and accent information using a technique that centres on an alternative form of cognitive elaboration: soundalikes. The respondents in G2 were invited to identify anyone known personally to them or famous, that the speakers they heard may sound like. The respondents were re-assured that they did not have to provide a name if nothing came to mind, and accordingly, certain voices did not attract many comments. Results The first of the research questions centres on non-linguists’ ability to perceive differences in accent. Respondents in G1 were asked to provide possible estimations of where they thought the speakers could be from, with no maps or other visual cues provided. G2 was provided with a map of the UK, demarcated and labelled into regions, plus a separate tabulated list of the UK regions with each region’s largest urban centres listed. Table 2 displays the results of the labelling task.

-251-

Table 2 Nonlinguists’ labelling of speaker accents - % ‘recognition’ rates

G1

G2

Speaker 1 labelled as ‘Wales’

66%

71%

Speaker 1 labelled as ‘the city of Cardiff’

36%

39%

Speaker 2 labelled as ‘South West’

67%

70%

Speaker 2 labelled as ‘the city of Bristol’

20%

43%

Speaker 3 labelled as ‘South East’

72%

72%

Speaker 3 labelled as the ‘city of London’

62%

52%

Speaker 4 labelled as South/South East/London

49%

20%

Speaker 4 labelled as London

13%

7%

0%

44%

Speaker 4 ‘don’t know’

44%

24%

Speaker 5 labelled as South/South East/London

34%

14%

Speaker 5 labelled as the Midlands

14%

18%

0%

24%

44%

21%

Speaker 4 labelled as Oxford/Cambridge

Speaker 5 labelled as Oxford/Cambridge Speaker 5 ‘don’t know’

Taking first Speakers 1-3, we see that at least 2/3 of the respondents in both groups identify the speakers as having accents that match their native region—this, despite the two groups having completed alternative questionnaires, and G1 not being given any maps or tabulated lists of regions or cities. At the regional level, the results remain (in percentage terms) very similar across the two groups. (N.B. due to the nature of the experiment, it is not possible to run independent or repeated measure statistical treatments, and for now we must satisfy ourselves with percentages.) This would suggest that rather than being bereft of intuitions regarding the origins of a speaker accent, the non-linguists bring with them a relatively strong areal taxonomy of accent variation in the regions under investigation. However, there is more to this picture—Speaker 3 is estimated as coming from the city of London, despite being from the south coast of England, some 60 miles away. We can only speculate at this stage as to what might direct people’s focus to the largest conurbation in the region: it may be that non-linguists from outside any one particular region are not sensitised to the range of intraregion accent variation. Alternatively, it may be that if pressed for the name of a town/city, non-linguists will simply identify the largest, best-known urban centre in any particular region. Certainly, this could form a revealing line of enquiry in future research designs. Regardless of the possible explanations, however, it must be noted that the non-linguists could have suggested any urban centre in the UK, but did not, and the cities that were selected were remarkably accurate, given the range of urban centres within the country. For Speaker 4, we see quite a different picture. The data appear to suggest that the regional labels provided by the respondents to describe social accents such as that of Speaker 4 are a function of the question and questionnaire. G1, which was provided only with open questions, estimated Speaker 4 to be from somewhere in the South—a potentially very large region encompassing both the South West and South East—or London. The use of London as a label for the accent also exemplifies the semantic duality of labels like London, where the name can connote a traditional regional accent as seen for Speaker 3, or an accent associated

-252-

with social/political/educational prestige such as RP-like accent of Speaker 4, as both accents are apparently seen to occupy the same urban space. However, it is the data from G2, who were provided with a guided perceptual mapping task, that are perhaps most intriguing. We see that in this task when the cities of Oxford and Cambridge are provided on the questionnaire in map and tabulated form, nearly 50% of respondents choose these cities as the label they associate with the accent, markedly different from the South/London labels from the respondents in G1. We can certainly speculate that what we are seeing here is the Oxford/Cambridge labels serving as a convenient shorthand for non-linguists to articulate the accent’s RP-ness—that RP has a particular sociourban associations in the minds of non-linguists, even if these associations are not based on linguist fact. If this is the case, this is of value to researchers and legal professionals, as the use of the Oxford/Cambridge labels as a repository for accent associations also serves as a strategy for disambiguating non-linguist responses. The data for Speaker 5 in some ways shows similar tendencies as the data for Speaker 4, in that we see that the Oxford/Cambridge label has been utilised by a quarter of respondents from G2 in contrast to none from G1. We can also see that where guided perceptual mapping has been utilised for G2, there are 50% fewer ‘don’t know’ responses. But what is also clear is that the non-linguists do not agree on a label which articulates their associations with a large spread of associations. What is needed is to investigate an alternative paradigm for articulating accent associations. To explore an alternative paradigm, the non-linguists in G2 were also presented with an instrument, upon which they were invited to plot their intuitions regarding the strength of the speakers’ accents.

7

Mean score

6 5 4 3 2 1 0 Accent Strength

Speaker 1 2.55

Speaker 2 2.81

Speaker 3 2.71

Speaker 4 5.26

Speaker 5 4.31

Figure 2 Means for ‘strength of accent’

The results of an ANOVA on the differences between the mean ratings for these speakers show that in all cases, the differences were significantly different ( < 0.05), with the exception of between Speakers 1-3 and 2-3, where no significant differences were found. The ‘accent strength’ attributions for Speakers 1, 2 and 3 are remarkably loaded towards the ‘regional’ end of the scale whilst for Speaker 4, 86% of respondents placed on the ‘well-spoken’ side of the scale. Speaker 5 is the only example of the five speakers where neither points 1 nor 7 are employed by the respondents and Speaker 5 was viewed differently from other speakers: over 75% of the respondents rated the speaker around the middle three points on the scale, suggesting that Speaker 5 is seen as somehow less ‘well-spoken’ than -253-

Speaker 4, but also less ‘regional’ than Speakers 1, 2 and 3. The results of the ANOVA and the data in the bar chart suggest that the ‘no accent’ scale was meaningful to the respondents and methodologically valid, and that it may have a role to play in any future audiofit. There is a third pillar to this research which investigates what role the respondents’ own origin may have to play in deciding where they believe the speakers to come from: to what extent does inter-community salience mediate recognition and labelling of accents? Across the two groups, most of the respondents were drawn from four regions of the south of the United Kingdom: Wales, South West England, South East England and the Midlands. Respondents were coded according to their region, with any respondents from outside of these four catchment areas being excluded from the treatment. This meant that for G1 n=136, and for G2 n=112. Respondents’ labelling of speaker accent were grouped by region and presented in tables of frequency categorised as ‘yes’ and ‘no’ for the range of the popular responses in each group, and a chi-square test was performed to check whether the labelling was distributed evenly or whether there were any statistically significant differences in the labelling of speaker accent according to respondents' origins. What emerges from the chi-square tests is that for G1, significant associations can be observed between respondent origin and estimations that: Speaker 1 comes from Wales ( <0.001); Speaker 3 comes from the South East ( <0.046); Speaker 4 comes from the South/South East/London ( <0.001). For Speaker 1, as many as 94% of Welsh respondents identify the speaker as Welsh, compared with 60% or below for respondents from English other regions. Conversely, Speaker 3, from the South East, is most frequently estimated to be from the South East (including London) by respondents from the South East. An association between respondent and speaker label is also seen for Speaker 4, the RP-like speaker. Speaker 4 is most frequently identified as from the South/South East/London by respondents from the South East. It is interesting to note that it is the Welsh respondents who return much lower estimation frequencies for an RP-South/South East/London origin than respondents from English regions. This may indicate less familiarity with the hierarchical nature of language varieties in England. For G2, significant associations ( <0.05) were observed between respondent origin and accent labelling for: Speaker 3 being from the South East/London ( <0.016) and Speaker 4 being from the South East/London ( <0.001). These are almost identical to those observed for Speaker 3 and 4 in the first group of respondents. The fourth pillar of this research was to go beyond accent and look at voice qualities. Running repeated measures ANOVAs for these categories on all speakers revealed that significant differences between speakers were found in all categories, suggesting that the categories were meaningful to the non-linguists and that they were able to utilise the 7-point scales to convey their impressions regarding the voices. Characteristics that stand out (based on observed significant differences ( < 0.05)) follow, including their mean scores. Speed: Speaker 5 (5.0) and Speaker 2 (4.8) are fast speakers, Speakers 3 (3.9) and 1 (3.8) are medium speakers and Speaker 4 (2.6) is a slow speaker. Confidence: Speaker 5 (4.45) is more confident than 1, 2 and 3, Speaker 4 (4.29) is more confident than 2 and 3, Speaker 1 is more confident than 3. Pitch: Speaker 2 has the highest pitch (4.6), followed by Speakers 5 (4.3) and 3, (4.2) followed by Speaker 1 (2.8) with a relatively low voice and Speaker 3 (2.1) rated as having a deep voice.

-254-

Melodiousness: Speaker 5 (4.8) is more than average and therefore melodious, followed by Speaker 2 (4.4), followed by Speakers 1 (3.9) and 3 (3.8) with Speaker 1 rated at a low (monotonous) 2.5. Loudness: Each was significantly different from each other: Speaker 1 (5.1), Speaker 5 (4.6), Speaker 2 (4.2), Speaker 3 (3.7) and Speaker 4 (3.3); Steadiness: Again, each was significantly different from each other: Speaker 4 (5.35), Speaker 1 (4.95), Speaker 5 (4.6), Speaker 3 (4.0) and Speaker 2 (3.5). Elaborating on this further, we can turn the data into boxplots for each speaker: 7

6

5

4

3

2

1 Speaker1 Slow-Fast

Speaker1 Speaker1 Speaker1 Speaker1 Speaker1 Hesitant - Deep Voice - Monotone - Soft-Loud Shaky-Steady Confident High-Pitched Melodious

Figure 3: Boxplot for Speaker 1 voice characteristics

-255-

7

6

5

4

3

2

1 Speaker 2 Slow-Fast

Speaker2 Hesitant Confident

Speaker 2 Deep voice High-Pitched

Speaker 2 Monotone Melodious

Speaker 2 Soft - Loud

Speaker 2 Shaky-Steady

Figure 4: Boxplot for Speaker 2 voice characteristics 7

6

5

4

3

2

1 Speaker 3 Slow-Fast

Speaker 3 Hesitant Confident

Speaker 3 Deep voice High-pitched

Figure 5: Boxplot for Speaker 3 voice characteristics

-256-

Speaker 3 Monotone Melodious

Speaker 3 Soft-Loud

Speaker3 Shaky-Steady

7

6

5

4

3

2

1

Speaker4 Slow-Fast

Speaker4 Hesitant Confident

Speaker4 Deep Voice High-pitched

Speaker4 Monotone Melodious

Speaker4 Soft-Loud

Speaker5 Monotone Melodious

Speaker5 Soft-Loud

Speaker4 Shaky-Steady

Figure 6: Boxplot for Speaker 4 voice characteristics

7

6

5

4

3

2

1 Speaker5 Slow-Fast

Speaker5 Hesitant Confident

Speaker5 Deep voice High-pitched

Speaker5 Shaky-Steady

Figure 7: Boxplot for Speaker 5 voice characteristics

The boxplot for Speaker 1 demonstrates that ‘deep voice—high-pitched’ provokes a very clear reaction, with 50% of respondents placing the voice between points 2 and 3, and we can

-257-

observe the direct opposite for evaluations of the speaker on ‘soft—loud’, with the speaker attracting ratings between 5 and 6. For Speaker 2, the boxplot is a little less dramatic showing fewer extremes of ratings although higher than medium ratings for speed, pitch, melodiousness and loudness. There is quite a wide range of opinion for hesitant—confident. The boxplot for Speaker 3 indicates a concentration of opinion on three features: medium hesitancy—confidence and loudness, but a higher pitched voice. For Speaker 4 we see some strong characterisation of a slow speaker with a deep, monotonous generally steady voice and for Speaker 5 there is quite a range of opinion on some points, but there is certainly a characterisation of the speaker as having a higher than medium voice, louder than medium and steadier than medium voice. The fifth and final pillar of this research is to consider an alternative way of eliciting voice and accent information from non-linguists using a technique that uses an alternative form of cognitive elaboration: soundalikes. Where soundalikes were provided, the names associated with the voices provide interesting reading. Speaker 1 was compared with ‘Goldie Lookin’ Chain’ (rap group from Newport, South Wales), ‘Dirty Sanchez’ (South Wales 'stunt' TV show), ‘Kelly Jones from Stereophonics' (South Wales pop group), a friend (respondent was from South Wales), ‘my brother’ (respondent was from South Wales). What is clear is that every soundalike mentioned has an association with South Wales and the three famous people mentioned all have accents that would be associated with industrialised urban areas of South Wales. Speaker 2 was compared with Gareth from 'The Office' (British TV comedy), ‘Vickie Pollard’ (Bristol character from 'Little Britain'), ‘Little Britain woman’ (presumably the same character from the TV comedy programme), ‘a farmer’, ‘people from my hometown’ (the respondent was from South West), ‘a friend’ (the respondents were from the South West, although we cannot be certain where the friends are from). The regional identification data in the earlier sections confirm that most respondents believed Speaker 2 to be from the South West of England, many of them from Bristol. An obvious connection appears in that all the famous soundalikes identified are famous for their regional accents, in particular their realisation of the postvocalic [r] shibboleth that can so readily identify those southern English speakers that come from the West or South West of England. This accent is also readily stereotyped as a ‘farmer’s’ accent, an example of which also appears. Speaker 3 attracted by far the highest frequency of single soundalike: ‘David Beckham’ (English footballer), with 36 comparisons. Also included were ‘Frank Lampard’ (English Footballer) ‘Mike Skinner’ (rapper from 'The Streets' pop group) and ‘my parents from London.’ The 36 soundalikes for David Beckham is a remarkable figure for a non-essential category within the questionnaire. But what was it in Speaker 3’s voice that was so salient to the respondents? It would be uncontroversial to say that Beckham, a famous footballer from just outside London (Essex) is well known for his strong regional, estuary English speech variety. And these are the very same features that have been identified with Speaker 3 throughout by both groups of respondents. The choice of Frank Lampard may echo the same reasoning as for David Beckham. It may be that the respondent perceives Lampard to share similar, high-pitched estuary English tones. But David Beckham as a label could be conveying more than simple accent information: Beckham in particular is impersonated on television by comedians, etc., and a constant characteristic to be mimicked is not just his accent but soft, high, voice; and these are features described by a number of respondents in the polar voice descriptions. Speaker 4, who was characterised as having a deep monotonous voice with an RP-like accent was compared to famous TV comedian with an RP-like voice, Stephen Fry, and the famously dead-pan, monotonous deep-voiced comedian Jack Dee. Comparisons were also

-258-

made with the TV actor/comedian David Walliams and the singer Damon Albarn from the pop group ‘Blur’. Dee, Albarn and Walliams can all be characterised as having a somewhat understated and unmelodious delivery and again, it would appear that the respondents are conveying more than accent information with their soundalikes. Speaker 5, whose accent had been identified with his home region by very few people and who appears to be more likely to be identified as having no accent—neither strongly regional, nor RP—on the accent scale, was identified as sounding like just two people: Frank Skinner (TV comedian) and Scott Mills (radio presenter) who many would describe as having an ‘average’ to ‘high’ pitched voice, and it may be these salient features noted by the respondent that suggested this soundalike. It is also worth noting that in this research, the speaker without an overt regionally or socially marked accent attracted the fewest soundalikes. Discussion and conclusion Given the need for the police and legal professionals to elicit accent descriptions from nonlinguists, the first research question in this article was to investigate whether non-linguists have the ability to perceive regional and social accents. The data from this study demonstrate that the non-linguist respondents are capable of perceptual discrimination between the accents of the speakers presented. However, variation was observed in the accuracy and focus of the estimations of speaker origin, with those speakers using traditional regional accents (Speakers 1, 2 and 3) receiving the highest frequency of accurate estimations of origin. Those speakers who were not strongly associated with any one region (Speakers 4 and 5) received more disparate estimations of origin. Moreover, the introduction in this study of guided perceptual mapping as an elaboration of Preston’s (1996) and Niedzielski and Preston’s (1999) approach to folk linguistics research has shown itself to have applications in the forensic field, in that it assists not only in the perceptual differentiation of regions and accents, but also in the expression of social associations with an accent that has no overt geographical origin. Researchers and legal professionals may wish to draw upon and further develop such a technique. With regard to the second research question—whether accent perception is a function of respondent origin—this study has provided evidence that social and perceptual distance and not just Euclidian measures between a respondent and a speaker’s origin have a role to play in the accuracy of the identification. The evidence of perceptual asymmetry between, for example, Wales and England and the South East of England compared with other regions is consistent with other data from the UK context ( Garrett et al., 2004; Kerswill and Williams, 2002; Williams et al., 1996). Perceptual distance and within-community variation and salience should play a role in future research designs. The use of the ‘strength of accent’ scale served to illustrate a non-geographical medium for eliciting accent perceptions and labels from non-linguists, addressing the third research question. For example, Speaker 5 was identified as having ‘no accent’, and whilst linguists will know it to be impossible to speak a language without communicating some regional or social information about oneself, the data from the ‘strength of accent’ scale lend support to the conclusion that, in the UK, the concept of ‘no accent’ is a very real nonlinguist alternative to, on the one hand, the socially-marked RP/well-spoken/Oxford/‘posh’ English and on the other hand, regional/accented English. Turning to the fourth research question regarding what non-linguists notice about a voice, inviting non-linguists to rely on their own cognitive representations and processes when describing voices and accents endows resultant descriptions with a level of community authenticity essential to the field of forensic speaker description. Identifying and engaging

-259-

with non-linguists’ own perceptual scales should inform practice amongst forensic linguists and legal professionals. Additionally, Kerswill (2002) has suggested that the recognition of non-community accents may be mediated by whether the voice sounds like someone the judge happens to know. The data in this study have shown that there is a clear potential for non-linguists to articulate their perceptions by describing a person known to them— soundalikes. The remarkably high rate of ‘David Beckham’ soundalike descriptions served as an overwhelmingly powerful alternative to description through other, more technicallyoriented channels of accent and voice description. Linguists may baulk at the idea of introducing such non-technical opinions into research but it is non-linguist police officers and non-linguist earwitnesses that we are focusing on, and so long as such terms are universally salient and have wide currency, they remain viable ways of communicating what has been heard from one non-linguist to another. A further avenue of research may be to investigate what role linguists could play in relating the description of the earwitnesses to technical descriptions of language. In conclusion, I discussed at the beginning how around the world, non-linguist police officers routinely take statements from non-linguist earwitnesses in relation to linguistic issues, meaning witnesses to and victims of crime are being questioned by almost exclusively linguistically untrained people. In answer to the fifth research question, whilst only a first step in researching the potential for FSD, the data presented here suggest that there is indeed potential for the creation of an audiofit to augment current practice. The data suggest that the illustrated gap between current practices and non-linguist abilities should be reviewed and that researchers should assume that non-linguists are capable of perceiving linguistic detail and providing relatively meaningful data. No doubt, this first venture into researching FSD cannot be assumed to present a finished methodology and further testing on a range of respondents populations, using a wider cabinet of exemplars, is required in order to arrive at a reliable and valid audiofit instrument. However, if the situation is to be improved, researchers, police and legal professionals need to acknowledge the weaknesses in the current system and develop a coordinated and systemic typology for eliciting descriptive forensic linguistic data—an appropriate audiofit instrument with which enhance the evidence gathering process and affect the course of social justice. References Braun, A. (1996) Age estimation by different listener groups. Forensic Linguistics: The International Journal of Speech, Language and the Law 3(1): 65—73. Broeders, A.P. (1996) Earwitness identification: common ground, disputed territory and uncharted areas. Forensic Linguistics: The International Journal of Speech, Language and the Law 3(1): 3—15. Bull, R. and Clifford, B. (1984) Earwitness voice recognition accuracy. In G. L. Wells and E. F. Loftus (eds) Eyewitness Testimony: Psychological Perspectives. Cambridge: Cambridge University Press, 92—123. Dixon, J.A. and Mahoney, B. (2004) The effect of accent evaluation and evidence on a suspect's perceived guilt and criminality. Journal of Social Psychology 144(1): 63—73. Fisher, R.P. & Geiselman, R.E. (1992) Memory Enhancing Techniques for Investigative Interviewing: The Cognitive Interview. Springfield: Charles C. Thomas.

-260-

Gallois, C. and Callan, V. (1981) Personality impression elicited by English accented speech. Journal of Cross-Cultural Psychology 12: 347—359. Garrett, P., Coupland, N. and Williams, A. (2004) Adolescents' lexical repertoires of peer evaluation: boring prats and English snobs. In A. Jaworski, N. Coupland and D. Galasinski (eds) The Sociolinguistics of Metalanguage. Berlin: Mouton de Gruyer, 193—226 Hollien, H. (1990). The Acoustics of Crime. New York: Plenum Press. Hollien, H. and Schwartz, R. (2000) Aural-perceptual speaker identification: problems with non-contemporary samples. Forensic Linguistics: The International Journal of Speech, Language and the Law 7(2): 199—211. Kerswill, P. and Williams, A. (2002) Dialect recognition and speech community focussing in new and old towns in England. In D. Long and D. Preston (eds.) Handbook of Perceptual Dialectology Vol. 2 Michigan: John Benjamins Publishing Company, 173—204. Niedzielski, N. and Preston, D. (1999) Folk Linguistics. Berlin: Mouton de Gruyer. Nolan, F. (2003) A recent voice parade. International Journal of Speech, Language and the Law 10(2): 277—291. Nolan, F. and Grabe, E. (1996) Preparing a voice lineup. Forensic Linguistics: The International Journal of Speech, Language and Law 3(1): 74—94. Philippon, A.C., Cherryman, J., Bull, R. and Vrij, A. (2007) Lay people's and police officers’ attitudes towards the usefulness of perpetrator voice identification. Applied Cognitive Psychology 21(1): 103—115. Preston, D. (1996) Whaddayaknow? The modes of linguistic awareness. Language Awareness 5(1): 40—74. Seggie, I. (1983) Attribution of guilt as a function of ethnic accent and type of crime. Journal of Multilingual and Multicultural Development 4(2): 197—206. Shuy, R. (1993) Language Crimes: The Use and Abuse of Language Evidence in the Courtroom. Cambridge MA: Blackwell. van Bezooijen, R. and Ytsma, J. (1999) Accents of Dutch: Personality impression, divergence and identifiability. Belgian Journal of Linguistics 13(1): 105—129. van Bezooijen, R. and van Hout, R. (1985) Accentedness ratings and phonological variables as measures of variation in pronunciation. Language and Speech 28(2): 129—142.

-261-

The study of inter- and intra-speaker variation towards an index of idiolectal similitude Núria Gavaldà-Ferré ForensicLab, IULA, Universitat Pompeu Fabra, Spain [email protected] Abstract This paper is fundamentally concerned with the study of an individual’s ‘idiolectal style’ for forensic purposes. The main objective is to compare several oral samples and to calculate the linguistic distance between them so that an Index of Idiolectal Similitude (IIS) for the phonological module of English can be created. It is assumed that this IIS can help establish what kind of idiolectal similarity is needed to reliably say that two oral samples have been produced or not by the same speaker of English. The development of this IIS is based on the study of the realisation and distribution of certain phonological variables from the variety of Southern British English. The data that are analysed have been extracted from a panel study in real time, following the Labovian framework, by which the same speakers were recorded at two points in their lives with a time lag of 10-15 years. The main hypotheses of the study are a) that the variation observed in an individual’s idiolectal style will be lower than the variation that exists between different individuals; and b) that an individual’s idiolectal style remains relatively stable throughout time. So far, two different methods for the development of this IIS have been explored. The results for the first method were presented at IAFL 09. This paper deals with the application and the results obtained for the second method, which is a statistical technique based on the calculation of the difference between the expected and the observed frequency adjusted to the total number of tokens analysed (Adjusted Residual Value). These results seem to confirm the initial hypotheses in that a) the variation in the phonological variables used by the same speaker—Intra-speaker variation—appears to be lower than the variation in the same variables between different speakers—inter-speaker variation; and b) the phonological patterns observed within the same speakers seem to remain quite stable throughout time. Keywords:

INTRA-SPEAKER VARIATION; INTER-SPEAKER VARIATION; FORENSIC VOICE COMPARISON; IDIOLECTAL STYLE

1. Linguistic variation and forensic linguistics As variationist sociolinguistics demonstrates, variation is inherent in all languages and all linguistic levels. Thus, the idiolect, which had been applied by structuralists and generativists to the study of language universals, has been shown to be as variable as language itself and therefore not suitable to represent a whole linguistic community (Labov 1973, Trudgill 1974). Moreover, this variation is not random, as it was previously considered to be, but it correlates with internal (linguistic) and external (social) factors. Variation is present in two main levels within a speech community. On the one hand, there is variation across different individuals (inter-speaker variation), which is the result of different speakers having different voices and using language in a different way. On the other hand, there is also intra-speaker variation, since no linguistic production by the same individual can ever be repeated in exactly the same way. Consequently, forensic expert witnesses are required to discern whether the differences found between two (sets of) samples are more likely to be down to inter- or intra-speaker variation. As Rose (2002) explains: [F]orensic speaker identification involves being able to tell whether the inevitable differences between samples are more likely to be within-speaker differences or betweenspeaker differences. […] [T]he greater the ratio of between-speaker to within-speaker variation, the easier the identification. (p. 26) -262-

In this sense, one of the major challenges for the forensic phonetician is to show that despite intra-speaker variation, forensic voice comparison is still possible because each individual has their own distinct linguistic preferences. Several authors have used the term ‘idiolect’ to refer to this individual use of language that each person has (Baldwin 1979; Coulthard 2004). As Coulthard (2004) explains: The linguist approaches the problem of questioned authorship from the theoretical position that every native speaker has their own distinct and individual version of the language they speak and write, their own idiolect, and the assumption that this idiolect will manifest itself through distinctive and idiosyncratic choices in texts. (p. 431—432) However, Turell (2010: 217) argues that this use of the term ‘idiolect’ by forensic linguists may not be accurate for two main reasons. Firstly, through the use of the term ‘idiolect’, the controversy around this notion within 20th Century linguistics seems to be ignored. As mentioned above, structuralists and generativists claimed that a single speaker’s idiolect reflected the homogeneous nature of a linguistic community, whereas Labovian sociolinguists argued that this homogeneity of language was a fallacy and that a speaker’s idiolect could not be used to analyse the speech of a whole community because it was as variable as language itself. Secondly, it would only be possible to describe idiolects through innumerable databases containing a speaker’s linguistic production, which is an impossible task. Therefore, Turell proposes that the term ‘idiolectal style’ should be used in forensic contexts in order to focus on the fact that each person favours certain linguistic features which constitute their individual use of language. What follows is a study that aims at establishing the border between intra- and inter-speaker variation, thus separating different idiolectal styles. 2. Index of idiolectal similitude The study that is described in this paper1 is concerned with the analysis of intra- and interspeaker variation and aims at defining the border between these two types of variation so as to separate between different idiolectal styles. The main objective of this analysis is to create an Index of Idiolectal Similitude (IIS from now on) that can compare different oral samples and calculate the linguistic distance between them so that it can be established whether they have been produced by the same or by a different speaker.

Figure 1: Representation of the Index of Idiolectal Similitude as a continuum. 1

This study is part of a research project being conducted at ForensicLab, IULA, Universitat Pompeu Fabra: ‘Forensic Idiolectometry and Index of Idiolectal Similitude’ (FFI2008-03583/FILO; PI: Prof M. Teresa Turell). The project is funded by the Spanish Ministry of Science and Innovation.

-263-

As represented in Figure 1, this IIS is conceived as a continuum between 0 and 1, where 0 indicates minimum similarity and 1 corresponds to maximum similarity. Ideally, in a forensic context when comparing two (sets of) oral samples so as to establish whether they have been uttered or not by the same individual, a near-zero result would indicate that the samples would belong to different speakers, in other words, the samples would show inter-speaker variation. A result at an intermediate position within this continuum would still show inter-speaker variation, but from speakers belonging to the same linguistic variety. Finally, a value close to 1 would show the existence of intra-speaker variation and suggest that the compared samples were very similar and they could have been uttered by the same person. 2.1. Corpus of study The corpus used in this study contains data on spontaneous speech from nine speakers of the same variety of English, born around the same dates (between 1951 and 1961) and raised in the same area of the South-East of England, namely London and the Home Counties. For each speaker, two different sets of samples were collected at two different measurement times with a time slot of 15 to 20 years between the two. Measurement time one (MT1) corresponds to the end of the eighties and beginning of the nineties whereas measurement time two (MT2) is from 2005 to 2010. The speakers are divided into two main sub-corpora according to whether they have been in a situation of language contact or not. The two main sub-corpora are the following: a) LanCon contains sociolinguistic interviews conducted at two different times in the lives of three English speakers (two men and one woman) who settled in Catalonia (Spain) during the 90s and are still living there working as teachers of English. The data were collected following Labov’s proposal of real time longitudinal studies (Labov 1994). The study conducted is a panel study, which consists of locating the same subjects of a first study and analysing the changes in their linguistic patterns after a period of time. b) InSit includes data from six speakers (three men and three women) collected at two different times. The difference with the previous sub-corpus is twofold: these speakers have not been in such long-term situation of language contact as the speakers in LanCon; and the data were not collected through sociolinguistic interviews, but they consist of speech (from interviews) extracted from on-line sources (such us Youtube or bbc.co.uk), since the informants are famous TV or radio presenters and singers. 2.2. Variables of study The variables that have been analysed in this study correspond to phonological processes from the variety of English under consideration—English from the South-East of England—, and they are all related to processes of reduction (or elision), addition or shift of sounds. Table 1 shows a list of the variables under analysis, which were selected for their idiosyncratic nature after several preliminary studies.

-264-

Table 1: List of phonological variables analysed.

2.3. Main hypotheses The experiment presented in this paper is based on three main hypotheses, which are formulated as follows: 1. A speaker’s idiolectal style will remain relatively stable despite the course of time. Therefore, results from comparisons of the same speakers should be closer to 1 in the IIS continuum. 2. The intra-speaker variation of the informants that have been in a situation of language contact will be higher than the intra-speaker variation of those who have not. Thus, results from intra-speaker comparisons from speakers that have not been in a situation of language contact are expected to be closer to 1 than those from speakers who have not. 3. Inter-speaker variation will be higher than intra-speaker variation. Consequently, results from comparisons between different speakers are expected to be closer to 0. 2.4. Methods of analysis The qualitative study of the variables was undertaken through an auditory-acoustic analysis, which is the method most widely used in forensic voice comparison2. This approach implies that expert witnesses use both their linguistic knowledge of the phonetic-phonological system that is being considered and also the techniques especially developed for speech analysis -F0 analysis, vowel formant measurements, use of spectrograms, waveforms and spectra, etc. 2

See French 1994, Nolan 1997 and Watt 2010 among others.

-265-

The quantitative analysis was performed by means of one of the methods that have been tested for the calculation of the IIS. Results for a first method based on the calculation of the percentage of occurrence of each variant were presented in IAFL 09. This paper addresses the application and the results obtained for the second method, which is based on the analysis of the chi-square test -since the variables are categorical- and the calculation of the Adjusted Residual Value (ARV) for the distribution of every variable. This ARV shows the difference between the expected value and the observed value adjusted to the total number of realisations of each variable and indicates how similar the distribution of the variables between the two speakers being compared is. The results are subsequently normalised in order to get a value between 0 and 1 so that the IIS can be obtained. The calculations of the IIS were obtained by doing comparisons of two main types: 1. Speaker A in measurement time one (MT1) was compared with speaker A in measurement time two (MT2), so as to assess whether a speaker’s idiolectal style remains relatively stable throughout time. 2. Speaker A in MT1 was compared with speaker B in MT1, with the purpose of testing whether inter-speaker variation is higher than intra-speaker variation. These comparisons can be: a) comparisons of speakers within the same sub-corpora (LanCon and InSit); and b) comparisons of speakers across the two sub-corpora. Besides, two main factors were considered when comparing different speakers, namely the sub-corpus each speaker belongs to (LanCon or InSit) and the gender of the speakers. 3. Results The experiments conducted to evaluate hypothesis one, which states that a speaker’s idiolectal style will remain relatively stable despite the course of time, show results that would validate this hypothesis. Figure 2 shows the results for comparisons between the same speaker at two different times, and every point in the graph correspond to the IIS result of one comparison. All these IIS results are close to 1, more precisely between 0.81 and 0.91, which indicates that the variation found when comparing samples produced by the same speaker is quite low - the closer to the one endpoint of the continuum, the less the variation. In the light of these preliminary results, the boundary between intra- and inter-speaker IIS values could be set at 0.81, which is the lowest intra-speaker result.

-266-

Each point in the graph corresponds to the IIS result of a comparison between two samples.

Figure 2: Summary of all intra-speaker results.

Contrary to hypothesis 1, hypothesis 2 does not seem to be validated. It was initially predicted that the intra-speaker variation of the informants that have been in a situation of language contact would be higher than the intra-speaker variation of those who have not. However, as can be seen in Figure 2, the intra-speaker values from sub-corpus LanCon are not lower (and therefore show as much variation) than those from sub-corpus InSit. This suggests that a long-term situation of language contact does not seem to have an effect on the phonological patterns under consideration on these speakers.

Each point in the graph corresponds to the IIS result of a comparison between two samples.

Figure 3: Summary of all inter- and intra-speaker values.

The third hypothesis, which states that inter-speaker variation will be higher than intra-speaker variation, also seems to be validated. Figure 3 shows a summary of inter- and intra-speaker values and there is a clear tendency for inter- and intra-speaker values to separate towards different directions: intra-speaker values, as seen in the previous figure, go towards the 1 endpoint of the continuum, thus showing less variation. On the other hand, 90% of the interspeaker results show lower values than the 0.81 that had been established as the border between intra- and inter-speaker variation. There is, however, a remaining 10% of inter-speaker values -267-

that overlap with the intra-speaker ones, and we will try to explain the cause of this overlap by looking at inter-speaker variation from two perspectives: a) whether or not the speakers have been in a situation of language contact; and b) whether the comparisons are between people of the same gender or otherwise.

Each point in the graph corresponds to the IIS result of a comparison between two samples.

Figure 4: Inter-and intra-speaker values from corpus LanCon (language contact).

Figure 4 shows the results within the sub-corpus of language contact (LanCon). 50% of the comparisons between the different speakers are lower than 0.81, and therefore show more variation. However, another 50% of the comparisons return results higher than 0.81, showing variation that could erroneously be classified as intra-speaker. It would be necessary to have more data in order to make a reliable interpretation of these results. Yet, it is interesting that the speakers who show less variation between each other are the ones that have been in the same situation of language contact.

Each point in the graph corresponds to the IIS result of a comparison between two samples.

Figure 5: Inter- and Intra-speaker values from corpus InSit (no language contact).

The comparisons done within the InSit sub-corpus, shown in Figure 5, demonstrate that interspeaker values clearly tend to be lower than 0.81, since 93% of inter-speaker comparisons show lower values than the intra-speaker comparisons, i.e. between 0.44 and 0.80. The other seven per -268-

cent, which in absolute numbers means two comparisons out of thirty return values that could be interpreted as intra-speaker variation since they are higher than 0.81.

Each point in the graph corresponds to the IIS result of a comparison between two samples.

Figure 6: Inter-speaker values crossing the two sub-corpora.

If we make comparisons across the two sub-corpora, as seen in Figure 6, we see the same tendency for intra- and inter-speaker values moving away from each other in the IIS continuum. Thirty-two out of thirty-six comparisons (89%) between speakers show high variation and return results above 0.81, more precisely they range between 0.59 and 0.79. Again, some of the comparisons, in this case four of them, show higher values in the continuum that could be interpreted as intra-speaker values.

Each point in the graph corresponds to the IIS result of a comparison between two samples.

Figure 7: Summary of all inter-speaker values regarding gender.

Considering gender as a factor that can affect inter-speaker variation, we see in Figure 7 that, although far from being a categorical difference, there is a tendency for comparisons between speakers of the same gender to be higher (and therefore show less variation) than those comparisons between speakers with different biological gender. The comparisons between -269-

subjects of the same gender range from 0.44 to 0.77 whereas the inter-speaker values from individuals of different gender range from 0.61 to 0.81. This tendency could be accounted for by the gender stratification that some of the variables considered show. For example, the spirantisation of /t/ between vowels is described in the literature as being a more prestigious variant than [ʔ]or [ɾ] in British English and it is more usual for women to use it than for men3. To sum up, the results for the experiments that have been conducted show two main important factors to be considered. On one hand, intra-speaker comparisons give results that are closer to one, and therefore show low variation. These results confirm the hypothesis that a speaker’s ‘idiolectal style’ seems to remain quite stable despite the course of time and a longterm situation of language contact. On the other hand, the inter-speaker variation has proved to be higher than intra-speaker variation which confirms the proposal formulated in section 1 that although there is intra-speaker variation, each speaker has a unique ‘idiolectal style’ that separates them from the rest of speakers from the same community. 4. Conclusion The research presented in this paper is a preliminary approach to the study of the ‘idiolectal style’ as regards several phonological variables of Southern British English. Thus, the study of inter- and intra-speaker variation is proposed as a tool that can help expert witnesses conduct forensic voice comparison in a more rigorous and reliable way. An experimental method towards the establishment of an Index of Idiolectal Similitude has been proposed that can determine the idiolectal distance between the phonological patterns of different speakers. Results have shown that inter-speaker variation seems to be higher than intra-speaker variation and that an individual’s ‘idiolectal style’ seems to remain relatively stable throughout the course of time. Moreover, a contribution has been made to the Base Rate Knowledge, necessary for forensic voice and text comparison, for fourteen phonological variables belonging to Southern British English. Yet, it is necessary to continue exploring the idiosyncratic nature of the variables that have been analysed so far and to evaluate the hypotheses formulated with a larger corpus of study and with real forensic cases. References Baldwin, J. (1979) Phonetics and speaker identification. Medicine, Science and the Law 19: 231—2. Coulthard, M. (2004) Author identification, idiolect, and linguistic uniqueness. Applied Linguistics 25 (4): 431—447. French, P. (1994) An overview of forensic phonetics with particular reference to speaker identification, Forensic Linguistics 1: 169—181. Haslerud, V. (1995) The variable (t) in Sydney adolescent speech, Unpublished Cand. Philol. dissertation, University of Bergen. Labov, W. (1973) Sociolinguistic Patterns. Philadelphia: University of Pennsylvania Press. Labov, W. (1994) Principles of Linguistic Change: Internal Factors. Oxford: Blackwell. 3

Loakes and McDougall state that although all variants ([t], [ts], [ɾ] and [ʔ]) are used to some degree by all socioeconomic groups, the fricated variant [ts] is ‘overwhelmingly used by female speakers, in more formal speech, and in speakers from a higher socioeconomic background’ (2010: 159). Haslerud (1995) quoted in Loakes and McDougall also points out that ‘[t]here are indications that [t] and [ts] are high prestige markers’.

-270-

Loakes, D. and McDougall, K. (2010) Individual variation in the frication of voiceless plosives in Australian English: a study of twins’ speech. Australian Journal of Linguistics 30 (2): 155—181. Nolan, F. (1997) Speaker recognition and forensic phonetics. In W. Hardcastle and J. Laver (eds) The Handbook of Phonetic Sciences. Wiley-Blackwell, 744—67. Trudgill, P. (1974) Sociolinguistics: An Introduction to Language and Society. Great Britain: Penguin Books. Turell, M.T. (2010) The use of textual, grammatical and sociolinguistic evidence in forensic text comparison. The International Journal of Speech, Language and the Law 17(2): 211—250. Rose, P. (2002) Forensic Speaker Identification. London: Taylor & Francis. Watt, D. (2010) The identification of the individual through speech. C. Llamas & D. Watts (eds) Language and Identities. Edinburgh: Edinburgh University Press, 76—85.

-271-

Computer-mediated investigative interviews: A potential screening tool for the detection of insider threat Marisa Jenkins1 and Coral J. Dando2 Department of Psychology, Lancaster University, UK 1 [email protected] 2 [email protected] Abstract Within secure organisations, the knowledge that employees have of policies, procedures, and technologies means that they are also often aware of potential vulnerabilities. Hence, the threat of insider attack, when legitimate users maliciously leverage their privileges, familiarity, and proximity to negatively affect the confidentiality and/or integrity of an organisation’s information, is real. When an insider event occurs, the number of witnesses and potential suspects to be interviewed, sometimes by non-experts, is a significant problem for investigators. We suggest that the substitution of a written text for a spoken, face-to-face interview may be usefully applied in this investigative context in obtaining information from witnesses and suspects and, further, that such an interview could be computer-mediated. We propose that the computer-mediated Self-Administered Veracity Questionnaire (SAVQ) designed for this purpose may also prove to be a useful tool in other investigative contexts. This research was funded by the UK Government Communications Planning Directorate. Keywords:

INVESTIGATIVE INTERVIEWING; COMPUTER-MEDIATED INTERVIEW; SELFADMINISTERED VERACITY QUESTIONNAIRE; INSIDER THREAT

Introduction Professional investigators are tasked with answering two primary questions, namely what has happened and who might be responsible. To answer these questions they need information, which is typically gathered during face-to-face interviews with witnesses, victims, and suspects. Worldwide, there exists much legislation, and numerous organisation-specific guidelines as to how interviews should be conducted, which investigators are duty bound to follow. In addition, there is an ever-increasing body of psychological and linguistic empirical literature that presents additional methods of improving these dynamic and complex social interactions. Accordingly, tensions often exist between limited resources and strict time pressures, and the practices and procedures advocated by linguists and psychologists (for example, Poyser & Milne, 2011). However, investigators, psychologists and linguists are also usefully working together to advance investigative interview practice, one example being the development of a Self Administered Interview for witnesses and victims of crime (SAI) (Gabbert, Hope, & Fisher, 2009). The SAI is a standardised witness interview in the form of a written text (rather than spoken face-to-face communication), which witnesses complete independently, immediately post-event. The benefits of the SAI have been demonstrated during trials endorsed by the Association of Chief Police Officers (ACPO) and conducted by Greater Manchester Police in 2009 (Gabbert, Hope & Fisher, in press), and it has now been adopted by a number of UK police forces. In its current form the SAI is designed to consolidate and protect long-term eyewitness episodic information. However, we suggest that the substitution of a written text for a spoken, face-to-face interview can be usefully applied in other investigative contexts, -272-

one of these being the investigation of insider threat occurrences, where interviewers are faced with considering how to preserve episodic information (for evidential purposes), while concurrently maximising opportunities to detect wrongdoing. Within secure organisations, the knowledge that employees have of policies, procedures, and technologies means that they are also aware of vulnerabilities. Hence, the threat of insider attack, when legitimate users maliciously leverage their privileges, familiarity, and proximity to negatively affect the confidentiality and/or integrity of an organisation’s information, is real. Keeney et al have identified the ‘critical infrastructure sectors’ (2005: 5) which they argue should be addressed in the ‘securing’ of cyberspace. These include banking and finance, information and telecommunications, transportation, postal and shipping, emergency services, continuity of government, food, energy and water. However, for some organisations such as government authorities and agencies, insider threat is an ongoing problem with exceedingly severe risks. Of the numerous problems that arise when investigating insider threat occurrences, managing the large number of witnesses and suspects, in terms of how to expeditiously elicit information from both groups, represents a significant challenge. Moreover, that the initial information gathering process is likely to be conducted by non-experts in a work environment (e.g., by internal security officers, line managers, work colleagues, etc.) lends support to our contention that in the first instance, a written text (produced by a computer-mediated procedure) will offer real additional advantages while retaining the benefits of a more traditional face-to-face interview. There now follows a brief overview of a research project concerning insider threat occurrences. Designed to mimic a secure working environment, including the presence of insider threats, the project sought to investigate, among other things, the efficacy of a novel self administered (computerised) veracity assessment tool (SAVQ). Pilot data suggests that it is informational content, rather than paralinguistic features, that is most salient in the process of screening witnesses and prioritising potential suspects in this context, and that this process could be achieved via a computer-mediated SAVQ. Accordingly, we offer the SAVQ as suitable for use by both specialist and non-specialist investigators, as a ‘first layer’ insider threat investigative tool. Detecting insider threats (DIT) From its inception the DIT project sought to fill an apparent gap in the empirical literature pertaining to the investigation of insider threat occurrences. Existing research (for example, Carnegie Mellon, 2011; Chapman, Appleyard, Christie, D’Silva & Glanville, 2007; Hoffman, Meyer, Schwarz. & Duncan, 1990; Miller, 2003) has advocated approaches that simply combine existing behavioural approaches (where mock suspects use researcher presented deceptions) and technical perspectives. While providing useful insights into the motivations for, and typical activities associated with insider threats, these approaches take no account of the need to prioritise investigative resources, and fail to consider how to protect event information to best effect. We suggest that only when organisations have access to a ‘tool box’ of techniques can they fully investigate an insider event. The DIT project, on the other hand, not only mimicked a secure working environment to ‘observe’ multiple behaviours (including linguistic behaviour) and generate much-needed data, but it also sought to develop novel investigative tools. Unlike other research in this domain which has focused on researcher-led deceptions (as cited above), participants (whether they be insiders or not) were given no instructions as to how they should undertake their tasks, communicate with other participants, or construct any deceptions during the simulation. Neither were they given any information or guidance as to how they might act or what strategies they might use for convincing an interviewer that they were telling the truth, thereby generating, as far as -273-

possible in a laboratory environment, a similar cognitive load as might be experienced by genuine insiders (for more on generating and maintaining lies in an interview setting see Dando & Bull, 2011: 194). Our approach to witness and/or suspect behaviour does not rely on those behavioural indicators that have been found to be both subjective and over-sensitive as a means of identifying insider threat (Chapman et al, 2007). Rather, it relies on past behaviour and expected patterns of work as more reliable indicators of expected future behaviour, with employers themselves being able to set the limit for variance from an expected ‘norm’ before an alert is raised. If all employees’ work activities are constantly monitored by computer software, there is less risk of human error compromising an otherwise effective security system, as well as reducing the need for staff monitoring, either by management, or by specifically trained behavioural experts. Furthermore, under the relatively inconspicuous but universal monitoring of computer software, employees may feel reassured of the objectivity of such a security system, without feeling the potential lack of trust apparent in the level of human observation required to produce accurate judgements of complex and subjective behavioural indicators. It is hoped that the DIT project will demonstrate that a combination of task-specific technology with behavioural analysis can identify and perhaps even predict an insider threat. Our incorporation of an investigative solution to confirm or repudiate the potential risks identified is not only an innovative contribution to the management of insider threat, but is specifically tailored to the problems of time constraints and high numbers of witnesses/suspects faced by secure organisations at risk from insider threat. Method Simulation To simulate a secure working environment three ‘office’ locations were set up, in which twelve participants played in three teams of four (randomly allocated with each team member having an equal status role) to solve a number of fictitious crimes. Immediately prior to the simulation, all participants received half an hour of training which provided an overview of the research, role information, and instructions concerning the sensitive and confidential nature of their working environment: that is they were to imagine working with information about real crimes and real criminals, and to act accordingly. The simulation was divided into a ‘game’ of four 60-minute rounds. At the beginning of each round, teams were provided with information about fictional gangs of criminals in the form of a crime ‘narrative’ of clues. These clues, along with instructions such as completing their task as quickly as possible, were presented as a video-recorded briefing (via their personal laptop) from their boss. In addition, participants received both individual and group emails, text messages and phonecalls, some containing instructions, and some that were ‘social’ and unrelated to the simulation, but were intended to more realistically represent the flow of email traffic in a workplace. The information needed to solve the ‘crimes’ was contained within a number of databases, which held ‘intelligence’ about fictional gangs and gang members. Each player had access to a unique combination of the databases, some of which were shared-access and some of which were single-user. No one team member was able to solve the puzzle using only the information in their own databases, hence team members had to interact with each other in order to complete their task. Players were tasked with working in their teams to identify and locate gang members, and with determining an appropriate arrest order. Teams’ -274-

tasks were identically structured so that roles were comparable across teams, and each player’s role was structured to ensure that each player within a team had a similar workload. To simulate a secure working environment a silent-working ‘Office’ rule was imposed, but with an adjacent ‘Meeting Room’ in which participants could talk (both were covered by audio- and digital video-recorders). A ‘Printer Room’ was nearby and contained a printer which all participants needed to use if they wished to print notes, maps, etc, since no writing materials were provided (a security measure sometimes utilised by secure organisations). In order to represent the departmentalisation of (secure) working environments, and to successfully record as much written (email and text message) and spoken language as possible, the silent office was separated (by room dividers) into four sections, with participants being grouped according to role, rather than team. Participants were not permitted to talk in this room, but were provided with a laptop and a mobile phone, and so could contact any other participant using the latter (by text and/or voice call), and by email. Participants then either arranged to go to the adjacent Meeting Room if they wished to talk face-to-face, or were advised/instructed (by an email from their boss) to hold a team meeting. The Meeting Room included only tables and chairs for each team, with no stationery or other facilities for note-taking. Participants were encouraged to move freely between these rooms, and the Printer Room, in order to compile their notes, talk to each other, and to reach a solution to the puzzle with which they were presented. Although a completely silent working office may not be realistic, pilot data suggests that a meaningful analysis of participants’ language use requires this instruction. Some initial linguistic data was lost or rendered unintelligible by participants talking simultaneously, and by participants moving to other desks in order to talk to their team members, where they could indicate their searches, etc. by physically pointing to their laptop screen. Also, the silent office may add to the simulation’s realism in that employees might email a colleague in another department rather than walk to their desk, if those departments are in different locations within an office. To add to the realism of the simulation, participants were asked to sign up for the study in groups of two or more. Pre-existing friendship groups amongst participants may add to the realism of a working environment, in which colleagues have friends within their organisation who are not necessarily on their working team or in their department, and to encourage inter-team communication. The data collected from the Simulation consists of emails, audio- and dvd-recordings of both the Office and Meeting rooms, text messages, mobile phone calls (sender, receiver and duration only) and computer use (keystroke data, database searched, password attempts, internet sites visited, etc.) (see below). The insider threat Participants self-selected the opportunity to act as an insider when they chose their role at random at the beginning of the game (a maximum of 6 per game). During the game participants were approached outside the offices by a member of the research team, and asked if they would like to take part in this additional aspect of the research (to incentivise, potential insiders were offered additional financial rewards for each piece of insider information collected). Participants were informed at this time that if they wished to assist in testing the security of the information and the system they were working with, they must also try to conceal their insider activities (from their team and from an investigator).

-275-

The tactical interview Immediately after the simulation, participants were informed that there had been a breach of security, and that their interview the following day would consist of questions about what they did during the simulation. Participants were interviewed within a set time of completing the simulation—24 hours, 48 hours, or 96 hours. Interviewers were tasked with making veracity judgements, rather than simply gathering as much information as possible from witnesses and suspects, since this is representative of the majority of investigative responses to insider threat; if the insider threat does not constitute a criminal offence, it is employers or whomever is investigating the event who must decide upon (the extent of) a suspect’s guilt and the suitable course of action. In the morning of the day following the simulation, interviewers were presented with the printout from the DIT software, along with a list of all gang member names, as a ‘case file’. Interviewers were familiar with the simulation, but had no knowledge of the crime narratives, the data contained in the databases, or the way in which participants behaved during the simulation (other than that recorded by the software). For the round of interviews following an initial pilot study, interviewers were two experienced police interviewers, and in the interviews following the second pilot study, the interviewers were one experienced police interviewer and one novice interviewer. Interviews lasted for a maximum of thirty minutes. At the start of each interview, after an explanation phase, interviewees were asked a set of ‘standardised’ questions with reference to ‘case file’ electronic footprint information. Following this the interviewers moved seamlessly into a tactical phase, using a ‘drip feed approach to the revelation of information items to maximise opportunities to detect verbal deception’ (Dando & Bull, 2011: 190). The initial set of ‘standardised’ questions were generally closed in nature (see Appendix A for interview format), seemingly in contrast to the best practice advocated within investigative interviewing. However, although interviews began with WH- questions functioning as closed questions, there were no leading questions, no multiple questions, and no opinion/statement questions (cf. Oxburgh, Myklebust and Grant, 2010). Questions such as ‘What was your role?’, ‘What databases did that role allow you to access?’ and, ‘What databases did you access?’ all provided truth-telling interviewees with opportunities to provide a detailed and truthful account. However, for deceptive interviewees (those involved in an insider event) this approach reduced the opportunities to reveal as little information as possible, and placed them in the position of ‘having’ to deceive in order not to implicate themselves in wrongdoing in the first instance (Bull & Dando, 2010). Following the tactical questioning phase, a free recall phase was introduced using an open invitation question, although this was still structured chronologically by the interviewer, from Round 1 to Round 4. This chronological ordering enabled interviewers to compare responses against their case files, and to note any inconsistencies between interviewees’ accounts and the evidence presented in the case file which they could follow up with further questions. Pilot data suggest that it is informational content, rather than paralinguistic features, that are most salient in the screening and prioritising of potential suspects in this context, and that this process could be achieved via a computer-mediated investigative interview. Results In pilot studies, all participants who were offered the role of insider decided to undertake the additional tasks. All insiders were able to complete the first task (introduced in Round 2), and most were able to complete the second and third tasks, with all participants attempting all -276-

tasks, and returning at least some ‘illegal’ information from their searches. In future iterations of DIT, the insider task may involve another level of threat—that of an employee deliberately feeding misinformation into the databases to achieve a particular goal. For example, an information storage and analysis system such as VICAP or ViCLAS (used to store sensitive information in the US and UK, respectively) might not connect offences committed by the same offender if information was deleted, or misinformation entered, resulting in a criminal evading detection or even legal practices and policies being implemented that are based on such misinformation. Currently, DIT includes a shared database that some participants may update with any information they choose, as part of the routine tasks for their role, but currently the insider threat does not include the feeding of misinformation into the simulation via this database. It is hoped that DIT will facilitate the development of software that will provide an electronic footprint that may be able to flag up potentially anomalous activity. During the pilot phase of data collection, every participant was interviewed the day after the simulation. Two pilot studies involving 12 participants meant that 24 interviews took place. In the first round of interviews, carried out by two experienced police interviewers, interviewers identified insiders and non-insiders with 100% accuracy. In the second round of pilot data collection, interviews were carried out by one experienced police interviewer and one novice interviewer. In this round, the police interviewer again achieved 100% accuracy and the novice interviewer achieved 100% accuracy for identification of non-insiders, but misidentified one insider as a non-insider (although the interviewer would have recommended a second interview with this participant). Accounting for this unusually high detection rate is one objective to be addressed by further testing which, additionally, will include interviews carried out after longer post-event delays (of 48 hours and 96-hours). After the first round of pilot data collection, interviewers observed that a great deal of the information on which they had based their decision was the informational content, rather than the non-verbal and paralinguistic cues often associated with deception. This ‘trade-off’ of non-verbal and paralinguistic data for speed and scope of interviewing suggested the possibility of a written, self-administered interview tool. The SAI is described by its creators as ‘a powerful investigative tool that can be used to elicit comprehensive initial statements from witnesses, quickly and efficiently. It takes the form of a standardised protocol of clear instructions and questions that enable witnesses to provide their own statement’ (Gabbert, Hope & Fisher, 2009). They suggest the following benefits of the SAI to an investigation:      

Facilitates an investigation by obtaining information quickly—at the scene or soon after—investigators do not always have the resources in terms of time, expertise or personnel (particularly when there are many witnesses) Elicits significantly more accurate and detailed information than a free-recall request Strengthens witness memory, meaning that witnesses are protected against forgetting, and against exposure to potentially distorting post-event information Enables investigators to ‘prioritise’ witnesses for follow-up interview, meaning that key witnesses are less likely to be overlooked and that resources can be appropriately allocated The use of a standardised tool to elicit witness accounts limits any suggestion of ‘improper’ interviewing techniques and reduces any pressure from police or other interviewers which might be inadvertently perceived by witnesses Suitable for use in any environment where a statement is required (Gabbert, Hope & Fisher, 2009)

We would argue that there are additional benefits of the SAI that have not been explicitly noted by its designers: -277-

   

 

Reduces the likelihood of sociolinguistic differences and/or ideological prejudices affecting interviewer/interviewee behaviour Reduces the interviewer/interviewee power imbalance which may reduce stress for witnesses providing statements, while maintaining the power BEHIND the instructions of a written document Reduces the potential for power IN discourse through verbal and non-verbal interviewer behaviours Reduces the questionability of authorship in the statement-taking process—provides an example text of all and only the interviewee’s words (although the suggestibility of question-wording is not discounted), or at least a clear record of the Q-A sequence which produced the answers (cf, for example, Coulthard , 2007; Rock, 2001) This original account may benefit any subsequent trial process if it ‘supersedes’, or at least provides reference points for, later accounts—as we know, lawyers may take advantage of small inconsistencies between accounts (for example, Matoesian (2001)) Suitable for deaf and otherwise hearing-impaired witnesses (not including questions about what a witness heard) and witnesses with speech production problems.

Such an interview, we would argue, could be conducted via a computer, with a programme of questions being put to a witness, or many witnesses simultaneously. This interview, while perhaps not providing enough information to determine the veracity of responses, could at least be compared with both the individual workflow information recorded, and the expected responses for a person in that particular role, for example a witness stating their role, and then that they had access to a database which is extraneous to that role. This comparison could quickly highlight anomalies between recorded and reported behaviour which might be useful in prioritising which witnesses, out of the many likely to be interviewed in a secure organisation, might pose the greatest risk. In addition to the benefits of a written text-interview argued above, we would suggest the following benefits of a computerised Self-Administered Veracity Questionnaire (SAVQ): 

      

Can adapt to the interviewee’s responses via automatic text-analysis, in terms of producing witness-compatible questioning (Dando & Milne, 2009: 14) or, for example, looking for expected lexical items or strings within answers (see Appendix B) Can be administered by a non-expert, for example employers or HR departments, without the immediate threat of non-anonymity that staff might feel in divulging information via hand-written statements Can simultaneously record non-verbal behaviour via digital recorder or webcam— facial expressions, bodily movement, etc, along with the length of time taken to answer a question Can specify the type of answer required—Y/N, multiple choice, lengthy text—and can specify required answer length Records and stores written language, leaving an electronic interview record without the need for costly and time consuming transcription Can be used to develop a ‘baseline’ of typical linguistic and/or physical behaviour Software already exists that administers computer-mediated interviewing, and analyses computer-mediated interview data, including basic text analysis (e.g. SurveyMonkey) Interviewees are not permitted to change their answers once they have been submitted, but the opportunity to make any changes or convey uncertainty are contained in a final

-278-



question; any changes or areas of uncertainty can therefore be noted immediately and before any statement-taking process (cf. Rock, 2001) Can quickly be available in other languages (although of course translation is not a ‘neutral’ process) and in other formats (for example ‘large print’ font size)

Discussion (of SAVQ) and conclusions As with all laboratory research, there are of course limitations, and none more so then when developing novel methods. Firstly, the SAVQ is dependent upon an appropriate level of computer skills, which would arguably have to be measured in a standard way in order for the information gathered to be ‘admissible’ within a legal system. However, in its intended use in an office-type setting to aid employers who need information from employees, the SAVQ is likely to be simple enough for all to use. For interviewees with sight impairment, it is likely that the SAVQ might be administrable using a computerised audio format, a development which future research could investigate. Second, to maintain the advantage of many witnesses and/or suspects providing information simultaneously, the SAVQ requires access to multiple computers. Again, in an office environment this is unlikely to be problematic. Indeed, that interviewees are able to complete the SAVQ at their own desk may reduce the probable stress associated with an insider event, for innocent/truthful interviewees. This in turn may help to maintain employeremployee relations, which might otherwise be threatened by a series of needless face-to-face interviews; witnesses may feel less singled-out or threatened if they are undertaking a SAVQ in a familiar setting and can observe that all of their colleagues are also doing so. One drawback of the SAVQ that is not alleviated by an office setting is that paralinguistic interview data is lost. However, this is also true of the SAI, and of any face-toface interview that is not audio- or video-recorded. Equally, although designed to comply with the current Home Office investigative interview model, the SAVQ may be less useful in gathering information about a task or activity that does not have a ‘standard’. For example, if not enough ‘testable’ questions could be generated to capture the information required, or if the questions cannot be organised into a meaningful sequence, the SAVQ may be less advantageous. We consider, for example, a crime—the law sets out condition/s that must be met for an action or series of actions to constitute that crime, but there will also be a set of circumstances that cannot be predicted, and which the law must be flexible enough to encompass. If one cannot anticipate the questions needed, the SAVQ may still be effective, though, in using the more typical open-question formats and aspects of the Cognitive Interview (Fisher & Geiselman, 1992) such as those used in the SAI. Whether the printed ‘transcript’ of a computer-mediated interview could ever take the place of a police statement in court is uncertain, but since there is clear intertextuality between the police interview, the SAI, and the SAVQ, it would seem logical to consider the possibility of this process as circular rather than linear, and that at some point in the future the Self-Administered Veracity Questionnaire could have a place in police procedure.

-279-

Appendix A - Interview format 1) Explanation Phase 2) Closed question phase i. What team did you play on? ii. What role did you play in the team? iii. What tasks did your role entail? iv. What databases did your role allow you to access? v. What type of information is on that database? vi. Did you gain access or attempt to gain access to another database? vii. Did you use your mobile phone? viii. Did you use your mobile phone to make calls? ix. Did you use your mobile phone to text? x. Did you use your mobile phone to contact anybody outside your team? xi. Did anybody outside your team contact you by phone? xii. Did you email anybody? xiii. Did you email anybody outside your team? xiv. Did you receive an email from anybody outside your team? xv. Did you visit the meeting room? xvi. Did you speak to anybody in the meeting room? xvii. Did you speak to anybody outside your team in the meeting room? xviii. Did you visit the printer? xix. Did you print anything? xx. Did you speak to anybody in the printer room? Etc. 3) Tactical phase 4) Free recall phase (including follow-up questions) i. Describe what happened in Round 1 ii. Describe what happened in Round 2 iii. Describe what happened in Round 3 iv. Describe what happened in Round 4 5) Closing

-280-

Appendix B - Example of question adaptation in SAVQ ‘Describe what happened in Round 1’ Interviewee describes in detail

No names mentioned ‘Do you remember the name of the gang you investigated?’

Yes, interviewee remembers

No, interviewee doesn’t remember

‘What was the name of the gang you investigated?’

‘Do you remember the names of any of the gang’s members?’ Yes, interviewee remembers

No, interviewee doesn’t remember

'What are the names you remember?'

References Carnegie Mellon University (2011) Detecting insider threat requires a multi-faceted approach. Available at: http://www.cert.org/insider_threat/docs/workshop.pdf [Accessed 31/08/2011]. Chapman, V., Appleyard, S., Christie, M., D’Silva, C. & Glanville, H. (2007) Indicators of the ‘disaffected employee’ type of insider threat QinetiQ Ltd, UK. Clark, R., Dando, C.J., Gabbert, F. & Hope, L. (2011) Toward assisting older witnesses and victims to give best evidence. Paper presented at the 4th International Investigative Interviewing Group, Dundee. Coulthard, M. (2007) An Introduction to Forensic Linguistics: Language in evidence. London: Routledge. Dando, C.J., & Bull, R. (2011) Maximising opportunities to detect verbal deception: training police officers to interview tactically. Journal of Investigative Psychology and Offender Profiling 8(2): 189—202. Dando, C. J. & Milne, R. (2009) The Cognitive Interview. In R. N. Kocsis (ed.) Applied Criminal Psychology: A Guide to Forensic Behavioural Sciences. Sydney: Charles Thomas. Fisher, R. & Geiselman, R. (1992). Memory-Enhancing Techniques for Investigative Interviewing: The cognitive interview. Springfield: Charles Thomas. Gabbert, F., Hope, L. and Fisher, R. (2009) The self-administered interview. Available at: http://www.selfadministeredinterview.com [Accessed 31/08/2011]. -281-

Gabbert, F., Hope, L., & Fisher, R. (in press) Protecting eyewitness evidence: examining the efficacy of a self-administered interview tool. Law and Human Behavior. Hoffman, B., Meyer, C., Schwarz. B. & Duncan, J. (1990) Insider Crime: The Threat to Nuclear Facilities and Programmes. For the US Department of Energy Santa Monica, CA: RAND. Keeney, M., Cappelli, D., Kowalski, E., Moore, A., Shimehall, T., and Rogers, S. (2005) Computer System Sabotage in Critical Infrastructure Sectors. Pittsburgh, PA: Carnegie Mellon. Matoesian, G. (2001) Law and the Language of Identity: Discourse in the William Kennedy Smith Rape Trial. New York: Oxford University Press. Miller, J. (2003) Police Corruption in England and Wales: An Assessment of Current Evidence. Home Office Online Report. Oxburgh, G.E, Myklebust, T., & Grant, T. (2010) The question of question types in police interviews: a review of the literature from a psychological and linguistic perspective. The International Journal of Speech, Language and the Law 17(1):45— 66. Poyser, S. & Milne, B. (2011) Miscarriages of justice: a call for continued research focussing on reforming the investigative process. British Journal of Forensic Practice 13(2): 61—71. Rock, F. (2001) The genesis of a witness statement. Forensic Linguistics 8(2):44—72.

-282-

Imaginative leaps in trademark law Ronald R. Butters Duke University, USA [email protected] Abstract In American courts, linguists are often called upon to testify about whether trademarks in litigation are best classified lexicosemantically as ‘suggestive’ or ‘descriptive’ (only the former category is afforded automatic legal protection). Legally, the central issue is imaginativeness: suggestive terms differ from merely descriptive ones in that they require a significant ‘cognitive’ or ‘imaginative’ ‘leap’ to connect the name with the referent. But imaginativeness is a seldom-addressed concept in linguistic science, and has been little discussed in the forensic linguistic literature. I suggest that precision in formulating the forensic linguistic concept of ‘imaginativeness’ lies in viewing putatively suggestive marks within the framework of linguistic notions of figurative language and phrasal anomalousness, drawing also upon cognitive linguistic concepts of ‘semantic leaps.’ Keywords:

DESCRIPTIVE TRADEMARKS; LEXICOSEMANTICS; SEMANTIC LEAPS; SUGGESTIVE MARKERS; TRADEMARKS

In determining intellectual property rights, trademarks are judged according to their ‘strength’, that is, where they can best be placed on a continuum of lexicosemantic categories from weakest to strongest. There are two categories of ‘weak’ marks: generic, that is, the name of the kind of thing rather than the thing itself; for example, package delivery service, bread descriptive, that is, terms that allude to some intrinsic and/or laudatory quality of the product or service; for example, speedy delivery service, tasty bread. The addition of geographical terms, general terms of praise (known as LAUDATORY terms), and words that name basic ingredients may sometimes be adjudged to rescue a descriptive mark from being generic; however, the law views such terms as likely to be merely descriptive themselves. There are three categories of ‘strong’ marks: suggestive, that is, terms with only an oblique semantic or pragmatic connection with the product or service; for example, Federal Express, Panera Bread arbitrary, that is, terms with no apparent semantic or pragmatic connection with the product or service; for example, Rainbow Delivery Service, Sunbeam Bread fanciful, made-up or coined terms; for example, Purlator Delivery Service, Bimbo Bread Unlike the weak marks, suggestive marks are considered ‘imaginative’ in that speakers of English must make some mental effort to figure out the semantic connection between the mark and the service or product so named. Such an act of mental effort is described by the legal term of art, imaginative leap. In general, only suggestive, arbitrary, or fanciful marks have full legal protection from infringement by competitors. (However, even if a trademark is fanciful or arbitrary, it may be -283-

denied protection if it is adjudged to be obscene, scandalous, or disparaging). In contrast, those determined to be generic have no legal protection, and those determined to be descriptive are accorded protection only if they have been officially registered for a period of years and/or are adjudged to be famous enough to have acquired what the law terms secondary meaning—that is, a strong independent semantic association of the mark and the particular company that offers the product and service. Fame is a legal term of art used to describe marks that have acquired secondary meaning. For example, Post Office is arguably a descriptive mark, but in the United States it remains a protected, registered trademark of the United States Postal Service, owing to the secondary meaning that results from the public’s longstanding strong association of the term with the owner. Likewise, assuming hypothetically that the following marks have not been registered long enough to be accorded protection by that criterion, Best Foods, International Screw, Fluffy Bread, American Tool and Die, and United Parcel Service are all descriptive terms that will have legal protection only if it can be demonstrated that they are sufficiently famous among the relevant public— i.e., that the relevant public associates the marks with particular companies. As long as it remains legally live (the opposite of being legally dead, the term of art that is used to describe the state of a trademark that is no longer in commercial use), a legally protected trademark not only cannot be used by others in the same line of business, it also cannot normally be rescinded by legal action; if somebody else uses the mark in connection with products or services that are similar to the mark’s owner’s, the owner can sue the infringer for damages. However, alleged infringers sometimes mount a defense in which they allege that the mark does not actually belong in one of the three strongest categories of marks, but is really just descriptive (without secondary meaning) or even generic. If the defense is successful, the owner’s mark is then subject to legal cancellation, which generally means that anyone can use it. When an alleged infringer’s defense is based on the putative genericness or descriptiveness of the owner’s mark, the issue of the imaginative leap thus becomes critical. That is, the mark’s owner will seek to demonstrate that the average potential purchaser of the goods or services in question will not instantly recognize the semantic link between the mark and the goods or services, but must instead perform some kind of a mental analysis—an ‘imaginative leap’—to see the connection that relates the term and the referent. Linguists who are professionally qualified to opine on issues of lexis and semantics are therefore often retained to testify about aspects of a particular mark’s meaning that may be relevant to determining whether the mark is descriptive or suggestive—that is, whether or not an ‘imaginative leap’ is required. Of course, the linguist does not testify as to the ultimate legal determination of which category the mark legally belongs in; that is up to the jury or judge. Rather, the linguist testifies as to what the meaning of the mark is, including most particularly the nature of the meaning connection between the mark and the putative commercial referent. I have testified, for example, about linguistic aspects of the meaning of the term steakburger (as applied to fast-food sandwiches) that bear on the issue of whether it is a generic, descriptive, or imaginative mark, but it was up to the judge to decide if there were considerations other than the linguistic which had bearing on their final decision.1 As noted in Butters (2008a; 2008b), the linguistic concept of genericness, at least as used by lexicographers, is close enough to the legal definition that there is generally little problem for the linguist who testifies in genericness cases (and little else for the jury to consider). A problem arises, however, in the match-up between the legal and linguistic 1

The Steak N Shake Company, SNSTM, Inc., and Steak N Shake Operations, Inc., v. The Burger King Corporation and Burger King Brands, Inc., United States District Court for the Eastern District of Missouri, Civil Action No.4:2004-CV-00525-CDP, JULY 7, 2004.

-284-

concepts of categories of descriptiveness and suggestiveness, which are not terms of art at all in linguistics (at least not in the relevant sense). This problem, I believe, generally gets too little noticed in forensic linguistic theory. Shuy (2002) makes a good beginning, noting that there are four related legal tests for determining the difference between a suggestive mark and a merely descriptive one, ‘the most commonly used’ of which is ‘the imagination test’ in which the more imaginativeness that is required by the consumer to get some direct description of the product from the mark, the more likely the term is to be considered suggestive. … If the purchaser has to make use of mature thought and engage in a multistage reasoning process to connect the mark to the product or service, the term is judged to be suggestive. (p. 37—38) Shuy rightly characterizes this as a ‘cognitive leap beyond the descriptive category’—a ‘leap of imagination’; ‘… if the … [understanding of the meaning is] not instantaneous, then the word is suggestive’ (p. 65). However, Shuy’s otherwise exemplary introduction does not take up the issue of the possible linguistic correlates of the ‘leap of imagination,’ which itself is not a concept that linguistic science has usually addressed. Rather, he reports, he gave legal testimony that begins to develop the notion of ‘leap of imagination’ within the general framework of ‘metaphor.’ In his view, the use of the trademark Long Life when applied to antifreeze (or garments such as ‘sox’ [sic]) is suggestive, not descriptive, because ‘long life’ refer[s] to animate existence, except in its metaphorical uses [in which it means ‘extended life’]; which, being metaphorical, could in no way be considered descriptive. … [Moreover,] ‘extended life’ is not commonly used in reference to animate existence. … [Although that] some elements of the antifreeze/coolant industry misused the term ‘long life’ … is evident, … [that] does not render ‘long life’ descriptive or generic. It remains metaphoric. (p. 67) Aside from the oddly unscientific labeling of a usage as ‘misused,’ Shuy’s rationale here seems manifestly vulnerable, considering that established dictionaries give definitions of both long and life that indicate that the two are totally interpretable according to their ordinary usages. Here, for example, is the relevant portion of the definition for long found in the New Oxford American Dictionary (2001): lasting or taking a great amount of time: a long and distinguished career | she took a long time to dress relatively great in extent: write a long report | a long list of candidates And here is the relevant portion of the same dictionary’s definition of life: the period during which something inanimate or abstract continues to exist, function, or be valid: underlay helps to prolong the life of a carpet Thus bringing together the two words long and life in the context of ‘antifreeze’ requires no imaginative reading at all: the result is simply the construal of the two words according to ordinary lexicalized meanings that speakers will have recognized without thinking about it further. Indeed, ‘long life’ is being used to describe antifreeze in a laudatory way, which in

-285-

itself could be taken as arguing that the term is descriptive, since, as noted earlier, laudatory words are taken to be descriptive rather than suggestive. It may well be that the etymology of life in the relevant contemporary sense originally came about through a process of metaphorical extension; that is, long-life batteries may be a dead metaphor. But a dead metaphor underlying long-life batteries does not require any greater ‘imaginative leap’ than the New Oxford American Dictionary’s example of the ‘long life of carpet’. It would be unfair to conclude that Shuy does not acknowledge the essential irrelevance of dead metaphors to synchronic analysis, much less that Shuy reverted to the rhetorical strategy of framing an explanation in terms of a ‘metaphorical’ reading here because a direct confrontation of the criterion of the imaginative leap would not yield the results he wanted to obtain. That is, it seems unlikely that one could convince a jury that the interpretation of Long-Life Antifreeze would require that, for speakers of the English language who are alive today, ‘the purchaser … [must] make use of mature thought and engage in a multistage reasoning process to connect the mark to the product’ antifreeze any more than he or she would need to do so to connect long-life with carpet (or batteries, or suntan lotion). A more likely reason why Shuy took the approach he did was that, in framing his argument as grounded in the concept of metaphor, as opposed to ‘cognitive leaps’ or ‘leaps of the imagination,’ he was on more familiar linguistic grounds. The terminology of tropes may arguably be more within the purview of literary studies than linguistic, yet it nonetheless bleeds over into linguistics as a part of semantic and pragmatic interpretation, and has been widely used in cognitive linguistics, beginning with Lakoff and Johnson (1980). There is also historically a considerable body of scholarship on the interpretation of compounding and socalled ‘deviant utterances’ (for example, see Butters, 1969 and the references therein).2 Although degree-of-imaginativeness has been a seldom-addressed concept in linguistic science, I believe that we can sharpen the forensic linguistic definition of suggestive marks by pursuing the concept of ‘imaginative leap’ within the framework of relatively recent cognitive linguistic research that focuses on figurative language and phrasal anomalousness. Coulson (2001: 2) defines semantic leaps as resulting in ‘natural language constructions that yield nonobvious meanings’; the figurative meanings become clear only after one expends significant conscious thought upon the constructions. Coulson summarizes this turn in theorizing about figurative language within cognitive linguistics as follows: [C]ognitive semanticists … have argued that metaphor is, in fact, a pervasive phenomenon in everyday language and, moreover, that it represents the output of a cognitive process whereby we understand one domain in terms of another. Cognitive linguists define metaphor as reference to one domain (known as the target, theme, or base domain) with vocabulary more commonly associated with another domain (known as the source domain, phoros, or vehicle). On this construal, metaphoric language is the manifestation of conceptual structure organized by a cross-domain mapping: a systematic set of correspondences between the source and the target that result from mapping frames or cognitive models across domains.

2

Note that a standard work of reference in linguistics such as Crystal (2003) does not define cognitive leap or imaginative but does contain entries for deviance and cognitive metaphor (but in the sense of Lakoff and Johnson 1980).

-286-

On this view, … a speaker invokes a metaphor whenever she refers to one domain, such as verbal argumentation, with vocabulary from another domain, such as physical combat3. (p. 162) Colson notes, for example (p. 167), that for a hearer understanding for the first time the wellknown recently coined compound, computer virus, the target domain (‘a computer program written for the express purpose of damaging other peoples’ computational resources’) is conceptualized with source domain models (biological viruses). Although ‘the two domains share structure and vocabulary’ (e.g., both are ‘invasive’; both ‘attack’ ‘hosts’; both are ‘unwanted’, both ‘replicate’, both typically result in diminished capacity for the host) they also are significantly different in readily identifiable ways: in the end, ‘a computer virus is a very different kind of entity from a biological virus.’ One advantage to referring to the work of Coulson and other cognitive psychologists in thinking about what constitute suggestive marks from a linguistic perspective is simply that it ties an important legal concept directly into a linguistic one. Courts are generally less comfortable, and rightfully so, when experts speak of legal concepts in ways that do not seem immediately relevant to legal concepts. More important, however, is that examining trademark strength from the perspective of the imaginative leap generates a kind of analysis the results of which offer greater scientific clarity. By way of illustration, I will briefly outline arguments that emerged in a case in which I was a paid forensic linguistic consultant (the type of industry involved has been changed here for purposes of anonymity). The case involved whether or not XX and XY are descriptive or suggestive when used as trademarks applied to sports clothing and equipment for men and women (respectively). One can look at dictionaries and find generic definitions for shoes, hats and the like, and even XX ‘symbol for female chromosome’ and XY ‘symbol for male chromosome’, and conclude that the use of the terms as applied to garments would be ‘metaphorical’ rather than ‘literal,’ given that clothing does not directly involve chromosomes—and hence that the marks XX BOWLING BALL and XY TENNIS SHOES are suggestive rather than descriptive. However, the dictionary definitions really don’t inform us very well about the collocational possibilities and interpretations of XX BOWLING BALL or XY TENNIS SHOES. Even if we assume for the sake of argument that XX and XY in today’s English centrally and lexically mean, for a number of speakers who are likely buy sports clothing and equipment, ‘symbol for female chromosome’ and ‘symbol for male chromosome’, it would still arguably be the case that the only-slightly-more-general meanings ‘female’ for XX and ‘male’ for XY are so obvious that the speakers will not require significant thought in order to interpret them in those senses, especially in contexts such as the terms used for products that normally are designed with the sex of the target customer very much in mind. Rather, one might argue that it is much more in accordance with psycholinguistic reality that such speakers will—without giving it much thought at all—recognize that XX and XY in the various brand names and discourse uses fit well within the understandable semantic boundaries of XX and XY. That is to say, we can in this way conclude that the necessity for an imaginative leap is not

3

Although cognitive linguists often stress the differences between their views of metaphor and more traditional views, in fact, at least in terms of discussing individual metaphorical constructs, the differences are not necessarily great. See, for example, Harmon and Harmon, (2000, s.v. metaphor): ‘An analogy identifying one object with another and ascribing to the first object one or more qualities of the second. I. A. Richards’ distinction between the tenor and the vehicle of a metaphor may be useful. The tenor is the idea being expressed or the subject of comparison; the vehicle is image by which the idea is conveyed or the subject communicated.’

-287-

particularly great, and that XX BOWLING BALL or XY TENNIS SHOES are essentially descriptive rather than suggestive. Semanticists agree also that ordinary words co-exist with semantic border areas that are somewhat provisional—and overlapping with their neighbors’. Indeed, the meanings of words are characteristically so ‘fuzzy’ at the edges that fuzzy has become a well-known term of art for psycholinguists. Aitchison (1994) describes the nature of word meaning for normal speakers of a human language as follows: Words cannot be assigned a firm meaning … ‘natural language concepts have vague boundaries and fuzzy edges’ [quoting Lakoff, 1972]. Word meanings cannot be pinned down, as if they were dead insects. Instead, they flutter around elusively like live butterflies (p. 39—40). … [Word meanings] have fuzzy edges in the sense that there is often no clear point at which one word ends and another [in the same family of words] begins (p. 46). … [The majority of words have] one or more of the following problems: first, it may be hard to specify a hard core meaning at all. Second, it may be impossible to tell where ‘true meaning’ ends and encyclopedic knowledge begins. Third, the words may have ‘fuzzy boundaries’ in that there may be no clear point at which the meaning of one word ends and another begins. Fourth, a single word may apply to a ‘family’ of items which all overlap in meaning but do not share any one common characteristic (p. 48—49). The sociolinguist William Labov illustrated the underlying principle involved here with words in the general category of containers (1973: 340): In any kitchen, there are many containers that are obviously bowls, cups, mugs, and dishes. But there are others that might be called cups or might not; or might be a kind of cup, according to some, but a kind of bowl, according to others. Thus the mental organization of words and meanings is not a set of sharply delineated entities in which each word has a unique set of defining elements all and only of which must be present in a real-world entity to qualify that entity to be legitimately named by that word. The connection between a brand name XY and the meaning ‘for men’ is an easy one because the semantic boundaries of the term XY chromosome are, like the boundaries for words in general, somewhat fuzzy. In other cases involving the difference between descriptive and suggestive marks, the idea of ‘metaphor’ seems of little help, yet the idea of ‘imaginative leap’ seems at least somewhat illuminating. Suppose, for example, that a company calls itself National Southern Trucking Company.4 Is this a descriptive or a suggestive mark? Certainly, trucking and company are generic, and will afford no protection. And national and southern are in the law generally considered descriptive by virtue of falling into the descriptive class of general geographical terms. Only if the term National Southern Trucking Company requires some kind of metaphorical interpretation or cognitive leap will it satisfy the criteria for inclusion in the class of suggestive marks. However, while National Southern is not what one would usually describe as a metaphor, if one focuses on the apparent meaning of National Southern, one sees something of a semantic incongruity: while Southern National can readily be understood to mean ‘the southern part of the nation’, the parallel semantic interpretation of National Southern—‘the national part of the South’—does not make much sense. A potential 4

Again, this example refers to issues about which I have consulted with attorneys; however, I am changing the names and the industries so as to preserve the anonymity of the proceedings.

-288-

customer of National Southern Trucking Company must give it some thought—must engage the imagination—and impose some kind of interpretation: perhaps it really means ‘southern national’, making use of a deviant syntax; or perhaps it means ‘characteristic of the South but extended throughout the nation’; or perhaps it means ‘national, but with especial emphasis on the South’. In short, the potential customer must make some kind of imaginative/cognitive leap to interpret the semantically anomalous phrase. This is not to say that linguists ought not to introduce the concept of metaphor at all in discussing the intersection of genericness and descriptiveness. In many cases, the absence of a clear dictionary definition for a word or phrase is helpful in ascertaining the degree of mental effort that might be necessary to understand it. However, making use of the linguistic concept of the imaginative leap will, I think, become increasingly important in future linguistic consultation in trademark cases. References Aitchison, J. (1994) Words in the Mind: An Introduction to the Mental Lexicon, 2nd edition. Cornwall: Blackwell. Butters, R. R. (1969) On the interpretation of ‘deviant utterances’. Journal of Linguistics, 6: 105—110. Butters, R. R. (2008a) A linguistic look at trademark dilution. Santa Clara Computer & High Technology Law Journal 24(3): 101—113. Butters, R. R. (2008b) Trademarks and other proprietary terms. In J. Gibbons & M. T. Turell (eds) Dimensions of Forensic Linguistics. Amsterdam/Philadelphia: Benjamins. Coulson, S. (2001) Semantic Leaps: Frame-Shifting and Conceptual Blending in Meaning Construction. Cambridge: Cambridge University Press. Crystal, D. (2003) A Dictionary of Linguistics and Phonetics, 5th edition. Padstow, Cornwall: Blackwell. Harmon, W. and C. H. Holman (2000) A Handbook to Literature, 8th edition. New York: Prentice Hall. Labov, W. (1973) The boundaries of words and their meanings. In C.-J. Bailey and R. W. Shuy (eds) New Ways of Analyzing Variation in English. Washington, DC: Center for Applied Linguistics, 340—373. Lakoff, G. (1972) Hedges: a study in meaning criteria and the logic of fuzzy concepts. In Papers from the Eighth Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society. Lakoff, G. and M. Johnson. (1980) Metaphors We Live By, Univ. of Chicago Press. New Oxford American Dictionary (2001) Oxford: Oxford University Press. Shuy, R. (2002). Linguistic Battles in Trademark Disputes. Houndmills: Palgrave.

-289-

Arabic varieties and LADO: how can LADO deal with variance? Judith Rosenhouse Technion Israel Institute of Technology, Haifa, Israel [email protected] Abstract This paper approaches problems of Language Analysis for the Determination of Origin (LADO) associated with Arabic. Arabic-speaking communities are found in the Middle East and North Africa, as well as in immigrant communities in Europe and elsewhere. The discussion consists of four major parts. The first part begins with a background description of Colloquial Arabic (CA) and Literary Arabic (LA). Next, three problem types are raised relating to linguistically mixed speech samples. The areas of Diglossia, Code Switching and Variety Switching are considered as a single psycholinguistic process. Some examples of the complexity involved in CA dialects follow. These difficulties lead us to suggest the development of a computer program which might help in arranging features of recorded Arabic speech samples into linguistic categories and calculating their statistical likelihoods as an aid to assessing the origin of the sample. Finally, consequences for LADO are discussed. Keywords:

LADO; ARABIC DIALECTS; LANGUAGE CONTACT; KOINÈZATION; COMPUTATIONAL LINGUISTICS

1. Introduction This paper examines problems relating to Language Analysis for the Determination of Origin (LADO), using Arabic as a ‘test bed.’ The very concept of LADO and its tasks raises many questions (e.g., Eades, 2009; Fraser, 2009; Muysken et al., 2010: 2—4) but we will not discuss here whether the LADO goal of determining a person’s origin (and nationality) based on linguistic criteria is justified or not. Rather, we intend to pursue the linguistic aspect, by asking how a language expert can be sure that a given person is a native speaker of a certain dialect. The following discussion includes four parts. Firstly, it will provide a background description of Arabic with its dichotomy of Colloquial Arabic (CA) and Literary (LA) Arabic. Secondly, it will focus on LADO problems as viewed from the perspective of the Arabic language expert. These refer mainly to Diglossia, Code Switching and Variety Switching. Section (3) presents examples of shared and different linguistic features in the Arabic dialects, which leads to a suggestion for a computer program in section (4). The paper ends with some conclusions in section (5). Arabic has been expanding from its original home, the Arabian Peninsula, since the 7th century CE. Thus, the major populations of native speakers of Arabic at present, in addition to the Arabian Peninsula, are in the Middle East and North Africa (MENA). In addition, ‘Fringe’ dialects of Arabic are spoken in Central Asia and Sub-Saharan Africa. Recent migrations have created numerous Arabic-speaking communities in the rest of the world—in North and South America, Europe, Australia and Africa (cf. Holes, 2004; Kaye & Rosenhouse, 1997). The Arabic language ranks fifth by number of speakers worldwide with more than 220 million native and 2nd language speakers (Ethnologue, 2009; Lewis, 2009). Since many of the Arabic speaking countries in the Middle East and North Africa are great oil producers, interest in the Arabic countries, Arabic culture, language and dialects seems to have been growing all over the world. Arabic has a rich and prestigious written literature, which began to flourish around the 7th century CE. In addition, Arabic is the most prevalent spoken language in the Middle East and North Africa, and due to -290-

the number of Arabic speakers and their geographical dispersion Arabic has many spoken dialects. Thus, the language is known for its two basic registers: the ‘low’ (basilect) vernacular, i.e., the spoken informal Colloquial Arabic (CA), and the ‘high’ (acrolect) Literary Arabic which refers to the written, literary modern standard variety (LA).1 The colloquial variety is the speakers’ mother tongue, whereas the literary variety has to be studied at school and is not a mother tongue. Yet LA is considered the Arabs’ prestigious language mainly because the Koran, the holy book of the Muslims, is written in LA, whereas the colloquial Arabic varieties are (still) often considered a debased language form in comparison. This attitude is partly due to the natural linguistic development of the Arabic language through the centuries, which has yielded many differences between CA varieties and LA on all linguistic domains (at various rates). This dichotomy has become known in the linguistic world as Diglossia, this term having been coined by Krumbacher (1902) for the situation of Ancient vs. Modern Greek. Later on, Marçais (1930) applied it to Arabic, but only Ferguson (1959a) drew general linguistic attention to this inherent dual nature of Arabic (and of other languages with a similar dichotomy). To understand the case of Arabic in relation to LADO, a short survey of its dialects is in order. Since CA dialects are numerous, they are classified by several categories. First, they are classified geographically into Eastern and Western dialects, with the Egyptian dialects as the transitional area between the two groups. They are also classified demographically, into sedentary urban (towns) and rural (villages) dialects on the one hand, and into nomadic tribal dialects on the other hand (Fischer and Jastrow, 1980; Kaye and Rosenhouse, 1997)2. In addition to geography and demography, other factors that affect Colloquial Arabic varieties are mainly social status, religion, sex, and education (see Kaye and Rosenhouse, 1997). Social status and religion are often combined in the MENA, where Muslims tend to become the majority and politically influential layer, while Christians and adherents to other faiths (Druze, Jews, etc.) constitute minorities in these countries3. The social and cultural effects of male—female differences in the MENA also affect linguistic features, forming subvarieties within local dialects (see Rosenhouse, 1998). Education, which has been increasingly spreading in the MENA in the 20th century, is particularly important, because education enhances speakers’ proficiency in LA and affects their daily communication in CA (cf. Al-Wer, 2002), thus modifying the structure of Arabic Diglossia.4 Additional, external causes have affected Arabic in the MENA in the 20th century. These include the political and cultural changes which were mainly due to strong Western influence, technological progress, and general globalization processes (Holes, 2004). These pressures caused the immigration of numerous natives of the MENA to other Arabicspeaking countries (for example, from Egypt to Saudi Arabia, and from Iraq to Jordan) and to non-Arab countries in other continents, mainly because of economical or personal security reasons. In their new environments these immigrants naturally make contacts with new varieties of Arabic and/or foreign languages. Such language contacts expand their linguistic scope which is discussed below with reference to particular Arabic-speaking immigrants.

1

Literary Arabic (LA) is also known as Modern Standard Arabic, but we use here LA to distinguish it from the colloquial, spoken dialects. 2 In the 20th century many nomadic dialects have been changing (urbanizing) due to sedentarization. Such developments are not unique to this period, or this language, of course. 3 This is not the case in Israel, where Arabic is a minority language. 4 Note that in parallel, CA simultaneously affects LA being the speakers’ mother tongue.

-291-

2. Problems with Arabic in LADO: diglossia, code switching and variety switching Many immigrants become LADO clients in their new locations. Although the very concept of LADO and its tasks is questionable (Muysken et al., 2010: 2—4) we will not be concerned with this here. I intend to focus only on the aspect of the language expert’s certainty that a given person is or is not a native speaker of a certain dialect, when this opinion depends only on the analysis of a single recorded sample of that speaker’s speech. The discussion reflects professional experience in this area. The recordings discussed were usually around 30 minutes long. Most of them involved questions put to the immigrant by a governmental representative or a lawyer, and the immigrant’s answers. In such cases the governmental representative was aided by a translator, to ensure that the immigrant understood the questions and that the governmental representative understood the immigrant’s answer (given in his/her claimed native language/ dialect). In a few cases the LADO client spoke a spontaneous “monologue”. A both practical and theoretical linguistic problem arose when the recorded speech sample revealed linguistic features of more than one dialect, that is, when the recording revealed features of additional and different language varieties than the claimed one. Such features often included LA or elements of other (non-Arabic) languages. In other cases specific typical forms of the claimed dialect were missing where they could have been used. The question of what to do in such cases is the focus of this paper. Immigrants’ contacts with speakers of new languages usually cause language contact processes such as borrowing, code switching/mixing, erring, accommodation, mother tongue attrition, etc. (Trudgil, 1986; Muysken, 2000). These phenomena are also found in Arabicspeaking immigrants’ speech as described in the literature (e.g., Bentahila and Davies, 1995; Miller, Al-Wer, Caubet and Watson, 2007; Muysken, 2000). It is important to consider the fact that like any other living language, CA and LA keep developing and changing in a natural and unavoidable process, but the CA literature usually describes the structure and vocabulary of dialects in terms of a static picture. The linguistic changes that have been taking place during the 20th century include LA effects on CA and CA elements that penetrate LA (‘reverse diglossia’). Due to growing literacy (cf. Diem, 1974; Meiseles, 1980; Kaye, 1994; Mejdell, 2006) this process ‘lowers’ LA level, and converges LA and CA. Even highly literate writers cannot avoid diglossia, because their L1 is CA and not LA (see Rosenhouse, submitted). The increasing contacts between LA and CA are described in the following excerpt: The phenomenon of the renaissance of the Arabic language must not be construed in the sense of the death and subsequent revival of the [LA] language… but rather as the penetration of a language which has previously existed in written form exclusively into widespread areas of everyday social communication. (Shraybom-Shivtiel, 1995) Shraybom-Shivtiel also shows that CA affects various lexical decisions concerning modern LA vocabulary made by the Egyptian Language Academy, a citadel of LA purism and prestige (and opponent of CA). More about CA/LA contacts in Egypt, for example, can be seen in papers discussing Egyptian Arabic as a written language (Rosenbaum, 2000) and about language mixture of CA and European languages in Egyptian literature (Rosenbaum, 2000—2002). Rosenbaum (2000) begins his article with these words:

-292-

The change of attitude in Egyptian culture toward the colloquial and the change of stylistic norms in the literary system have encouraged the appearance of various kinds of mixed styles, not only in oral performance but in written texts as well. (Rosenbaum 2000: 68) Rosenbaum (2000—2002) describes in his paper a play by the Egyptian playwright Mahmoud Diab in which four languages and varieties are used: CA, LA, English and the rarely-used German and remarks as follows: ‘The German is written in Arabic script with a translation into standard Arabic in parentheses (which may be of help to the reader but hardly solves the problem for the spectator in the theater!)’ (Rosenbaum 2000—2002:24). These few references demonstrate the kinds of linguistic mixtures that exist both in speech and in writing in all the Arabic speaking countries, and not only in Egypt or in immigrants’ communities. The CA—LA diglossic back-and-forth switching involves different registers within a single language. But two additional major processes occur in Arabic-speakers’ speech: Code Switching, which involves mixing two languages (interlingually), and Variety Switching (or Dialect Switching)5, which involves inter-dialect (intra-linguistic) processes. For native Arabic speakers, Variety Switching occurs when they move to a different CA-speaking environment. This occurs when, for example, a rural Egyptian moves from his or her village to Cairo, or when immigrants travel further away to Saudi Arabia. Code Switching occurs when they get in touch with different language communities, whether in the Arabic speaking country, such as in the very frequent case of Moroccan Arabic/French code switching in Morocco (cf. Bentahila and Davies, 1995; Sadiqi, 2003) or elsewhere (Gafaranga, 2007; Mejdell, 2006). Psycholinguistically, these processes are very similar (if not identical—cf. Rosenhouse, submitted). For the linguist whose task is to confirm (or refute) the recorded speaker’s claimed origin or mother tongue for LADO, all these natural linguistic and sociolinguistic processes are problematic factors due to their ‘fluid,’ unstable nature; but they must be taken into consideration. 3. Linguistic problems As noted, one of the LADO tasks related to Arabic refers to Dialect Verification. This implies that a language expert should (be able to) confirm that the recording reflects a certain claimed language community (or not), whatever the speech forms (of various different languages or dialects) found in the recorded speech sample. The issue of Dialect Verification for LADO cases can raise in our opinion at least three interrelated questions: a) How can one decide the speaker’s original dialect based only on a recorded text in which linguistic features of more than one dialect may occur? b) How does one decide which features are more salient than others, if features of several dialects occur in a single recording? (On feature saliency see Hickey, 2002). c) Should the expert form an opinion of the speaker’s original dialect (i.e., mother tongue) by considering only what is in the recording or also by what is not found in it? We will now move on to address each of these questions in turn.

5

Variety switching refers to the linguistic process, whereas dialect verification refers to the expert’s task.

-293-

Question (a) Question (a) relates to the decision that a linguist must take about a recorded speaker’s speech. For this task at least three basic optional approaches may be applied: 1. Counting and classifying the number of all the features of each dialect that occurs in the text and deciding that the original dialect is the one that presents the majority of linguistic features; 2. Assessing the saliency of the individual dialect features, counting them and deciding that the original dialect is the one that presents the majority of the salient features; 3. Combining these methods. The third option appears to be the most preferable. Indeed, these methods can be applied to both intra- and inter-lingual switching. However, one may argue that the counted features may reflect the dominance of some dialect features in the speech sample rather than ascertaining that it is the speaker’s original mother tongue or ‘mother dialect.’ This case may occur if the speaker has been using the claimed dialect/language for years prior to the recording, so that s/he speaks it fluently. If s/he has not spoken that dialect continually, mother tongue attrition features and other language contact features need to be considered. The speaker may also use features of a dialect s/he wants to adopt or a language variety which s/he knows at mother tongue level; this can be, for example, the prestigious Muslim Baghdadi Arabic, even if the speaker is not a native of that city, or if s/he is a native of Baghdad but not Muslim. But language experts are limited to only one recording, and cannot explore such different options. We may conclude that the above methods do not always provide a fully satisfactory answer to Question (a). Question (b) Question (b) relates to feature salience. Here we ask how one decides which features are salient, in light of there being so many linguistic features in each variety, and more than a single dialect is involved in the speech sample. Before proceeding we should note that: 1. A linguistic feature may be salient for some dialects (i.e., characterize them) but not for others; 2. A feature that occurs in several dialects may hardly be salient as a distinctive marker between those dialects. As an aside in this context, let us consider that on the way to the analysis of the formation of new dialects, Hickey (2002) defines some active and negative saliency in dialects and actually presents many salient features in dialects of many languages. Some examples of salient and non-salient CA features (here referred to as shared or distinctive) are presented in Section 3 below. Returning to Question (b) it seems necessary to look into the problem of defining different dialects and their borders, i.e., geographical dialectology and dialectometry (Behnstedt, 2006; Belinkov, 2010). To define a dialect and its borders, dialect atlases are often used in the linguistic studies of many languages. Linguistic atlases may study lexical, phonological, morphological and/or syntactic elements. Examples of such atlases are the atlases of the American English, Dutch or German languages (American Heritage -294-

Dictionaries, 2006; Barbiers and Bennis, 2007; Koenig and Paul, 1978). Arabic dialectology and dialect atlases have made much progress since the second half of the 20th century (e.g. Behnstedt and Woidich, 2005; TAVO, 1978—1992) but altogether, in comparison with European or American atlases, only few dialects of various Arabic countries have been comprehensively studied and described. In marking linguistic differences within geographical dialectology, vocabulary has traditionally played an important role. In this respect concerning Arabic dialects Cadora’s study of the Syro-Lebanese dialects found that the urban dialects in this region revealed 96% non-contrastive compatibility on average (Cadora, 1979:32). This means that almost all the vocabulary is shared by many SyroLebanese dialects (at least those studied by Cadora). Certain phonetic features may help in distinguishing some of the different dialects, but not all of them, because most of these dialects also share much of their phonological systems. Cadora also writes (following Swadesh, 1952a; 1952b; 1971) that a 70% minimal requirement of non-contrastive compatibility between dialects is enough for them to be still considered varieties of the same language. By this definition most of the Syro-Lebanese dialects belong to one and the same language, i.e., Arabic, which is correct, as is well known. But knowing this does not help the complicated LADO tasks which require distinctions rather than similarities. The compatibility thus actually complicates things for language experts who work on LADO cases. Because of this compatibility, or inherent similarity, feature salience (of vocabulary and phonology, in particular) is difficult to tackle in Arabic dialects (but see Tables 1—3 in Section 4 below). We will try now a different approach to answering Question (b). A full examination of the features of a language would usually refer to syntax, morphology, phonology and vocabulary features (in this order). This approach is due to the generally accepted view that syntax and morphology constitute the core of a language system (and therefore they are also the last to be affected by external linguistic forces). Phonology and vocabulary, on the other hand, usually change (both diachronically and synchronically) as a result of external linguistic pressures more easily than do syntax and morphology. Now, every dialect has some feature(s), whether syntactic, morphological, phonological or lexical, which differ from other dialects and thus the combination of such features may distinguish it from others. It appears that being able to consult and compare full lists of such features of the relevant dialects could help the linguist’s LADO task. We continue this line of thought below. Question (c) We turn now to our third and somewhat philosophical question: Should one define a speaker’s original dialect only by what is in the recorded texts or also by what is not? In other words, should one consider the fact that certain typical features of the claimed dialect are missing from the speaker’s recording (where theoretically they could or should occur) in trying to confirm or refute a client’s claim? This question arises since speakers may try to disguise certain features of their original dialect by levelling (i.e., by using avoidance strategies—see, e.g., Milroy, 2002), or imitating other dialects by koinèization (i.e., by using features which also occur in dialects other than their native dialects—cf. Ferguson, 1959b; Miller, 2006). Speakers often use such strategies—at times unconsciously. Levelling and koinèization function to conceal what speakers may consider some weakness of their native dialect (e.g., a well-known local dialect feature which is ridiculed by speakers of a more prestigious dialect). Conversely, speakers may use koinèization, i.e., features which they share with their interlocutors’ dialect, to show their bonding or closeness with them, or due to the others’ perceived superior social status. Over 50 years ago, Blanc (1960) described these -295-

two inter-dialectal communication phenomena in conversations of speakers of Arabic from various dialect communities. Such behavior has also been described as part of speakers’ sociolinguistic accommodation to each other due to factors such as situation, status, sex, or origin (Trudgill, 1986). Knowing this, should an expert’s work rely on explicit facts only? Or can the lack (or avoidance) of a feature also be taken into consideration? This issue complicates LADO work, because a speaker cannot really control sociolinguistic or psycholinguistic factors even in a recorded monologue. We consider this an important issue in the make-up of a speaker’s speech and accordingly believe that it needs to be taken into consideration some way or another (see below). 4. Linguistic features of CA dialects: some examples Most of the above considerations are relevant for other languages as well as for Arabic. To demonstrate some of these aspects in Arabic we present here Tables 1, 2, 3 with examples of shared and distinctive features of three linguistic domains: syntax, morphology and phonology. These examples do not constitute a comprehensive list of such features in any dialect, but exemplify the large variety of features that need to be considered. The examples are based on the literature, including EALL (2006—2008), Fischer and Jastrow (1980), Kaye and Rosenhouse, (1997), among others. Table 1: Shared and distinctive syntactic features of Arabic dialects

Shared syntactic features Distinctive syntactic features The Arabic system distinguishes gender The copula is usually a simple 3rd person and number in nouns, adjectives and verbs. pronoun between the subject and predicate. Only in few dialects does a copula occur at the end of the sentence; e.g., Baghdad: [ya:nu]; Anatolia: [we]. Noun gender is either male or female (there A bound personal particle preceding the is no neuter in Arabic). direct object and the possessive structure is typical of Iraqi Arabic; it occurs also in some other Syro-Levantine dialects, but not outside this region: Iraq: [ʃufto la-mħammad ] Saw-I-him to-Muhammad ‘I saw Muhammad’ [be:t-o la-mħammad] house-his to-Muhammad ‘Muhammad’s house’ Noun number is singular, dual (in LA, less Imperfect indicative particles occur in frequently used in CA) and plural. many CA dialects (though not in all of them); but lexically these particles vary and thus characterize and distinguish different dialects: Damascus: [bi]-; Iraq: [qa]-; Morocco: [da]-, [ka]-, [ta]. Arabic has nominal and verbal sentences Some forms of the negating particles of -296-

(i.e., the predicate can, but does not have to nouns and adjectives distinguish various be a verb or verb phrase). dialects, e.g.: Baghdad: [ma:] Damascus: [mu:] Cairo: [muʃ] Jerusalem: [ma:],[muʃ],[miʃ ]. Verb and adjective number and gender Genitival exponents (meaning ‘of’) exist in agree with noun heads. a large variety in many dialects and thus distinguish them; see e.g.: Iraq [ma:l]; Syria: [tabaʕ]; Cairo: [bita:ʕ]; Galilee Bedouin: [ħagg]; Morocco: [ntaʕ], [djal]; Malta: [tagħ], [tiegħ]. For marking possessives, the construct state pattern (noun + noun) is very common, but also analytical genitive structures exist in the dialects (using varying prepositions to mark them). Attributed adjectives follow nouns. Table 2: Shared and distinctive morphological features of Arabic dialects

Shared morphological features Basic noun patterns have usually three radical (root) consonants, though there are also words with two or four consonantal radicals.

Distinctive morphological features Pronouns are basically similar but vary (phonetically) in many dialects; e.g.: ‘I’: Casablanca, Cairo, Damascus: [ʔana]; Yemen and various Bedouin tribes: [ʔani] [ʔa:ni][ʔani:].

Noun patterns are derived by changing vowels or adding syllabic affixes to root Bound pronouns (suffixes) are basically consonants. similar but vary phonetically in the dialects; e.g.: ‘his’: Cairo: -[uh]; Damascus: -[o]; Negev Bedouins: [ih]/[ah]/[uh]; Morocco: -[u]/[eh]. Affixes in nouns provide information about gender, number and semantic elements. Verb conjugation patterns of the perfect tense may vary due to different phonetic rules:‘I wrote:’ Cairo, Damascus: [katabt]; Morocco: [ktəbt]; Beoduins: [tʃitabt]. Basic verb patterns have usually three radical (root) consonants, though there are also verbs with two or four consonantal radicals Verb forms are formed by changing vowels and adding syllabic affixes to root -297-

consonants. Verb affixes provide information about agent gender, number, tense, mode or aspect (depending on the function of the very form). Table 3: Shared and distinctive phonological/phonetic features

Shared phonological/phonetic features The Arabic phonological system has more consonants (basically 28) than vowels (basically 3).

Distinctive phonological/phonetic features LA vowels are [i], [a], [u]; in CA there may be more vowels, for example added [e], [o] and [ə].

The consonant system includes laryngeal, Vowel length is retained in urban Eastern pharyngeal and velarized/pharyngealized dialects but in urban Western dialects vowel (‘emphatic’) consonants [ʔ], [h], [ħ], length often changes (long>short; short>null). [ʃ], [s ʔ], [t ʔ], [ðʔ], [d ʔ]. Diphthongs /ai/and/au/ are retained in few dialects (e.g., Lebanon); they are often monophthongized: /ai/>[e:] and /au/>[o:] in Syro-Levantine dialects, and /ai/>[i]or[a] and /au/>[u]) in urban Western dialects. The [ʔima:la] process ([a] > [e] or [i] in various morphophonological conditions) differs in different dialects; e.g.: Cairo: no [ʔima:la]; Damascus: conditioned; Iraq: very frequent (conditioned). The consonant /q/ is pronounced as [ʔ] in urban Eastern dialects; [q] is often retained in urban Western dialects; and /q/ is voiced (and affricated) as [g], [ʤ] or other variants in Bedouin and Bedouinized dialects (including certain dialects of Jordan, Iraq and in the Arabian Peninsula). The consonant /k/ is pronounced as [k] in most urban Eastern and Western dialects but is affricated [ʧ] in certain rural and Bedouin dialects. Pharyngealized consonants (e.g., [sˤ], [tˤ] and[dˤ ] are weakened in urban dialects, and are strong in Bedouin dialects. -298-

Secondary pharyngealization occurs in certain consonants (e.g., [lˤ] and [rˤ], in various dialects (many Bedouin dialects use more secondary pharyngealization than sedentary dialects, and with more consonants). Syllable structures differ greatly between Western and Eastern dialects (mainly following vowel system differences). Intonation patterns differ greatly between Western and Eastern dialects. The field of vocabulary includes a huge volume of data that an Arabic expert is expected to handle. Even in pre-Islamic Arabic the nomadic tribes of the Arabian Peninsula spoke different dialects, which differed from LA (Rabin, 1951). Arabic vocabulary naturally expanded in space and time all over the MENA up to the present. Thus, the Arabic vocabulary is basically similar in its lexical semantic fields throughout its many geographical regions, but also has many distinctive elements in different dialects (cf. Behnstedt and Woidich, 2005; Cadora, 1979; Ibrahim, 2008). We therefore mention here only three issues that have yielded distinctive CA elements. As a first example see the verb ‘speak’. The map provided by Behnstedt and Woidich (2005:209, Map 98) shows three main varieties in different large regions of the Middle East (indicated by circles, squares and triangles representing three roots: klm, ħky, hdr). In addition, there are ten other roots, each used by fewer dialects dispersed in the large Arabic-speaking MENA territory. This map is a characteristic example of the way certain lexical items can be shared by some dialects while differing among others. For LADO, then, if a speaker uses one of the above main verb roots it is harder to define the original dialect than if that speaker uses one of the more localized forms. Another linguistic phenomenon is lexical contamination, i.e., mixing different lexical items with some linguistic (e.g., phonological, morphological) feature that does not belong to that word form. For example, in the region of Lebanon-Syria we find the following variants for the word ‘lip:’ ʃiffe, burt ˤum, as well as bartˤ u:me, burtˤme; another example (from Egypt) is the ‘cooking oven:’ furn and tˤabu:na; but also furnı (Behnstedt and Woidich, 2005:214—215). In these examples, the feminine suffix /-e/ of ʃiffe and /ı/ in furnı has been added to a word which originally does not end with this suffixed vowel (this process also involves gender changes in the words). The semantic splitting of a lexeme to several meanings is a frequent process in many languages. In Arabic dialects this occurs as well, as in e.g. the word batˤtˤi:x. In various Syrian dialects it may mean watermelon (red), melon (yellow), or either (i.e., it is neutral, and the specific meaning is made explicit by adding an adjective—‘red’ or ‘yellow’) (Behnstedt and Woidich, 2005:216—217). Thus, while certain features are shared by various Arabic dialects, other dialects distinguish different features. This makes it extremely difficult, if not impossible, to know all the details of the different dialects. Even less possible is it to know the dialects (and LA) as they are changing in speakers who reside outside their native locations and are thus influenced by many new linguistic sources.

-299-

5. A suggestion for LADO Up to now we have seen that the linguistic analysis of Arabic dialect recordings for LADO goals may be very difficult in certain cases, because of linguistic instability as well as lack of knowledge of the features of this instability. This situation seems to call for some new method of analysis to help answer our questions and enable reliable LADO decisions. In this context computer-assisted work comes to mind. In contrast with many computer-aided works on speaker and speech recognition, and few studies of dialect identification and dialectometrics (e.g. Barkat, 2001; Barkat et al., 2004; Belinkov, 2010; CASL, 2008; Rose, 2002) little effort has been invested in designing or applying computer programs for LADO. As a first step towards achieving computerized LADO it seems necessary to divide the topic into two parts: a) A feature array (database) of as many Arabic dialects features as possible. b) A computer program which calculates the probabilities of origin by features and feature combinations. At the first stage the lists should include: a) The basic features which are shared by all the Arabic dialects. b) A list (or lists) of the distinctive features of each and every Arabic dialect. At the second stage the program will register the data, classify and list the features by dialects, and calculate the probability that a recorded text which has a certain number of features reflects a claimed dialect. The scoring of each feature for the calculations may differ according to the salience of the features: ‘typical’ (or specific) syntactic and morphological elements could be considered generally more salient than lexical ones, so that the score of a syntactic feature would be higher than that of a lexical item. Phonetic elements such as segmental features (i.e., phonemes and their acoustic-phonetic features) and prosodic features (such as intonation, pauses and speech rate) should have different but relatively high scores. Specific lexical elements (highly representing a certain dialect) could have a higher score than less uniquely used lexical items (i.e., items that are shared by a number of dialects). Although it is difficult to define scoring details, scores are a fundamental requirement in this envisioned computer program. It will apparently require several stages of ‘trial and error’ until an adequate (reasonable) scoring is found. Since it has been suggested here that the computer program should also account for ‘missing’ features, the next question is how this can be achieved. One way is that typical features that do not appear in the LADO client’s recording (where they could or should occur, according to the dialect features) will be taken into account as a negative factor (score) for the final assessment. What the effect of such deductions on the final score (and the LADO decision) can be—remains for the time being a topic of future research. As a baseline for the calculation, Swadesh’s rate of 70% (Cadora, 1979) could be considered: if the sum of features in the recording match about 70% or more of the dialect features which the speaker claims to be speaking, some certainty as to his/her claim may be assumed. Higher certainty rates are better, of course (as required in many computational linguistics programs). Thus, if a recording is claimed to reflect some CA dialect, then:

-300-

a) If the final score of the features in the recording matches 70% or more of the features that are present in the database features, it can be considered the speaker’s original dialect. (This might be considered a first, and very rough, estimate, and higher compatibility rates are required. These requirements will have to be developed when the project actually materializes). b) In the case that a recording shows features of more than one dialect (implying foreign dialect features, whether by koinèization or otherwise), the claimed dialect would be confirmed if the sum of the claimed dialect features are more salient than the sum of the foreign dialectal features (e.g., if there are more ‘original’ syntactic and morphological features than ‘foreign’ ones, but many ‘foreign’ lexical items.) c) If the recorded text lacks salient features of the applicant’s claimed dialect (due to levelling), a ‘negative’ consideration of the speaker’s claim could take place, as mentioned above (Section 2, Question c). Speakers’ speech accommodation needs to be taken into consideration in this context via the scoring method. But deciding the rate of negative, as well as positive, scoring is beyond the present paper and is left for future research. Such a design presents the possibility of an accurate calculation of the rate that a text matches some claimed dialect. Clearly, this should be a large-scale project which will involve a great amount of work, time, and financial investment, but with contemporary programming abilities and memory scopes it might be technically feasible. It would be worthwhile to test such a project initially with two dialects and then see how to proceed. In spite of the benefits of the above suggestion for LADO, and even if this program fully materializes, and all the programming difficulties are overcome, we wish to stress that it can be an important aid to LADO, but not the sole tool for LADO decisions, i.e., it will not replace human language experts. This view is based on several reasons. Firstly, as already noted, not all the Arabic dialects are documented (Behnstedt, 2006); nor will they be fully documented in the foreseeable future. Secondly, since speakers use idiolects, which vary also in phonetic features and other consciously-uncontrollable properties, it is not easy to assess a speaker’s personal vs. dialectal features. Thirdly, dialects change all the time, and no program can be constantly (automatically) updated to follow all the changes. Moreover, beyond definite language rules, human verbal communication options are almost unlimited (in the use of synonyms, paraphrases, innovated structures and lexemes, etc.), and speakers will continually make original expressions, including such that mix with ‘foreign’ dialects or languages. In addition, there are surface communication problems such as hesitations and pauses in spontaneous speech that are still unsolved in speech analysis studies and which have not been considered here. It therefore remains to be seen whether any institute or another (international) research body will undertake such a huge project to assist LADO. 6. Conclusion In this paper we have analyzed some linguistic aspects related to LADO tasks. First, we noted the numerous Arabic dialects and varieties. Dialect variance exists not only due to Arabic speakers’ number and dispersion around the world, but also because languages develop and change due to speakers’ sometimes unexpected, random or even ‘chaotic’ language mixtures following hectic life histories. Due to internal and external non-linguistic and linguistic processes (including code switching, koinèization, borrowing and levelling, etc.) Arabic varieties continue changing everywhere, as is ultimately expressed in the speech of an -301-

individual. These natural processes complicate a language expert’s LADO work, but must be taken into consideration. In order to facilitate LADO tasks, we have suggested the design of a computer program which might follow the statistical likelihood approach (Rose 2002) and adapt it to LADO needs. Such a program would require a large scale study and long-term execution, but if undertaken at gradual stages it may be achieved. Teams of linguists, programmers and statisticians should, of course, be involved in such a project. This envisioned program may at least support the human language expert’s often semi-scientific opinion about the origin of a recorded text, and thus contribute to LADO tasks. The final human touch and control is inevitable in any task of this kind, however, and cannot be dispensed with, as explained above. Our three initial questions seem to remain without satisfactory answers: in this area, as in many other cognitive fields, humans can act to some level of precision, but rarely to perfection (cf. Fraser, 2009; Kreiman and Sidtis, 2011). That is, one cannot really be 100% certain that a linguist’s report in a complicated case will always be correct. Adding statistical calculations of probabilities and likelihoods may increase the expert’s certainty of his/her opinion. Hopefully, such a program will produce (in time) results with sufficiently high probabilities; for the time being we rely on human skills in LADO tasks. References Al-Wer, E. (2002) Education as a speaker variable. In A. Rouchdy (ed.) Language Contact and Language Conflict in Arabic: Variation on a Sociolinguistic Theme. London: RoutledgeCurzon, 41—53. Barbiers, S. and H. Bennis (2007) The syntactic atlas of the Dutch dialects: a discussion of choices in the SAND—Project paper. Nordlyd 34: 53—72. http://www.ub.uit.no/baser/septentrio/index.php/nordlyd/article/viewFile/89/85 Barkat-Defradas, M. (2001) Vers l’identification automatique des parlers Arabes. Revue Langue et Linguistique 7: 47—95. Barkat, M., Hamdi, R., Pellegrino, F. (2004) De la caractérisation linguistique à l’identification automatique des dialectes Arabes. MIDL Workshop, 29—30th November 2004, Paris.http://archives.limsi.fr/2004/MIDL//actes/session%20II/Barkat_Defradas&Ham di&Pellegrino_MIDL2004.pdf Behnstedt, P. (1997—2000) Sprachatlas von Syrien. Wiesbaden : Harrassowitz Verlag. Behnstedt, P. and Woidich M. (2005) Arabische Dialektgeographie. Eine Einführung. Leiden: Brill. Behnstedt, P. (2006) Dialect geography. In K. Versteegh (ed.) Encyclopedia of Arabic Language and Linguistics Volume 1. Leiden: Brill, 583—593. Belinkov, Y. (2010) Computational dialectology. NLP Dialectometrics Seminar, January 2010, Tel-Aviv University.

-302-

Bentahila, A. and Davies, E. (1995) Patterns of code-switching and patterns of language contact. Lingua 96 (2—3): 75—93. Blanc, H. (1960) Style variations in spoken Arabic. In C.A. Ferguson (ed.) Contributions to Arabic Linguistics. Cambridge: Cambridge University Press, 81—158. Cadora, F.J. (1979) Interdialectal Lexical Compatibility in Arabic: an Analytical Study of the Lexical Relationships among the Major Syro-Lebanese Varieties. Leiden: Brill. CASL (2008) AVIA (Arabic Variant Identification AID). http://www.casl.umd.edu/1227 Diem, W. (1974) Hochspraceh und Dialekt in Arabischen: Untersuchungen zur Heutigen Arabischen Zewisprachigkeit. Wiesbaden: Steiner. Eades, D. (2009) Testing the claims of asylum seekers: the role of language analysis. Language Assessment Quarterly 6(1): 30—40. Editors of the American Heritage Dictionaries (2006) The American Heritage Dictionary of the English Language, 4th ed. Boston: Houghton Mifflin Ethnologue (2005) Languages of the World, 15th ed. http://www.vistawide.com/languages/top_30_languages.htm Ethnologue (2009) World Ethnologue, Summer Institute of Linguistics. http://www.numberof.net/number-of-arabic-speakers/ Ferguson, C.A. (1959a) Diglossia. Word 15: 325—340. Ferguson, C.A. (1959b) The Arabic koinè. Language 35: 616—630. Fischer, W. and O. Jastrow (eds) (1980) Handbuch der Arabischen Dialekte. Wiesbaden: Otto. Fraser, H.(2009) The role of ‘educated native speakers’ in providing language analysis for the determination of the origin of asylum seekers. The International Journal of Speech, Language and the Law 16 (1): 113—138. Gafaranga, J. (2007) Code switching as a conversational strategy. In P. Auer and L. Wei (eds) Handbook of Multlingualism and Multilingual Communication. Berlin, New York: Mouton de Gruyter, 279—313. Hickey, R. (2002) How do dialects get the features they have? On the process of new dialect formation. In R. Hickey (ed.) Motives for Language Change. Cambridge: Cambridge University Press, 213—239. Holes, C. (2004) Modern Arabic: Structures, Functions, and Varieties. Washington, D.C.: Georgetown University Press (revised edition). Ibrahim, Z. (2008) Lexical variation: Modern Standard Arabic. In K. Versteegh (ed.) Encyclopedia of Arabic Language and Linguistics Volume 3. Leiden: Brill, 13—21. -303-

Kaye, A.S. (1994) Formal vs. informal in Arabic: diglossia, triglossia, tetraglossia etc., polyglossia–multiglossia viewed as a continuum. Zeitschrift für Arabische Linguistik 27: 47—66. Kaye, A.S. and Rosenhouse. J. (1997) Arabic dialects and Maltese. In R. Hetzron (ed.) The Semitic Languages. London: Routledge, 263—311. Koenig, W. and Paul, H.J. (1978) DTV—Atlas zur deutschen Sprache: Tafeln und Texte. München: Deutscher Taschenbuch Verlag. Kreiman, J. and Sidtis, D. (2011) Foundations of Voice Studies: an Interdisciplinary Approach to Voice Production and Perception. Chichester, UK: Wiley-Blackwell. Krumbacher, C. (1902) Das Problem der Modernen Griechischen Schriftsprache. Munich: KB Akademie der Wissenschaften. Lewis, M. P. (ed.) (2009) Ethnologue: Languages of the World, 16th edition. Dallas, Tex.: SIL International. Marçais, W. (1930) La diglossie Arabe. L'Enseignement Publique 97: 401—409. Meiseles, G. (1980) Educated spoken Arabic and the Arabic language continuum. Archivum Linguisticum 11: 118—148. Mejdell, G. (2006) Code-switching. In K. Versteegh (ed.) Encyclopedia of Arabic Language and Linguistics Volume 1. Leiden: Brill, 415—421. Miller, C. (2006) Dialect Koinè. In K. Versteegh (ed.) Encyclopedia of Arabic Language and Linguistics Volume 1. Leiden: Brill, 593—597. Miller, C., Al-Wer, E., Caubet, D. and Watson J. (eds) (2007) Arabic in the City: Issues in Dialect Contact and Language Variation. London: Routledge. Milroy, L. (2002) Introduction: mobility, contact and language change—working with contemporary speech communities. Journal of Sociolinguistics 6: 3—15. Muysken, P. (2000) Bilingual Speech: A Typology of Code Mixing. Cambridge, UK: Cambridge University Press. Muysken, P., Verrips, M. and Zwaan, K. (2010) Introduction. In K. Zwaan, M. Verrips and P. Muysken (eds) Language and Origin—The Role of Language in European Asylum Procedures: Linguistic and Legal Perspectives. Nijmegen: Wolf Legal Publishers, 1—6. Rabin, H. (1951) Ancient West Arabian. London: Taylor’s Foreign Press. Rose, P. (2002) Forensic Speaker Identification. London: Taylor & Francis.

-304-

Rosenbaum, G.M. (2000) ‘Fusˤħāmmiyya’: alternating style in Egyptian prose. Zeitschift für Arabische Linguistik 38: 68—87. Rosenbaum, G.M. (2000—2002) ‘Do you parler ‘Arabi?’ Mixing colloquial Arabic and European languages in Egyptian literature. Materiaux Arabes et SudarabiquesGELLAS, Nouvelle Serie 10: 11—47. Rosenbaum, G. M. (2004) Egyptian Arabic as a written language. Jerusalem Studies in Arabic and Islam 29: 281—340. Rosenhouse, J. (1998) Women's speech and language variation in Arabic dialects. Al‘Arabiyya 31: 123—152. Rosenhouse, J. (2009) Arabic in comparative forensic linguistics. Paper presented to the 4th Conference on Translation, Interpretation and Comparative Legilinguistics, 2nd—4th July 2009, Poznan, Poland. Rosenhouse, J. (submitted) Bilingualism/Multilingualism in the Middle East and North Africa: a focus on cross-national and diglossic bilingualism/multilingualism. To appear in W.C. Richie and T.K. Bhatia (eds) Handbook of Bilingualism/Multilingualism. Oxford: Blackwell. Sadiqi F. (2003) Women, Gender and Language in Morocco. Leiden and Boston: Brill. Shraybom-Shivtiel, Sh. (1995) The role of the colloquial in the renaissance of Standard Arabic. Israel Oriental Studies 15: 207—215. Swadesh, M. (1952a) Lexicostatistical dating of prehistoric ethnic contacts with special reference to North American Indians and Eskimos. Proceedings of the American Philosophical Society XCVI: 452—463. Swadesh, M. (1952b) Towards greater accuracy in lexicostatistics dating. International Journal of American Linguistics 28: 223—230. Swadesh, M. (1971) What is glottochronology? In Swadesh, M. (ed.) The Origin and Diversification of Language. Chicago, New York: Aldine—Atherton, 271—284. Trudgill, P. (1986) Dialects in Contact. Oxford: Basil Blackwell.

-305-

‘Faked’ or truthful second language proficiency: assessing claims Margaret van Naerssen Immaculata University, USA [email protected] Abstract In legal cases there is sometimes a question about whether a non-native speaker (NNS) is being truthful about problems in understanding or speaking. Some may try ‘faking’ low second language (L2) proficiency levels for legal advantages. An attorney using low proficiency as an argument faces the likely challenge: ‘How do you know he wasn’t faking?!’. Detecting possible ‘faking’ poses a significant challenge. This paper reports on a replication of a protocol developed initially for a federal drug trafficking case (in the USA) to determine the ease/difficulty of a non-native English speaker to ‘fake’ a lower than truthful language proficiency. Initial results from the original case suggested the degree of subconscious ‘leakage’ of content across languages might be evidence of truthful listening abilities in the L2. A replication study was carried out with 18 subjects outside the legal system identified through a metropolitan area Chinese community network of Taiwanese Chinese speakers. Keywords:

LANGUAGE PROFICIENCY; NON-NATIVE SPEAKER; FAKING; DECEPTION; INTERROGATION

Scenario (hypothetical) Ms Q, from a poor region of the world, entered the US illegally in a human trafficking scheme and was forced into prostitution. When the trafficking ring was uncovered, she was released. With legal assistance, she was eventually able to become a part of her ethnic community in the US and found a job which did not require any English. One day she was offered a chance to earn better pay, but she would have to learn how to drive. She was taught the English and content needed to pass the state driver’s license test which was given orally. She was given a car and ownership papers, and told she was to drive an ‘‘important’’ man half way across the US. She would be paid a big bonus when she returned. En route she was stopped by a police officer because her driving had briefly become erratic. (She later said she was taking off her jacket.) His video-cam was broken so there was no recording of the subsequent interactions. When asked for her license and registration, she pulled out some papers. The officer took what he needed. He became suspicious: the car and driver were from out-of-state. She and the backseat passenger looked foreign. He asked if he could look in the trunk/boot of the car, pointed toward the back of the car, and began walking to the rear of the car. At that moment the passenger shouted to her in their language, ‘‘Get out of here or we are dead’!’ She put her foot to the accelerator. A 20-minute car chase ensued. As she attempted to turn off the main road at a ramp, the car flew into the air, landed in a field, and rolled over. The car trunk flew open, the contents fell out and a bag split open, revealing packages of heroin. Both passengers were taken to a hospital. The passenger had minor injuries. Ms. Q was in the hospital for over two weeks. Both were charged with drug trafficking and had separate trials. -306-

As the police officer’s video-cam was in the repair shop there was no language evidence. The police officer testified that he thought she understood enough English to have agreed to his request to search the car. Ms. Z claimed she did not understand the request and that her English was very poor. She also claimed not to have known about the heroin. She had just been hired as a driver. The officer claimed she was lying about her English. 1. Overview In some legal cases questions arise about whether non-native speakers (NNSs) are being truthful about problems in understanding and/or speaking. Some may try ‘faking’ low second language (L2) proficiency levels for legal advantages. If an attorney uses low proficiency as an argument, the primary challenge is likely to be: ‘How do you know he wasn’t faking?’. Admittedly, detecting this type of deception poses a significant challenge. To try to address the issue of possible ‘faking,’ an experimental protocol was developed for determining the ease/difficulty of ‘faking’ a lower than truthful language proficiency. It was applied in a federal case involving a NNS charged with drug trafficking. The protocol involved an alternating language story retell task (Chinese and English) along with an English oral assessment. This was also tested on two other subjects at higher levels of English proficiency than the defendant. Initial results suggested the degree of subconscious ‘leakage’ of content across languages might be evidence of truthful listening abilities in the L2. The federal public defender encouraged research on it, thinking the protocol had very interesting possibilities. This is a brief research report on a replication of that protocol. Lacking additional relevant legal cases, a replication study was carried out with 18 subjects outside the legal system identified through a metropolitan area Chinese community network of Taiwanese Chinese speakers. There is little research on detecting the possibility of deception by non-native speakers in legal contexts, especially for the purpose of ‘faking’ a lower than truthful language proficiency for a legal advantage. In the behavioral sciences (i.e., criminology) there is extensive research and debate on detecting deception in law enforcement interviews, and in forensic psychology on malingering a psychological condition. Some legal practitioners have applied the term ‘malingering’ to assumed ‘faking’ of a lower than truthful language proficiency. However, while ‘faking’ is a form of deception, for this author psychologically-based ‘malingering’ and the assessment criteria used do not seem to fit the underlying cognitive processes that might be involved in attempting to manipulate a less than truthful second/ foreign language use. In the area of authorship analysis a linguist may attempt to identify, using linguistic analyses, the background of a person that authored a questioned text. The question might arise as to whether an author is attempting to disguise her/his linguistic background by writing like a non-native speaker, or is in fact likely to be a non-native speaker. Researchers in multilingual and multicultural forensic contexts include, among others, Eades (2010). Reports on individual cases involving NNSs provide valuable findings. However, few attempts have been made to distinguish between deceptive language performance for legal advantage on the one hand, and truthful language performance on the other. Eggington and colleagues (2011) and van Naerssen (2009, 2011) appear to be the only researchers primarily working in this area. These researchers are drawing on work in such areas as second language acquisition and assessment, sociolinguistics (including discourse analysis), language learning strategies, and cognitive psychology/ psycholinguistics. This paper is a report on one of the studies in this area.

-307-

Another area of research sharing a concern about the truthfulness/ authenticity of the language used by an individual surrounds the practice Language Analysis and Determination of Origin (LADO). LADO is used by some governments to evaluate the claims of asylum seekers about their regional or national origins. Researchers in this area have formed the Language and Asylum Research Group, an outgrowth of The National Language and Origin Group, which has developed guidelines to assist in the interviewing of asylum seekers (2004). However, while LADO also focuses on the individual, it has a different goal. Broeders (in Eades, in press) points out that in most cases involving forensic analyses, the concern is linking evidence to an individual. In contrast, the use of LADO involves a ‘classification process’ to attempt to determine whether the speaker belongs to a particular group of speakers. 2. Replication study1 In the federal case the aim was to try to identify how likely it was that a person at the defendant’s tested English proficiency could interact accurately, intelligibly, and easily in a law enforcement traffic stop involving (1) simple personal informational questions and (2) narrating short sequences of events. The expert testified that non-native English speaker (NNES) would probably have needed an American Council on the Teaching of Foreign Languages (ACTFL) OPI English proficiency level of at least Intermediate-Mid. The defendant tested at a Novice Low-Mid on an ACTFL-like Oral Proficiency Interview. Levels do not address cultural knowledge (including knowledge of legal rights/procedures) or the relevant language The original task used in the legal case was grounded in related research and practice in story retell tasks, language assessment, and second language acquisition. However, the data from only three individuals were not enough on which to begin developing firm hypotheses. 2.1. Aims The aims of the Replication study were to: 1. Make revisions to the initial alternating language story retell protocol and to test this revised protocol; 2. Determine whether the initial findings could be supported with 18 new participants at three general proficiency levels of English; 3. Explore, with two participants, intentional ‘faking’ of a lower than truthful English proficiency. The research questions were: 1. At what tested English proficiency level would the Taiwanese participants (TWPs) recognize there were two stories? 1

This study was supported by the Office of Sponsored Research, Immaculata University, Immaculata, PA USA. I am also grateful S. Marie Cooper, IHM, the director.

-308-

2. At what tested English proficiency level would TWPs NOT recognize there were two stories, but in retell, include some information from the English story? 3. At what tested English proficiency level would TWPs not include any English information? The details of this protocol are described later. To further explore the proposition that this protocol might catch those ‘faking’ a lower proficiency, two subjects were asked to intentionally try to ‘fake’ a lower proficiency. 2.2. Participants2 Of the 18 subjects, there were at least three subjects at each of the following general English proficiency levels: beginning, intermediate, and advanced intermediate/ advanced. The remaining subjects clustered primarily at mid-high Intermediate and Advanced levels. More at lower levels would have been desirable, but there was not a ‘ready pool’ of previously tested participants. Three other factors affected the choice/recruitment of participants: age, educational background, and regular language use of Taiwanese. Forty-four percent of the participants were above the age of 55. Two older persons (around 75 years old) specifically requested the chance to participate. After it was determined they were not hearing impaired, they were accepted for practical reasons: they were available, willing, and provided lower proficiency examples. Participants were from two general groups: settled immigrants and current/recent graduate students or students studying English at a university. A majority (10/18) had been in the US for ten years or more. 2.3. Project description/procedures3 A widely accepted oral proficiency interview used in adult education (BEST-Plus) was administered to 18 native speaker subjects of the Taiwanese He-Lo dialect of Chinese by a certified examiner (Investigator). Each subject was paid $23 for participating in the research session and was sent a copy of their official BEST-Plus score for possible use elsewhere. Table 1 is an overview of the Language Assessment Protocol. All tasks were audio-recorded.

2

Recruitment assistance came primarily from two Taiwanese community leaders in the TAAGP- Taiwanese American Association of Greater Philadelphia and congregation of the Taiwanese Grace Church of Christ in Philadelphia. The hospitality of the participants was very generous. Connections were also made by Immaculata faculty and alumni. 3 The Research Team consisted of the Principal Investigator, Dr Margaret van Naerssen, second rater and English transcriber, Marion Piccolomini, the Chinese Language expert Annie Hui-Ting Lai and two occasional assistants (YiChi Lien and Tz Shan Huang), and the research forms translator, Dr. Jiang-Yue Zhang.

-309-

Table 1: Ordered tasks & details of protocol Oral Proficiency Interview BEST-Plus

Practice Story Retell (all Taiwanese)

Alternating Story Retell Task

Follow-up: feedback on task

- Oral English proficiency test for adults ; - Computer-adapted; - 2 raters. Purposes: 1. Warm up English ‘channel’; 2. Provide reference for use in analysis.

Purposes: 1. Warm up Taiwanese ‘channel’; 2. Practice a retell task.

Steps: 1. Listen; 2. Do children’s puzzle (no talking, distractor); 3. Retell.

Purposes: 1. Provide retrospection on task & performance; 2. Answer questions about task; 3. Alert examiner to any problems.

2.3.1. English oral proficiency test A widely accepted oral proficiency interview used in adult education (BEST-Plus) was administered to 18 native speaker subjects of the Chinese Taiwanese He-Lo dialect by a certified examiner (Investigator). This task was completed first to warm up the English ‘channel.’ The computerized version of BEST-Plus was used. It reported the BEST Plus Scores and Student Performance Levels (SPLs). In this report the SPLs were primarily used (see Center for Applied Linguistics, 2005). Two subjects (one intermediate, one more advanced) were also asked to ‘fake’ a lower than truthful English proficiency and were given the role of ‘an arrested drug dealer’. 2.3.2. Warm-up retell in Chinese The next step was to practice a task similar to the final Alternating Language Story Retell. In the original case no warm-up task was carried out. Each participant was asked to listen to a story told in Chinese (presented via an audio recording), and then to retell the story in Chinese. This listen-retell warm-up differed from the final task as it was all carried out in Chinese and was a single story rather than two alternating subplots. It was decided that participants should not know in advance the actual format of the final task. Also, doing the warm-up task helped to warm-up the Chinese ‘channel’. Finally, they would be practicing the task using their native language, developing some initial comfort with the listen and retell activity, which is not a common task for adults. 2.3.3. The Alternating Language Story Retell Task The initial idea for the Alternating Language Story Retell Task came from Canadian cognitive psychologist, Bialystok, who indirectly pointed the way when reporting on a study involving a story retell task (The Frog Story) to examine domains of proficiency in bilingual children (2001). However, one drawback is that an adult intending to deceive could simply say, ‘I didn’t understand’. To address this, an alternating language story retell task was developed (see Figure 1). This task assumes subjects access the narrative content through their first language (L1) and whatever English they have. The question to be addressed is thus, what uniquely ‘English information’ shows up in the Chinese retell? Such leaking of information from the English story might provide evidence of a general level of their L2

-310-

listening comprehension. This information evidence was then compared with the language elicited in the L2 language proficiency interview.

Alternating Language Story Retelling Defendant Listens to story with 2 related storylines - told in an alternating pattern in - two languages, Then retell in own first language (L1)

Chinese

English

What English info would show up in retell in Chinese?

Any leaking?

Chinese

Figure 1: Alternating language story retelling process

The Alternating Language Story Retell Task was then administered by the investigator using an audio-taped protocol. The short passages of the two storylines were alternated to make it difficult for the subject to consciously shut down processing in one language. The general theme of the task was couples going to the hospital: each was expecting a baby. Figure 2 shows the beginnings of the two storylines. Beginning of Chinese Storyline 1. It was a very cold winter evening, and it was snowing. 2. Two people arrived at a hospital. 3. The husband was holding his wife by the arm. 4. She was very tall, had long black hair, and was leaning over. She was 9 months pregnant.

Beginning of English Storyline 1. An hour later the snow stopped. 2. Then another couple also arrived at the hospital. 3. They were in a hurry. 4. The woman was short and had red hair. 5. She was also pregnant.

Figure 2: Extracts from alternating storylines

Yuko Butler, whose research includes language acquisition and related cognitive factors, recommended a short, non-verbal distractor task to follow the listening but precede the retell to reduce effect of short-term memory (personal communication, 2004). In this case, a children’s wooden puzzle was used. Below are some guidelines used in developing the storyretelling task. 1. 2. 3. 4. 5. 6.

Coherence across alternating storyline Language Level: roughly mid-intermediate to high (based on original legal question) Theme: general knowledge/interest Simple transitions: within/across storylines to indicate two story-lines Length of alternating parts: avoid conscious control of language switching Story length: not so long as to overtax memory -311-

7. Sufficient information units for possibility of adequate recall 8. Attention-getting ending 9. Non-verbal distractor task before retell: avoid reliance on short-term memory 10. Instructions in Chinese, with audio cues for examiner Two participants were asked to pretend (‘fake’) a lower proficiency on their BEST-Plus test and on the Alternating Language Story Retell task. After the basic data were gathered, the participants were then given another form of the BEST-Plus test to reflect their truthful English performance on the test, in order to respect their dignity and to have data for comparisons across truthful and ‘faked’ performances. 3. Analysis Comparisons were made using English proficiency levels of participants, and the type of ‘leakage’ of content from the English story retold in Chinese. The types of leakage consisted of a) evidence of ability to recognize that there were two story lines, and b) for those who didn’t recognize two story lines, evidence of limited ‘leakage’. The English proficiency data was also analyzed for developmental patterns, for the cases in which English data appeared to have ‘leaked’ into the Chinese retell. The distribution of the tested levels of English listening comprehension paralleled the results from the retell task (with two exceptions). The research questions are repeated here along with the SPL equivalency listening sub-scores (1-10, with 10 being the highest level, see the BEST Plus administrator’s manual, Center for Applied Linguistics, 2005). Listening comprehension scores were key as listening is the ‘gateway’ for incoming information. More findings are available but space does not allow discussion of these. 1. At what level of proficiency on the English test would they be able to recognize that there were, in fact, two storylines? 7, 8, 9. 2. At what level would they not be able to recognize two storylines but did reveal some leaking of information from the English storyline? 4, 6, 7. 3. At what level would there be no leaking? 1, 2. Thus, a NNS with Level 7 listening may or may not recognize two storylines. There were two exceptions at Levels 8 and 9. In the follow-up a Level 9 participant noted that at first she had assumed the English storyline was a translation and did not pay attention to it. However, she was not able to completely shut out the input. So as the story progressed she realized it was not a translation, but by then she was too confused to figure out what had happened. The Level 8 participant (who did not recognize two storylines), later reported that while he usually did well on English tests, he had not had much opportunity to interact with English speakers. This was due to the (a) short time he had been in the US; (b) the isolating nature of his work in the library; and (c) his primarily Chinese friendship network. This exception caused the examiner to further consider the two key tasks: the retell task is cognitively different, perhaps more complex, than the BEST Plus interview. The findings relating to the two ‘fakers’ are summarized here. The goal was to examine the strategies and language the two participants may have used when attempting to lower their performance. Linguistic data, content choice, and other personal strategies the examinees appeared to use to control performance were studied. Data came from scores on the BEST Plus rubric for language complexity and recordings of their ‘faked’ and truthful language use. The Levels used here are from scoring rubrics in the BEST Plus administrator’s manual (2005). -312-

1. Both ‘fakers’ were able to perform at levels lower than their truthful performance, although they differed in their overall strategies. 2. Both displayed a significant reduction in the length of their responses in the Pretending mode, reducing opportunities for frequency/use of some language features. 3. While both lowered their performance to the lower border of a Level 2, neither was able to repress use of some language patterns typical of Level 3 that they had used during their truthful testing. In summary, the Retell task appeared to be cognitively complex enough to make it difficult for the participants to control and manipulate the task. With the BEST-Plus, it also appears that it might be difficult to intentionally (and successfully) avoid revealing evidence of comprehension and to pretend a significantly lower than truthful level of comprehension. 4. Conclusions While more research is clearly needed, the alternating language story retell task shows some promise. There appears to be a general proficiency level in English at which a person is likely to recognize that there were two storylines and show evidence of this in the retell in one’s first language. This probably involves skills similar to those needed in non-face-to-face communication and ability to recognize transitions in events being told. Below this level, recognition of two storylines is not likely. At the same time, below that level of recognition, some English information might be recalled and appear in the L1 retell. When analyzing the data, the differences in the two tasks (in the English test and in the alternating language retell task) are important to consider as well as the backgrounds of individuals in terms of their exposure to and use of the second language. Follow-up research should include a larger number of participants with more at the intermediate and high-beginning levels, and more ‘fakers.’ References American Council on the Teaching of Foreign Languages. www.actfl.org Bialystok, E. (2000) Against isolationism: cognitive perspectives on second language research. In X. Bonch-Bruevich et al. (eds) The Past, Present, and Future of Second Language Research: Selected Proceedings of the 2000 Second Language Research Forum. Somerville, MA: Cascadilla Press, 97—103. Center for Applied Linguistics (2005) BEST Plus Test Administrator Guide. Washington, DC: CAL. Eades, D. (2010) Sociolinguistics and the Legal Process. Tonawanda, NY: Multilingual Matters. Eades, D. (in press) Assessment in asylum-related language analysis. In A. J. Kunnan (ed.) The Companion to Language Assessment. Oxford: Blackwell-Wiley.

-313-

Eggington, W., T. Cox, & S. Wood. (2011) The consequences of faked comprehension in interrogation settings. Paper presented at the International Association of Forensic Linguists 10th Biennial Conference, Aston University, Birmingham, UK. van Naerssen, M. (2009) Exploring issues in ‘faking’/malingering of language proficiency: assessing claims. Paper presented at the International Association of Forensic Linguists 9th Biennial Conference, Vrije Universiteit, Amsterdam, Netherlands. van Naerssen, M. (2011) ‘Faking’ a lower than truthful language proficiency: how easy is it? Poster presented at the 33rd Language Testing Research Colloquium, University of Michigan, Ann Arbor, MI, USA.

-314-

Corruption in the Spanish news: verbalizing crime for public opinion María Ángeles Orts1 and Ángela Almela2 University of Murcia, Spain 1 [email protected] 2 [email protected] Abstract Drawing on an ad-hoc 500,000-word corpus of news-items in digital periodicals representing two different editorial groups, this paper scrutinizes the words deployed by the Spanish news to depict corrupt practices. Our ultimate goal was to explain whether the way in which public opinion is informed matches the reality posited in the new Spanish Penal Code. First, a lexical selection of the terms representing corruption crimes at the heart of the public administration was made. Subsequently, a quantitative analysis of the collected data in terms of frequency and collocation information was performed, plus a qualitative analysis to provide further insight into the corpus-driven data. Our results suggest that the way in which the official press treats corruption crimes is much more expressive, versatile and imaginative than the crude and hard-and-fast way in which law organizes such criminal behaviour. Keywords:

CORRUPTION; PENAL CODE; CORPUS ANALYSIS; PUBLIC OPINION

1. Introduction: sensitivity towards corruption and the building boom in Spain Corruption of civil servants and politicians is perceived as a major issue in Spain, and public opinion has been reflecting the concern for some time now (Jiménez, 2008). Specifically, higher regard has been taken in the country to the work of Transparency International (TI), a global civil society organization based in Germany leading the fight against corruption worldwide. Its 2010 Corruption Perceptions (CP) surveys show that nearly three quarters of the 178 countries in the index score below five, on a scale from 10 (highly clean) to 0 (highly corrupt). With a 6.8 score, Spain is today the 30th country in lowest corruption levels in the world (Transparency International, 2010). Previous surveys carried out by the agency point out that Spain has shown variations during the last few years: from a score of 7.1 in 2004 it plunged to 6.5 in 2008. Such figures, overall, actually mean a relative improvement in relation to more embarrassing previous results: in 1995, at the height of corruption, the country scored 4.35 out of 10. Still, the perception of the Spanish public opinion is that corruption occurs on a large scale in this country, and that Spain is still below advanced capitalist democracies and closer to the levels of corruption of developing nations with authoritarian regimes (Lapuente, 2009). Indeed, the relationship between corruption and governance is perceived as a problem of massive proportions in Spain, worsened by the building boom that has massively carpeted most coastal spots in the Peninsula with concrete. The interval since the advent of democracy in 1976 and the integration into the European Union saw a booming economy where power was decentralized and town halls were ostensibly run like personal fiefdoms by major civil servants and senior officials, who reclassified rural land for urbanization and construction purposes (Jiménez, 2009). In these years, buyers would make fortunes developing and selling the now urban land, then paying officials for their services with properties in the development. As a consequence─a culture of distrust has developed between society on the one hand, and its politicians and the jobs-forlife civil service lobby, on the other (Heywood, 2007; Jiménez, 2009). The situation has been -315-

further deteriorated by an impoverishing monetary crisis that, in Spain, has everything to do with bricks and mortar, since permissive laws have governed the real estate sector for a long time. But even if the indexes of TI surveys have become more sophisticated and demanding over the years, rendering more exacting results, some changes are taking place in Spain. The most salient of these have much to do with a new Land Law in force, encouraging transparency and control over urban development, as well as other changes in the legal panorama. Such changes have apparently been brought about by electorate punishments on the corruption cases that have come to light, materializing into legal measures like the creation of an Anticorruption fiscal and the recent passing of a newly reformed Criminal Code in December 2010, in effect since June 2011. The new code has been inspired by the attempt to implement the Criminal Law Convention on Corruption (ETS 173) of 10 May 2005, together with the Framework Directive 2003/568/JAI, which fight white-collar and real estate crimes, among others (Freshfields Bruckhaus Deringer LLP, 2010). But the work of the Spanish media has been monumental in this process, since it has always exerted a huge influence on public opinion, as it has repeatedly pounced on potential scandals of the alleged sleazy acts by politicians and public officials (Díaz-Herrera & Tijeras, 1996a; 1996b; Jiménez, 2000). Precisely, the goal of this paper is indeed to explain whether the way in which public opinion is informed in Spain matches the reality posited in the new Penal Code that the Spanish legislature has put into effect, which is mainly directed to the implementation of monumental reforms in the pursuing and punishing of corrupt behaviour of public officials and individuals (see Table 1). Our primary hypotheses are threefold: 1.

First, we hypothesise that the press echoes corruption scandals, spreading them and making them known to the public through the usage of highly specialised legal terms, or ‘terms of art’, in the field.

2.

Our second hypothesis is that, among the phenomena described and spread by the press, there are some which are more frequently mentioned than others. In other words, the press makes more significant or visible some phenomena as more relevant in their account of the events.

3.

The technical and hard-and-fast way in which the Code and the international instruments represent corrupt behaviour must sometimes differ from the way in which the official press treats corruption phenomena. Accordingly, a set of more general versions of the terms of art or legal terminology in the Code will be observable in the corpus.

2. Choosing and analysing the lexicon of Spanish corruption: corpus, taxonomy and method This paper reports on the analysis of an ad-hoc 500,000-word corpus gathered from newsitems in the digital version of two prestigious online newspapers: El Mundo and El País. Politically these represent the right and left wing, respectively. We intend to lexically illustrate the words deployed by the Spanish news to convey this information to the general public, and how they relate to the way in which the new penal law organizes the nomenclature of the acts of misconduct committed by government officials to attain -316-

illegitimate private gains. The corpus was gathered by means of Internet searches for news relating to the names of the scandals as nicknamed by the media, namely Malaya, Ibátur, Gúrtel, Barraca, Pretoria, Polop, Ninette, Umbra and Brugal. Most of the names for the cases are infamous in the country, and in most cases were assigned by the judicial agents that filtered ─purposefully or not─ the summary files of the cases to the press. Specifically, our sample period spans from January 2010 to April 2011. In order to handle our corpus, we needed a taxonomy of words, as a framework to work from. The purpose was to have the legislation on the matter as a background, and we studied the relevant instruments passed both in the international and national context. Indeed, we found out that the European Council Group of States against corruption (GRECO) undertook three exhaustive assessments from 2001 to 2009 about the situation in Spain, regarding its compliance of Criminal Law Convention on Corruption (ETS 173), Articles 1─6 of its Additional Protocol (ETS 191) and Guiding Principle 2 (criminalisation of corruption). More recently, it has endeavoured a recent tracking of compliance with the recommendations they issued. These analyses were carried out by the institution within the EU to ensure the compliance on the part of Spain with the requirements deriving from the provisions set in the UN Convention. It has exerted a major influence on the way in which the Criminal Code has been reformed, and the inspiration upon which its lettering has been worded. Therefore, even if there have always existed difficulties in establishing what is meant by the concept (Heywood, 2007), the attempts to measure the scale of corruption by various agencies, like the United Nations, the European Council, as well as the work developed by other international institutions like the IMF, the World Bank and the OECD (Heywood, 2007:695), have been significant in our task to delimit the offences and their place in the Code. Below we provide an account of the major crimes in the new Criminal Code, following the principles upon which the UN Criminal Convention and GRECO map out the area of white-collar corruption. As can be seen in our tables below, corruption of public officials is regulated by Titles VIII (On Falsehood), XIII (On Property and the Socioeconomic order), XVI (Regarding Urban Planning and Heritage) and, mainly, XIX, specifically controlling jobbery matters (On Public Administration). Together with the technical term for the offence, several data are supplied in our tables, such as the Title and Chapter and Articles where the crime is to be found and the less technical, more general version of the offence, as used by the sources consulted. Also the Articles which have been reformed in the new Spanish Code, in line with GRECO and the UN Convention normative, are highlighted (Valero, 2011). Table 1: Title VIII, On Falsehood OFFENCE

ENGLISH TRANSLATION Documentary falsehood

CHAPTER

ARTICLES

Falsedad documental

POPULAR VERSION Documentos falsos

II

390-394

ARTICLES REFORMED 392

Cobro de facturas falsas

Cobro de facturas falsas

False billing

II

395-6

none

-317-

Table 2: Title XIII, On Property and Socioeconomic Order OFFENCE Blanqueo de capitales

POPULAR VERSION Blanqueo de dinero

ENGLISH TRANSLATION Money laundering

CHAPTER

ARTICLES

XIV

301-304

ARTICLES REFORMED 301, 302

Table 3: Title XVI, On Urban Planning and Public Heritage OFFENCE Crimen contra la ordenación del territorio y el urbanismo

Crimen contra el patrimonio

POPULAR VERSION Construcción ilegal/irregular; Delitos urbanísticos; Adjudicaciones irregulares; Irregularidades en la construcción

ENGLISH TRANSLATION Against land and urban planning/Urban planning crimes

CHAPTER

ARTICLES

I

319, 320

ARTICLES REFORMED 319, 320

Prevaricación urbanística

Against the public heritage

II

321-324

none

Against natural resources and environment

III

325-331

325, 327, 328, 329

Irregularidades en la concesión de licencias Crimen contra los recursos naturales y el medio ambiente

Delitos medioambientales

-318-

Table 4: Title XIX, On the Public Administration OFFENCE

POPULAR VERSION Corrupción política/ administrativa

ENGLISH TRANSLATION Administrative breach of trust, jobbery

CHAPTER

ARTICLES

I

404-406

ARTICLES REFORMED none

Cohecho (activo, pasivo e impropio)

Soborno, Recibo de sobornos Cobro de comisiones

Bribery, Corruption

V

419-427

419-427

Tráfico de influencias

Trato de favor; Venta a bajo precio

Trading in influence

VI

428-431

428-430

Malversación (de fondos, de caudales públicos)

Desvío de fondos (de fondos públicos)

Embezzlement and misappropriation (of funds, of public funds)

VII

432-435

none

Fraude y exacciones ilegales

Estafa Cobro de impuestos

Fraud and illegal levying

VIII

436-438

436

Negociaciones y actividades prohibidas a los funcionarios

Negociaciones prohibidas

Illegal partaking of public officials in business activities or contracts

IX

439-444

439

Corrupción en transacciones comerciales internacionales

Corrupción en el comercio internacional

Corruption in international commercial transactions

X

445

445

Prevaricación administrativa

Table 5: Title XX, Crimes Against the Administration of Justice OFFENCE Prevaricación judicial

POPULAR VERSION Corrupción judicial

ENGLISH TRANSLATION Judicial breach of trust, jobbery

CHAPTER

ARTICLES

I

446-449

ARTICLES REFORMED none

From the nomenclature above, we narrowed our search to the following terms of art, selected through our scrutiny of the Spanish Criminal Code, as inspired by GRECO and the UN Convention against Corruption. As we look for the impact of the terms in our corpus, we will also seek to compare the results with those obtained from their more popular equivalents. The selection of the more general version of the term of art was made at random, through our common knowledge of usage, or our assumptions on public opinion usage. In some cases, like crimes ruling urban planning or bribery (cohecho), we presume that they have more than one unofficial version, due to the nature of the crime itself, the versatility in which it may be developed, and, probably, the stature they take in the speakers’ mind, given the importance of the events themselves: -319-

Table 6: Taxonomy of words for the corruption in Spain TERM OF ART 1. Falsedad documental 2. Cobro de facturas falsas 3. Blanqueo de capital/es 4. Crimen contra la organización del territorio y el urbanismo

5. Crimen contra los recursos naturales y el medio ambiente 6. Prevaricación administrativa 7. Cohecho

GENERAL VERSION/S Documentos falsos Cobro de facturas falsas Blanqueo de dinero Construcción ilegal/irregular Delitos urbanísticos Adjudicaciones irregulares Construcción irregular Delitos medioambientales Corrupción política/administrativa Soborno Recibo de sobornos Cobro de comisiones Trato de favor Venta a bajo precio Desvío de fondos (de fondos públicos) Estafa Cobro de impuestos Negociaciones prohibidas

8. Tráfico de influencias 9. Malversación (de fondos, de caudales públicos) 10. Fraude 11. Exacción ilegal 12. Negociaciones y actividades prohibidas a los funcionarios 13. Corrupción en transacciones comerciales internacionales 14. Prevaricación judicial

Corrupción en el comercio internacional Corrupción judicial

In order to perform a direct comparison of the terms at hand, we have firstly measured the absolute frequency of the terms of art and their corresponding general versions in the corpus by means of Wordsmith Tools 5.0. For the analysis of the data obtained, two significance tests have been performed in order to evaluate our working hypotheses: Wilcoxon signedrank and marginal homogeneity tests. The former is a non-parametric statistical hypothesis test used when comparing two related samples or repeated measurements on a single sample to assess whether their population means differ; in other words, it is a paired difference test. On the other hand, the latter tests whether combinations of values between two paired ordinal variables are equally likely. Both tests have been performed with the statistical package SPSS 15.0. 3. Results and discussion Beginning with our first hypothesis, Table 7 shows that the total frequency of the terms of art is four times that of their general equivalents, which confirms the spread of highly specialised terms for the journalistic description of corruption scandals. In addition, the marginal homogeneity test (see Table 8) shows an asymptotic significance of 0.032. Thus, the null hypothesis can be rejected, since p-value is less than the significance level α—as usual in Linguistics, this has been set at 0.05. This demonstrates that the values of the two paired ordinal variables analysed, namely the terms of art and their corresponding general versions, are significantly unequal. Accordingly, the absolute numerical superiority of the first set of terms supports the hypothesis that journalists have a clear preference for specialised terminological units in the matter of corruption, since the distribution of both sets in the corpus is skewed.

-320-

Table 7: Absolute frequency of the search terms in the corpus TERM OF ART Falsedad documental Cobro de facturas falsas Blanqueo de capital/es Crimen contra la organización del territorio y el urbanismo

Crimen contra los recursos naturales y el medio ambiente Prevaricación administrativa Cohecho

ABSOLUTE FREQUENCY 105 21 129 0

GENERAL VERSION Documentos falsos Cobro de facturas falsas Blanqueo de dinero Construcción ilegal

ABSOLUTE FREQUENCY 2 21 8 1

Construcción irregular

1

Delito/s urbanístico/s

12

Adjudicaciones irregulares

7

0

Delito/s medioambiental/es

0

15

Corrupción política

22

Corrupción administrativa

2

Soborno

77

Recibo de sobornos

0

Cobro de comisiones

22

Trato/s de favor

53

Venta a bajo precio

0

Desvío de fondos Estafa/s Cobro/s de impuestos

30 32 0

411

Tráfico de influencias

188

Malversación Fraude Exacción ilegal

200 200 5

Exacción fiscal

2

Exacción legal

1

Negociaciones y actividades prohibidas a los funcionarios Corrupción en transacciones comerciales internacionales Prevaricación judicial

0

Negociaciones prohibidas

31

0

0

2

Corrupción en el comercio internacional Corrupción judicial

TOTAL

1279

TOTAL

322

1

Table 8: Contrast statistics: marginal homogeneity test

Different values Non-diagonal cases Observed MH statistics MH statistics mean MH statistics standard deviation Typified MH statistics Asympt. Sig. (bilateral)

Term_of_art – gral_version 17 11 1258.000 779.500 223.322 2.143 .032

Indeed, our results show that the highly specific terminology, through which the Code organizes misconduct, is introduced in the realm of everyday popular culture through the usage and diffusion of jargon by the news-items of the periodicals. Such phenomena would be otherwise unknown but for the resonance that they have in the press, and the tremendous sensitivity with which such press reacts towards public officials committing misdemeanours. As shown in Table 7, the favoured terms are those included in Title XIX of the Criminal Code, i.e. crimes against the public administration, which, thus, seems to be the area of deepest concern of the press or of public opinion. Among those, there is one term, or crime, which stands out dramatically from the rest, this being bribery to public officials, formally designated in legal Spanish as cohecho, and popularly named with the terms -321-

soborno or cobro de comisiones. Although bribery, or cohecho, is the most prominent term in our corpus, there are some others which statistically occur more frequently than expected, notably malversación (embezzlement), fraude (fraud and illegal levying) and, finally, tráfico de influencias (trading in influence). Other terms that designate crimes in the area of falsehood or crimes against the property occur not so significantly, but in a sizeable amount, such as falsedad documental (documentary falsehood) and blanqueo de capitales (money laundering). Finally, in other cases, where, incidentally, the crime is not designated by just one noun but consists of a longer nominal group, such as crimes against urban planning or against natural resources, no coincidences were found in the corpus. The exceptions are where they appear in their shorter, more general or colloquial version of delitos urbanísticos (urban crimes). Furthermore, our findings also support our second hypothesis: some of the legal phenomena under study are more frequently reported by journalists, not only in their technical form, but also with their more general label. The statistical significance of this trend has been checked by means of the Wilcoxon signed-rank test, as outlined in the previous section. This quantitative analysis indicates that the asymptotic significance in the comparison of the paired terms is nearly meaningful (Z=-1.867; p=0.062). Finally, our observations do not support our third hypothesis, the treatment of which is more qualitative in nature. Our initial belief was that, despite journalists' preference for terms of art, the presence of a sizeable fraction of the popular versions conveys how the press, at times, uses more expressive, versatile and imaginative ways than legislation to render reality and trigger the attention of public opinion at large. In reality, and contrary to our hypothesis, the corpus shows a consolidation of very highly specialised, technical terms outside the domain of general language, as opposed to equivalent words of general knowledge. This is especially true of the most frequent terms in our corpus, those designating crimes within the public administration, namely cohecho, malversación, fraude and tráfico de influencias, which show a much more dramatic appearance in our journalistic texts, in contrast with their more colloquial, everyday versions within the common knowledge such as soborno, desvío de fondos, estafa and trato de favor. The opposite pattern is observable, though, when crimes against the property or the environment—whose longer, more technical version shows no appearance at all in the corpus—are represented by their shorter, more general, wide-embracing term such as ‘urban crimes’, delitos urbanísticos. The remainder of the terminology relating to crimes of corruption makes no significant appearance whatsoever in our results. 4. Conclusion All in all, our paper has endeavoured to explain the extent to which the public is meant to absorb the perception of crimes being committed by public officials by virtue of the press. Secondly, we aimed to scrutinise our corpus for evidence of the most visibly reported crimes, and thirdly, in doing so, we sought to discover whether formal, technical terminology is being used, or whether crimes are expressed by means of their more popular, widespread labels. Some further conclusions, parallel to the ones related to lexical choice, may be drawn. For instance, in our corpus there is a higher incidence of particular terms describing crimes than others. This may show that, within the limited scope of this research, the press appears to be involved in those crimes related to public administration. In comparison, terms relating to other crimes (on falsehood, against property) are moderately mentioned. Finally, crimes regarding the environment and illegal construction, in the small number of examples where they appear, are represented by less specialised, less technical labels. A further study, -322-

involving a larger volume of data and a wider variety of periodicals, may show more conclusive findings, since (to our knowledge) no account has been rendered by legal or sociological sources of the most frequent crimes as committed in the area of corruption in Spain. Our work constitutes a first effort to deploy corpus tools to analyse the fabric of the language of corruption. This effort is directed towards studying how the mass media is used as an influencing instrument regarding the latest political and legal developments in Spain, as far as dishonesty of public officials and politicians is concerned. Our corpus is still too limited to philosophise on the matter, but, in some ways, our limited findings show that the concern of the press, and consequently the public, is more on the malfeasance of public officials, and their methods and consequences (false documents, manhandling of public property), rather than illegal construction and other environmental matters, which seem to be absent from our corpus. There seems to be a bi-directionality here, in the sense that this concern about some specific crimes, or the lack of it, by the press, may alternatively reflect public opinion in Spain, as far as corruption is concerned. This work is only in its inception. We fully acknowledge the limitations of the present study, since more work is called for on this fascinating area of corruption and the press in Spain. We are, therefore, undertaking further analyses embracing larger corpora for more significant results and conclusions. References Díaz-Herrera, J. & Tijeras, R. (1996a) El Dinero Del Poder. La Trama Económica en la España Socialista. Madrid: Cambio 16. Díaz-Herrera, J. & Tijeras, R. (1996b) El saqueo de España. Madrid: Temas de hoy. Freshfields Bruckhaus Deringer LLP (2010) Annual http://www.freshfields.com/publications/pdfs/2010/July10/28417.pdf

Report.

Heywood, P. (2007) Corruption in contemporary Spain. PS: Political Science & Politics 40: 695—699. Jiménez, F. (2000) El caso Matesa: un escándalo político en un régimen autoritario. Historia y política: Ideas, procesos y movimientos sociales 4: 43—68. Jiménez, F. (2008) Un balance de la lucha contra la corrupción en la VIII legislatura. In J. Estefanía (ed.) Informe Sobre la Democracia en España/2008. Madrid: Fundación Alternativas. Jiménez, F. (2009) Building boom and political corruption in Spain. South European Society and Politics 14(3): 255—72. Lapuente, V. (2009, March 27) ¿Por qué hay tanta corrupción en España? [Editorial]. http://www.elpais.com/articulo/opinion/hay/corrupcion/Espana/elpepuopi/20090327el pepiopi_12/Tes Transparency International (2010) Annual Report. http://www.transparency.org/policy_research/surveys_indices/cpi/2010/results -323-

Valero, J. (2011) Índice Código Penal Español 2011 Actualizado. http://despachoabogados.fullblog.com.ar/indice-codigo-penal-espanol-2011actualizado.html

-324-

Wikiforensics and the transcoding of innocence and guilt in the Meredith Kercher case Maria Bortoluzzi1 and Lieve Gies2 1 University of Udine, Italy [email protected] 2 University of Leicester, UK [email protected] Abstract Focusing on the murder of the British exchange student Meredith Kercher in the Italian university town of Perugia, this paper studies the way in which social media have created a forum for ‘wikiforensics’ through which detractors and supporters, with the help of (self-declared) experts, speculate extensively about the guilt and innocence of the accused. Employing a theoretical framework that includes critical discourse analysis and metaphor studies, we look at how social networks seek to connect with forensic-legal evidence through lexical fields which denote dirt and pollution but which also act as a powerful proxy for the moral character of the suspects and the probity of forensic and legal procedure. Our preliminary findings raise questions about the significance of race and gender in online character constructions. Keywords:

SOCIAL MEDIA; CRIMINAL JUSTICE; COMMUNITIES OF PRACTICE; CRITICAL DISCOURSE ANALYSIS; METAPHOR STUDIES

Introduction When the British exchange student Meredith Kercher was murdered in the Italian university town of Perugia in late 2007, it soon emerged that one of the co-accused was her flatmate Amanda Knox, a US citizen. It was not long before the press in Britain, Italy and the US took an intense interest in the case. This being the age of Facebook and MySpace, the online world immediately followed suit. Dubbed ‘Foxy Knoxy’ by the British tabloid press, Amanda Knox quickly emerged as the main protagonist in media discourse, overshadowing her co-accused—Italian former boyfriend Raffaele Sollecito and Ivorian citizen Rudy Guede—and even the victim herself. Salacious stories about Knox’s sexual proclivities dominated the coverage. Her media notoriety continues to this day, even following her release on appeal in October 2011. Internet discussion of the case largely mirrors coverage in the mainstream media: it is Knox who commands most of the attention online. Guede, by contrast, is the least talked about. One of the most eye-catching features of relevant Internet discussions is the role of the amateur-expert who engages with complex legal and forensic evidence either to express a firm belief in Knox’s innocence or express the conviction that she is guilty. Online commentators are not in a position to examine any actual forensic evidence but are merely limited to making metadiscursive statements about documents which were produced in the legal context of prosecution, trial and appeal. This paper offers an explorative study of what we have termed ‘wikiforensics’ in relation to the Kercher case. We define wikiforensics as the activity whereby Internet users, as a pastime or on a freelance basis, extensively ponder the evidence, motives and modus operandi of the suspects. It also involves scrutiny and criticism of the workings of the criminal justice system. We aim to examine how innocence and guilt are constructed in online communities. Our preliminary findings suggest that the most ardent followers of the case relate to the legal-forensic process by constructing narratives which specifically strengthen their convictions. The structure of our paper is as follows: first, we will briefly outline some of the contextual elements that are relevant to our analysis. Secondly, we aim to show how online communities are -325-

starkly divided into two opposing camps, involving the so-called ‘guilters’ and ‘innocentisti’. Next, we explore how social networks seek to connect with the forensic-legal evidence through lexical fields which denote dirt and pollution, but which also act as a powerful proxy for the moral character of the suspects and the probity of forensic and legal procedure. Context If social media generally are becoming a player in the pursuit of justice, for example, by putting pressure on authorities to investigate and prosecute crime (Campbell and Stormark, 2011), wikiforensics represents a distinct category of Internet sleuthing which involves a form of forensic crowd-sourcing where users pool their knowledge with the aim of solving problematic crime cases. As we illustrate in this paper, some online communities emerge with the sole purpose to ponder forensic evidence and add their voices to the debate about the guilt and innocence of suspects. As one poster comments on Perugiamurderfile.org (PMF)1, one of the websites we are researching: ‘One thing about the internet; it makes everyone instant science experts’ (PMF, 8 Posted: Fri Oct 29, 2010 9:37 pm). The statement represents an interesting take on the notion of user-generated content and highlights the tension between the amateur and the expert. The label of ‘the instant science expert’ suggests minimal training and expertise but also signals a potential void in terms of professional ethics and the need for distance and objectivity. In the case we are studying, the quality of the wikiforensics is extremely variable, with users’ input ranging from wild speculation and aggressive stance-taking to in-depth scientific analysis and meta-analysis. While their professional background is hard to verify, some participants appear well qualified and experienced in the various matters they debate online. There can be little doubt that frequent posters on relevant websites are heavily invested in the Kercher case. The case has stirred up intense passion and controversy across the entire media ecology: journalists, scientists and an Italian MP have all written extensively about the case, resulting in the publication of several books. With the key participants originating from three different continents, the case is a transnational media event, generating widespread coverage in US, UK and Italian media. Guede was the first to be convicted for the murder and his sentence was reduced on appeal from 30 to 16 years. Knox and Sollecito, who were romantically involved with each other at the time of the murder, were convicted in a second trial: in December 2009, they were sentenced to 26 years and 25 years respectively. A lack of clarity regarding the circumstances in which Kercher was murdered adds an element of mystery to the case and undoubtedly increases its appeal. In February 2011, the US television network Lifetime screened a movie based on the case, much to the dismay of both the Kercher and Knox families who each threatened legal action to stop the movie from being broadcast. Online communities as communities of practice: the ‘guilters’ versus the ‘innocentisti’ Two online communities are the focus of our study: Perugia Murder File (PMF) and Injustice in Perugia (IP)2. While there are many more websites dedicated to the Kercher murder case, including accounts on Twitter, Facebook and a much disputed entry on Wikipedia, PMF and IP represent the two websites that are most closely preoccupied with the forensics of the case. The two online communities are heavily polarised: PMF followers are convinced that Knox is guilty, while for IP posters, she is the victim of a gross miscarriage of justice. The websites’ banners immediately give an idea of their distinct identities. The PMF banner highlights the connection between the ‘Given’ information of the city of Perugia, represented as a sort of new age place, and the ‘New’ information, Kercher’s smiling face, which is immersed in the 1 2

http://perugiamurderfile.org/ http://www.injusticeinperugia.org/

-326-

atmosphere of a dark tale. IP offers a more sober banner in which the ‘Given’ information is the main stance of the website (‘injustice’, ‘wrongful conviction’) and the ‘New’ information is contained in the faces of Knox and Sollecito. Interestingly, their gaze vectors avoid direct eye contact with each other and with the viewer (Kress and van Leeuwen, 2006).

Figure 1: Perugia Murder File banner

Figure 2: Injustice in Perugia banner

The two groups of users are communities of practice (CoP) who share competence/s and a commitment to a domain of interest, engage in joint activities and share a repertoire of resources. These CoPs of amateur-experts gain skills and knowledge for the whole community using blogs and discussion groups in a participatory way (Wenger, 1998). There is significant tension and antagonism between the two amateur-expert communities we are analysing: while they share the same focus of interest, their agenda is completely different and they compete to make themselves heard online. They are polarized into two distinct factions, with each claiming to search for the truth about Kercher’s murder: IP define themselves as ‘the people who need to understand and not to be manipulated’ and PMF as the community which ‘has truth on their side’. What sharply divides these two communities is their unwavering belief that Amanda Knox and Raffaele Sollecito are either guilty as charged (PMF) or entirely innocent (IP). The two communities refer to each other through the pejorative terms of ‘guilters’ and ‘innocentisti’. ‘Truth’ in both communities becomes a matter of distrust or faith in the moral character of the main participants. At the heart of each community of practice is a core group of users who are more active and appear to have a stronger influence on topics and textual choices for their community. In PMF, discussion threads are hybrid text-types of written highly informal exchanges similar to face-to-face conversation (see Baron, 2008 for a discussion of online text-types). The following posts illustrate the subtle hierarchy at work in PMF: a core group of participants considers themselves to be ‘genuine’ PMF members, suggesting that they have forged a bond that relegates other users to the status of outsiders: Extract 1: PMF, Posts: 1524. Posted: Sat Oct 30, 2010 11:55 pm

Thank you, P.! At the risk of sounding like an Over-the-Top Mutual Admiration Society.... (but, hey, what's wrong with that?!)

-327-

Extract 2: PMF, Posts: 572. Posted: Sun Oct 31, 2010 5:53 pm

A virtual group hug to you from all genuine PMF members. In IP, by contrast, the tone of the exchanges is significantly more formal and there is much less focus on communal bonding. The text-type is carefully edited and single authorial voices give coherence to stretches of texts in the website; posters openly refer to themselves as ‘experts’ and as authors. ‘Bruce Fisher’ owns the copyright of the contents of the website and introduces himself in the third person: Extract 3: 2010, IP, About Us, InjusticeInPerugia

Bruce Fisher created Injusticinperugia.org. Bruce followed the Meredith Kercher murder case from early on and could not believe what he was seeing. He quickly realized that two innocent people were being wrongly accused of murder. He was hopeful that the truth would prevail in the end and Amanda Knox and Raffaele Sollecito would be freed. After the ruling finding Amanda and Raffaele guilty, He felt compelled to find a way to help. This community of practice portrays itself as a group of experts rather than amateur-experts: Extract 4: 2010, IP, About Us, InjusticeInPerugia

He [Bruce Fisher] has helped to bring together an excellent group of people including various experts that have all worked very hard to provide the information that is available on Injusticinperugia.org. Elsewhere on the website Fisher openly voices his ideas and evaluations in the first person singular deixis, clearly presenting himself as the main authorial voice. Despite PMF’s informal atmosphere, its members too present themselves as experts. One important form of professional expertise which this CoP lays claim to is language and translation skills. A group of people belonging to the PMF community, realising the importance of the sentencing report by Judge Massei (hereafter MS or, when mentioned in translation, MST), decided to collaborate to produce a crowd-sourced translation. The translators identify themselves as follows: Extract 5: 2010, MST, 3

This translation was done by a group of unpaid volunteers who are regular posters on the Perugiamurderfile.org message board devoted to discussing the murder of Meredith Kercher in Perugia, Italy, in November 2007. The virtual community of translators defines itself by the communicative task they undertook. The collocation ‘unpaid volunteers’ emphasises the non-lucrative nature of the task the translators set for themselves, a striking form of self-characterisation which appears to suggest that their pursuit of truth and justice is placed above any kind of commercial interest. We know little about the identity of the translators: they define themselves as belonging to the main community of ‘regular posters’ on PMF which hosts the translation. The connection between the document and PMF is thus explicitly stated. The Massei sentencing report The Massei sentencing report (MS) represents only one of the official documents about the case, but it is of special interest to us because (at time of writing, August 2011) it is the only official -328-

document made available online in English. Discursively, it is a polyphonous whole of voices blended into the narrative of the judiciary in the person of Judge Massei who presided over the trial of Knox and Sollecito. MS constitutes the linchpin in the case. It contains a wealth of forensic details and it is the focus of attention in both PMF and IP. Following an independent first reading of MS in translation and in the original Italian, we noticed repeated references to cleaning, cleanliness and dirt in relation to two of the accused, namely Rudy Guede and Amanda Knox. The references are generally related to actual events, forensic evidence and witness reports. However, lexical priming, discoursal proximity and connection of these semantic areas with Guede and Knox (but less with Sollecito) are rather remarkable. We followed our initial readings up with a manual analysis which showed that in MS, the report of forensic analysis of data and the interrogation of witnesses and interpretation of the events relate Rudy Guede to the semantic field of cleanliness or, rather, uncleanliness in a prominent way. Our method involved identifying significant areas of semantic interest. We focused on the semantic areas of cleanliness and dirt which we believe to be highly relevant in the portrayal of guilt and innocence on the three texts we are studying: MS (165486 words)3, a discussion thread from PMF (56334 words) and a further sample drawn from IP (homepage and all main sections for a total of 46607 words). For each of the three texts, we have examined the following aspects: entities (cleaning materials, dirt, materials that become ‘dirt’ because they are out of place, places in which processes of cleaning or its opposite take place), attributes (related to cleanliness and dirt), processes (cleaning, polluting, contaminating), relationships (causal links between cleanliness and dirt, results of cleaning processes) and lexical priming. We looked at how the two CoPs intertextually transpose forensic and legal discourse to suit their own communicative agendas. We used WMatrix software (Deignan and Semino, 2010; Rayson, 2008) to verify the most frequently used semantic fields in the Massei Sentence (with significant difference of log-likelihood value above 7 for the USAS categories used in WMatrix)4 as compared to the Written Institutional Texts of the British National Corpus (BNC). Our main focus is the introductory section of MS which deals with setting the scene and the people involved in the event (38392 words). Thus, we were able separate the most highly technical jargon of the different fields of forensic investigation (see further below) from the more general legal discourse of the sentence. In the list of the fields that score over 100 of log-likelihood, some are clearly related to the text-type and the type of crime, but one stands out: the category of cleaning and personal care. It has a low occurrence in absolute terms, but it has a significant high log-likelihood of 124.70 compared with the BNC institutional texts. Focusing on the way in which Rudy Guede is discursively positioned in the report, we observed that his presence at the crime scene left behind biological traces which are related to dirt and disorder: Extract 6: 2010, MST, 41—42

Rudy had shown a certain interest in Amanda. That evening Rudy went to the bathroom leaving the door open; he fell asleep on the toilet and left excrement in the toilet bowl. Extract 7: 2010, MST, 44

Further biological traces of Rudy Guede were found on the toilet paper taken from the toilet of the larger bathroom. The faeces present in the toilet of that bathroom did not, however,

3

4

38392 words in introductory section, pp.10—100; 109179 words in investigative section, pp. 101—357; 17915 words in conclusion pp. 358—398. http://ucrel.lancs.ac.uk/usas/

-329-

yield any results, and Dr [S.], the biologist of the Scientific Police, explained that the presence of numerous bacteria easily destroys what DNA might be found in faeces. In the Introductory section of MST (pp. 1—100, 38392 words) out of 16 occurrences of the noun ‘toilet’, 10 are directly connected with Rudy Guede and 4 with Amanda Knox. Out of 47 occurrences of the bathrooms, 10 are connected with the presence of Rudy Guede, 22 with the presence of Amanda Knox, 1 with the presence of both Knox and Sollecito, 4 with the presence of Sollecito, while the remaining 10 occurrences are unrelated to any of them. Thus, the other person who is clearly associated to procedures of cleaning (or lack of them) is Amanda Knox. She is presented as washing herself (12 out of 18 occurrences of ‘having a shower’), but she is also portrayed as being oblivious to the uncleanliness of the bathroom (see example below). Significantly, she is portrayed as untidy and promiscuous, echoing the ‘Foxy Knoxy’ epithet used by British tabloids (see e.g. Churcher, 2007). Witness accounts play an important role in associating Amanda with dirt in a material and sexual sense: Extract 8: MST, 2010: 35

She had also met Amanda Knox on occasion. Meredith talked to her about her relationship with Amanda , telling her what was happening in the house , and said that at times Amanda did not leave the bathroom clean. Meredith had also told her about certain private objects that Amanda had in the house inside a beauty case which she kept in the bathroom: a vibrator and some condoms. Extract 9: MST, 2010: 38

Her relationship with Amanda was good, even though some things that annoyed Meredith. In particular, she was annoyed by Amanda's bathroom habits. Amanda Knox’s account as reported in the sentence is: Extract 10: MST, 2010: 39

She had found the entrance door open and this seemed strange to her: she had gone into the house and into her room and she had taken a shower and had seen drops of blood. She said that after the shower she got dressed and noticed that Meredith’s door was locked. She went into the other bathroom and said that there were faeces in the toilet. Amanda Knox is also presented as a verbal polluter and a manipulator. In the Massei report she is convicted of the offence of ‘calunnia’ as she falsely accused another person of Meredith’s murder. The repeated association of two of the co-accused with dirt and uncleanliness casts them in a negative light. While many of the descriptors may be relevant from a strictly legal perspective, they are a crucial source domain to the PMF community in expressing their hostility towards Amanda Knox. A second category of occurrences related to the semantic area of dirt is ‘contamination’ and (non)contamination. It plays a rather different role in the IP community which is predominantly preoccupied with the contamination of forensic evidence. The central section of the sentence, reporting the technical investigative and forensic activity of the police and the judiciary (pp. 101— 356, 109179 words) has a high occurrence of the lexical items related to the noun ‘contamination’ and its verb ‘contaminate’ (116 occurrences, 0.76 %). We shall see below how the lexical field of dirt/cleanliness and contamination/non-contamination performs a very different role in the discourses of the PMF and IP communities respectively. -330-

Discourses of cleanliness and dirt in the Perugia Murder File As part of our study of the PMF community, we selected a discussion thread which took place in the period after the translation of MS was published and in the run-up to the appeal by Knox and Sollecito in late 2010. In constructing Knox’s character, the references to dirt and cleanliness found in Massei are intertextually transposed by community members to express their moral stance. Since the translation originated in PMF, it can be assumed that several members of this community of practice are intimately acquainted with the language of MS. What, in terms of legal discourse, appears as statements relating to the forensic evidence and witness accounts is transformed into metaphors of guilt by the PMF community. As Kővecses (2010: 246) and Deignan (2010: 45 et passim) remark: ‘metaphorical dirt, or immorality, is, through the metaphor, opposed to metaphorical cleanliness, or virtue.’ (Deignan, 2010: 45). The source domain of lack of cleanliness is mapped in very complex ways onto the target domain of lack of morality. In the following post, for instance, two members (J. and T.) comment on their translation of some of Knox’s writings from Italian into English. Through the translation, play on words and metaphorical use of language, they relate dirt to moral impurity (sickening) so that Knox’s writings become a source of illness and pollution through her stories, fantasies and lies: Extract 11: PMF, Posts: 1524. Posted: Thu Oct 28, 2010 5:33 pm

Post subject: Toxic C. wrote: J. and T.: you may need to have rinse out your frontal lobes with more than a couple of spin cycles with all of that literary translating and scene dialogue going on! I'm glad you've done it! Caro C.: Not to worry! I ran to the Conad early this morning to purchase some bleach to mop up5 the black matter forming in my grey matter. No more impure thoughts coming from T.!!! I'm much more concerned about J., who's had FAR MORE INTENSE, PROLONGED exposure to those dangerous AK toxins. That J. is a tough cookie, but, really, how much can one person take?!!! You don't need a microbiologist to tell you that the not-so-onorevole “Raucus GirlRandyum” pathogen is extremely sickening and quite toxic!!!!!!!! blehhhhh J. would be better off drinking cone snail venom!!!! I only hope that Drive-by-Doc is on call-- to render medical assistance for J., should our brave Toxin Translator succumb... e speriamo di no.... The cleaning actions attributed to Amanda Knox in MST are repeatedly mentioned in the thread and become a clear accusation of murder. Investigative detail is transformed into moral judgement and evidence of guilt: Extract 10: PMF, Posts: 169. Posted: Fri Oct 29, 2010 10:27 am

I bet the bidet is a difficult one for Amanda. Every time she sees it she is reminded of washing off Meredith's blood. The following passage exemplifies many of the images used to represent immorality and guilt in the thread. The issue under discussion here again is Amanda Knox’s writing; the source domain of dirt 5

This reference goes back to the Massei sentencing report: the morning after Meredith’s murder, Amanda Knox allegedly went to the local supermarket Conad to purchase bleach to clean up traces of blood.

-331-

triggers a cluster of metaphorical expressions of disgust not only for her writing but also for her as a person: Extract 11: PMF, Posts: 1524 Posted: Fri Oct 29, 2010 4:34 pm

This is not AK about employing some literary technique—it’s a pathology, a disease. AK’s got a bad case of: Cacoethes scribendi Merda = an insatiable urge to write. KAK.... Evidently, even after THREE YEARS.... Anita STILL hasn’t learned TO FLUSH her KAK down the tubes. (PMF, 1524 Posted: Fri Oct 29, 2010 4:34 pm). The semantic referent of an inability to flush the toilet (attributed to Rudy Guede in MST) is here shifted onto Amanda Knox (who is nicknamed Anita). Her written output is equated with human waste (in a pun on her own initials, KAK) which is not only seen as immoral but also as sickening and pathological. Her lies and manipulation signify her status as a verbal polluter. Her writing and speaking are also referred to as ‘Knonsense’ and ‘Knoxsense’ (again an example of verbal disorder and pollution). Her writing is described as rambling stories, lies and fantasies overlapping in a representation of immorality and insanity. The following post, quoted verbatim, epitomises Amanda Knox’s character construction in PMF: Extract 12: PMF, Posts: 80. Posted: Thu Oct 28, 2010 5:27 pm

Judges: 1) ...a danger to the public without any inhibitions, disposed to follow any impulse even leading to violent conduct... 2) ...histrionic...an absolute liar... 3) ...restless, does not disdain ‘multiple frequentations’... (legalese for ‘sleeps around like a tart’) 4) ...had a ‘negative personality.... 5) ...crafty and cunning...a multi-faced personality unattached personality unattached to reality with an elevated fatal capacity to kill again... Attorneys: 1) Pacelli:...she lies in a blatant and shameless manner...depraved, scheming, murderous... 2) Pacelli:...a promiscuous young woman who enjoyed telling lies....venal attitudes 3) Maresca:...an immoral uncouth degenerate..... Others: 1) Patrick:...an exhibitionist and pathological liar...a vengeful college student...she does not even have a soul. She's empty; dead inside 2) Seattle Friends...she had fallen in with a bad crowd in Italy... 3) Polizia Ficarra:...does what she liked, when she liked without consideration for others 4) Madison: ...she was ‘not ashamed’ of having sex. 5) Robyn:...she literally seemed crazy... 6) Her Parents:...she sometimes lacked common sense... 7) Raffie’s Father:...cursed day son met her... Conclusion: Amanda may have shared some of her ‘traits’ with Scarlett O'Hara, but none of the above with the Virgin Mary or most nuns I have known (PMF, 80 Posted: Thu Oct 28, 2010 5:27 pm) -332-

In the excerpt, the legal text of Massei is extensively quoted and edited. Additional textual support is found in voices from a variety of text-types and sources which are mobilised to help cast Knox’s personality in the most negative light, while no other person involved in the case is discussed in such derogatory detail: the absence of a similarly strong moral judgment in relation to Sollecito (rather affectionately referred to as ‘Raffie’) is most striking. Discourses of dirt and pollution in Injustice in Perugia The portrayal of Amanda Knox in IP stands in stark contrast with the discourse in PMF which echoes the negative publicity in several mainstream media. The authorial voices and the community of IP portray Amanda Knox as a ‘normal’, well-adjusted young woman who is caught up in a terrible miscarriage of justice caused by an incompetent Italian judiciary and police. Her fragile mental and physical condition is largely attributed to the inhuman treatment inflicted on her by the Italian criminal justice system and the stress caused to her during her interrogation, trial and ongoing detention: Extract 13: IP, Amanda Knox, 33

Suffering from extreme exhaustion, after a long and grueling interrogation, twenty year old college student Amanda Knox gave in to the interrogators demands by describing an imaginary dream or vision Extract 14: IP, Amanda Knox, 42

Amanda’s behavior was grossly misrepresented in the trial just like it had been and continued to be in the media. Extract 15: IP, Amanda Knox, 45

Mignini [the prosecutor] described Amanda as a sex-and-drug-crazed sociopath. Remarkably, dirt and pollution still act as a dominant paradigm, but serve a very different purpose from what we have seen in PMF. The roles are reversed: the verbal polluters and manipulators are the media, the Italian legal system and the community of guilters (see IP, Guilters). Amanda is explicitly represented as a victim, which contrasts with her portrayal as main perpetrator in PMF: Extract 16: IP, The character of Amanda Knox, the motive, 111

Amanda was the victim of continual and unscrupulous attacks involving her personality, or what they wanted to believe was her personality. [...] With absolutely nothing to back up the claim, the prosecution described Amanda as a wicked calculator, a manipulator of men and envious at the same time of the purity of the victim. While the Massei report goes to great lengths to refute any criticism of the forensic investigation, IP puts the Italian criminal justice system under intense scrutiny. Note in the following passage the use of the adverbial ‘solely’ to reinforce the accusation that both the system and specific corrupt individuals are to blame for a serious miscarriage of justice:

-333-

Extract 17: IP, The Italian Justice System, 136

The wrongful conviction of Amanda Knox and Raffaele Sollecito is not solely the fault of the Italian justice system. This injustice was committed by a small group of individuals that abused their power to protect their own better interests. The focus should be on those individuals. No system of law is perfect. Unfortunately, corrupt individuals exist in every system. While PMF and IP are bitterly divided over the issue of Knox’s and, by extension, Sollecito’s guilt, there is little doubt as to Guede’s guilt. Echoing the Massei report, Guede is explicitly associated in IP with pollution and dirt: we learn how his genetic material is scattered all over the scene of the murder (yet again, his presence is connected with excrements): Extract 18: IP, Rudy Guede, 102

The evidence against Rudy Guede: Rudy admitted he was in the room. Rudy’s DNA was found all over the crime scene. Rudy's DNA was found in and on Meredith's body. Rudy’s DNA along with Meredith’s blood, was found on Meredith’s purse. Rudy’s excrement was found in the toilet. Rudy’s shoe prints, set in Meredith’s blood, were found in the bedroom and hallway. Rudy’s handprints, in Meredith’s blood, were found on a pillow case in Meredith's room and on her wall. Rudy had a cut on his right hand that was still visible when he was arrested. Rudy fled the country. On this website, in contrast with PMF, the main agents who scatter dirt and pollute evidence (both forensic and verbal) are the Italian police and the justice system. Below is a selection of many examples of the way in which the semantic area of cleaning and dirt creates the impression that the police and forensic team literally contaminated the evidence: Extract 19: IP, Case for appeal, 103

Contamination cannot be ruled out and is also highly likely with the methods used by Dr. [S]. Contamination is further possible by the fact there was no visible presence of biological traces on the blade. Tests also concluded that there was no blood on the blade. TMB testing showed negative for blood. Extract 20: IP, Case for appeal, 104

The court makes two contradictory observations regarding the bare footprints detected with luminol. Extract 21: IP, Case for appeal, 105

Many people walked throughout the cottage in the days that followed the murder. On the day the body was discovered, many people walked throughout the cottage. After the cottage was considered a crime scene many others walked throughout the cottage. Investigators did not change their shoe covers when they walked from room to room. Extract 22: IP, Prosecution case, 79

The blood samples in the bathroom were not collected properly. [P. S.] used the collection swabs like cleaning rags. She wiped large surface areas and also collected from multiple -334-

surfaces with one single swab. She also never changed her gloves. [S.] actually advanced the mixing process right on her swabs. Whereas the underlying accusation is that the forensic investigators were careless, ‘contamination’ is constructed as an endemic problem of the Italian criminal justice system. The semantic field of contamination is expanded to encompass corruption and impurity within the Italian criminal justice system. The Italian police and judiciary are presented as ‘claiming, boasting, insinuating, pretending, coaching witnesses’, verb processes which contain elements of misrepresentation or manipulation of reality. The tables are turned on the officials: the Italian legal system becomes an immoral polluter who behaves unethically and unprofessionally: Extract 23: IP, The character of Amanda Knox. The motive, 112

This imaginative reconstruction by the court has no basis. The failure to find key pathological elements pertaining to Amanda and Raffaele made it necessary for the court to insert additional elements such as ideas of perversion and depravity that controlled the every day life of Amanda. Extract 24: IP, The character of Amanda Knox. The motive, 116

The ruling of September 14, 2009 violates the principles of a fair trial, and rights of the defense. Contamination becomes an issue of the perverse behaviour of the prosecutor and the court (see example above), but it also extends to the language and methods used by the guilters: Extract 25: IP, The character of Amanda Knox. The motive, 118

One of the most notorious guilters is [H. R.] You will find his comments spread all over the internet like a cancer. Discussion and preliminary conclusions Mary Douglas (1966), in her pioneering work on the anthropology of purity and cleanliness, refers to dirt as ‘matter out of place’. This is an apt characterisation of what forensic science is fundamentally concerned with: it is the discipline of matter out of place. It deals with the chaos of displaced objects as well as that of body fluids, fingerprints and DNA at the crime scene. Dirt, however, is more than just matter. As Campkin and Cox (2007: 1) remark, dirt is ‘that which slips easily between concept, matter, experience and metaphor’. What is perceived as dirty and clean is culturally and historically specific. Feelings of disgust and revulsion are readily transferred from matter to people. Dirt plays a significant symbolic role across different cultures: those who pollute are those who transgress established social norms (Douglas, 1966). When considering descriptions of Guede’s toilet use or Knox’s attempts at washing away traces of blood in the Massei sentencing report, these are not simple statements of factual evidence but potentially powerful markers of their guilt and moral impurity. Similarly, the strong defence of the forensics in Massei is an implicit admission of the hypothetical possibility of pollution of evidence which is transformed by the IP community into a signifier of incompetence and corruption. Matter and metaphor collide in the online communities we have been studying. It is hard to see where material dirt ends and symbolic impurity begins: ‘guilters’ transform the presence of Guede’s excrement at the crime scene into a metaphor and a metonymy for Knox’s guilt-ridden and insane writings. In the innocentisti’s discourse, the forensic team’s failure to change gloves is a pars -335-

pro toto for the systemic failures of Italian justice, evoking the mani pulite (‘clean hands’) anticorruption campaign of the early 1990s. As Semino (2008: 33) writes, ‘When particular uses of metaphor become the dominant way of talking about a particular aspect of reality within a particular discourse, they may be extremely difficult to perceive and challenge, since they come to represent the “commonsense” or “natural” view of things’. Our analysis has focused on how seemingly mundane notions of cleanliness and dirt underpin the construction of innocence and guilt. Our preliminary findings raise questions about the significance of beliefs systems involving race and gender. Is it a coincidence that the white, middle-class, Italian-born Sollecito is much less prominently associated with cleanliness and dirt, while the black, foreign-born Guede is extensively depicted in legal and wikiforensic discourse as a polluter? Why is it Knox, not Sollecito, who is extensively portrayed in social media and also to some extent in the Massei sentencing report (‘multiple frequentations’) as promiscuous and hypersexual? How is the victim constructed and has she in effect been marginalised amidst the public outcry? Finally, to what extent does the slippage between forensic contamination and institutional corruption represent legitimate criticism of Italian justice, and to what extent does it harbour a negative cultural stereotype which depicts the Southern European official as untrustworthy, incompetent and corrupt? We believe that critical discourse analysis provides a useful framework for analysing and interpreting the multi-layered stories which online communities tell about a crime which has enthralled so many in cybersociety. References Baron N. 2008. Always On. Language in an Online and Mobile World. Oxford: O.U.P. Campbell, M. and Stormark, K. (2011) Facebook justice: friends of a woman murdered in London are using the web to hound the suspect’s family. Sunday Times April 3, 16. Campkin B. and Cox R. (eds) (2007) Dirt: New Geographies of Cleanliness and Contamination. London: I.B. Tauris. Churcher, S. (2007) ‘Foxy Knoxy, the girl who had to compete with her own mother for men’. Daily Mail, 10 November. Deignan A. (2010) The cognitive view of metaphor: conceptual metaphor theory. In L. Cameron and R. Maslen (eds) Metaphor Analysis. London: Equinox, 44—56 Deignan A. and Semino, E. (2010) Corpus techniques for metaphor analysis. In L. Cameron and R. Maslen (eds) Metaphor Analysis. London: Equinox, 161—179. Douglas M. (1966) Purity and Danger: An Analysis of the Concepts of Pollution and Taboo. London: Routledge. Kővecses Z. (2010) Metaphor: A Practical Introduction. Oxford: O.U.P. Kress G. and van Leeuween T. (2006) Reading Images. London: Routledge. Rayson, P. (2008) From key words to key semantic domains. International Journal of Corpus Linguistics. 13 (4): 519—549. Semino E. (2008) Metaphor in Discourse. Cambridge: C.U.P. -336-

Wenger, E. (1998) Communities of Practice. Cambridge: C.U.P.

-337-

This article has been reprinted from the original source with kind permission from the author: Brooklyn Law Review, 2011, 76:3, 1033—1048

Statutory interpretation, morality, and the text Lawrence M. Solan† Brooklyn Law School, USA [email protected] Abstract In this essay, I wish to explore the question of whether certain approaches to statutory interpretation can be regarded as wrongful. My argument concerns instances in which interpreters take advantage of linguistic accident to license arguments that flout the intent or purpose of a law. Philosopher Bernard Williams calls reliance on literal meaning in this manner “fetishizing assertion,”1 and considers it tantamount to lying. If linguistic practices that rely too heavily on linguistic accident are wrongful, then serious ethical questions present themselves to the legal system. For if we acknowledge the problem, we then are forced to ask ourselves how comfortable we are with a rule of law that cannot rely fully on the law as written to sustain its legitimacy. In this brief essay, I raise these issues, and comment on their relationship to questions of judicial candor in cases concerning the interpretation of statutes. I conclude that especially when there is doubt about meaning, or suspicion that the legislature has erred, it is essential to turn to the purpose of the law in order to avoid the moral consequences of assertive fetishism. I further argue that recourse to purpose, contrary to the views of many, actually reduces the range of judicial discretion, and that those who associate purposive interpretation with judicial activism appear to be subject to a cognitive bias—the conjunction fallacy. Keywords:

INTERPRETATION OF STATUTES, ETHICS, INDETERMINACY, COMMUNICATIVE INTENT

MORALITY,

LINGUISTIC

Morality, speech, and interpretation



Don Forchelli Professor of Law, Brooklyn Law School. My thanks to Ralf Poscher and to participants in the Brooklyn Law School symposium on statutory interpretation for helpful comments. This research was sponsored by a summer research stipend from Brooklyn Law School. 1 Bernard Williams, Truth and Truthfulness 100-10 (2002).

-338-

Lying is an immoral act—at least most of the time. It is prohibited in many legally-relevant contexts. Lying under oath is perjury.2 Lying in business affairs is fraud.3 Lying to government officials is a crime in itself.4 Lying in the course of acting as a lawyer is sanctionable misconduct.5 In some, but not all, of these contexts, misleading another person into believing to be true something the speaker believes to be false is also both immoral and legally prohibited. A truthful, but misleading, statement can be just as much a fraud as a false statement.6 In fact, fraud is defined to include both species of deception. In some instances, it might even seem worse for a person to scheme to misdirect his target through a series of truthful statements than it does to tell an outright lie. While the liar has to take responsibility for his falsehood, the deceiver can feel virtuous at not having said anything false while arriving at the same result: successfully leading someone to believe something to be true that the speaker knows is false.7 For this reason, many moral philosophers draw no distinction between the two.8 Bernard Williams goes even further, denouncing reliance on the truth for moral justification of a fraudulent act as immoral in its own right.9 How does deception work? Like persuasion, deception depends on a change in the state of mind of the hearer (or reader).10 To succeed, the deceiver calculates the inferences that a person is likely to draw from a speech act and leads the target to drawing just the inferences that will accomplish the task of deception.11 The better the calculation, the more likely the deception. And these calculations are easy enough, at least much of the time. It is not at all difficult to deceive, because we are all aware that people tend to draw the inferences we intend them to draw in everyday conversation. In fact, we say the things we do with those inferences in mind. If I ask you, as your guest, “where’s the telephone,” we both understand that I am asking you if I can use the phone, and the assumption is that I will not use it in a way that will cost you money (or at least no more money than our relationship would bear within the bounds of politeness). Neither of us has said any of this, but I know what you will infer before I speak. In essence, we both apply Grice’s cooperative principle, which says that we construe conversations to proceed as a cooperative interaction, drawing whatever

2

18 U.S.C. § 1621 (2006). For definitions, see RESTATEMENT (SECOND) OF TORTS §§ 525-26 (1977). 4 See 18 U.S.C. § 1001. 5 See Model Rules of Prof’l Conduct § 8.4 (2010). 6 See RESTATEMENT (SECOND) OF TORTS § 529 (“Representation Misleading Because Incomplete: A representation stating the truth so far as it goes but which the maker knows or believes to be materially misleading because of his failure to state additional or qualifying matter is a fraudulent misrepresentation.”). 7 See Jonathan E. Adler, Lying, Deceiving, or Falsely Implicating, 94 J. PHIL. 435, 439-40 (1997). 8 See, e.g., T.M. SCANLON, WHAT WE OWE TO EACH OTHER 320 (1998); Jennifer Saul, Lying, Misleading and What Is Said 1013 (forthcoming Oxford Univ. Press) (on file with author). 9 WILLIAMS, supra note 1, at 100-10. 10 Philosopher J.L. Austin refers to this as the “perlocutionary effect” of the statement. J.L. AUSTIN, HOW TO DO THINGS WITH WORDS 108 (J.O. Urmson ed., 1965). 11 This perspective is consistent with the signal approach to communication, described in Cheryl Boudreau et al., What Statutes Mean: Interpretive Lessons from Positive Theories of Communication and Legislation, 44 SAN DIEGO L. REV. 957 (2007). 3

-339-

inferences we need to draw for that to happen.12 You draw those inferences as the hearer, and I adjust what I say around the inferences that I (correctly, we hope) predict that you will draw. Now, let us assume that you and I are negotiating some kind of deal. I know that if you speak with Hannes before signing on the dotted line, you will find out that the deal is unfair to you, and I also know that you want to speak to Hannes because you value his counseling. It is the kind of conversation that you would only have with Hannes in person if he is available. I would rather that you and Hannes not speak. You ask: “Have you seen Hannes recently?” I answer truthfully: “I saw Hannes in Washington last week.” What I didn’t add was that Hannes is now staying at a hotel in New York two blocks from where we are having our conversation, and that I just had dinner with him there last night. With only a little luck, your trust in me will cause you to snap at the bait, and I will be home free. In our story, I have committed an immoral act, whether or not you believe that telling an actual lie would have been an even greater affront (I personally do not think so). Now let us ask whether, just as one can act immorally by misdirecting someone with statements that are literally truthful, one can act immorally by construing a statement of another in a manner consistent with the words, but inconsistent with the inferences that the context dictates would be reasonable to draw. That is, can I commit a reciprocal moral wrong if I make it your problem that your words can be construed in a manner contrary to your intended communicative desires, and perhaps, contrary to your interests, even when I know what you are trying to say? I think the answer is yes. Going back to our conversation about Hannes’s whereabouts, let us assume that, after seeing Hannes in Washington last week, I have not seen him since. I have no idea where he is now. You, however, would gain some advantage by my acknowledging that I had seen him yesterday, although I hadn’t. You then tell others that I did not deny having seen Hannes yesterday when I said that I had seen him last week, and that I spoke indirectly when you asked about Hannes, casting suspicion on my honesty. You never lie, but you have insincerely taken my words to imply things I never intended them to imply, and that the normal rules of implicature would not support. This makes you an immoral listener, just as the first story made me an immoral speaker. Of course, your immoral act does not end with your having intentionally misconstrued my words. For you have not only misconstrued them, but you have presented your interpretation to others knowing that you did not fairly report the substance of my message, and thus in order to deceive them without presenting them with a bald faced lie. I will not here address the morality of the person who privately perverts the intended meaning of a communication without letting anyone know about it. When the distorted interpretation is reported, though, the interpretive act has been transposed into a deceptive speech act, which is precisely what we saw to be uncontroversially immoral above. That is why the two acts are reciprocal moral wrongs.

12

H. P. Grice, Logic and Conversation, in SYNTAX AND SEMANTICS 41, 45 (P. Cole & J. Morgan eds., 1975).

-340-

Moreover, we would draw the same conclusions about your morality if you were to take advantage of an error I made in what I said. Assume that I met Hannes on Tuesday of last week. We both know this because you were also there. In a subsequent conversation with you, I accidentally refer to having seen Hannes on Wednesday. We both know that by Wednesday, Hannes had flown to London, and that I didn’t see him that day and could not have seen him that day even if I wanted to see him that day. Nonetheless, knowing that I had made a simple speech error,13 you find a way to take advantage of the fact that I said Wednesday instead of Tuesday, perhaps implying that I was scheming with Hannes, or that I am a liar, or something else that you know not to be true. Again, you never lie, but you construe my words both insincerely and ungenerously, and then report your construal to others. Lawyers do this all the time when they engage in aggressive cross-examination of an opposing witness. It is insincere in that context, whether or not it is justified or required by the lawyer’s obligation to advocate zealously. Should statutory interpreters behave similarly? Nothing in the nature of the adversarial system, which is what justifies the cross-examiner, suggests that they should be given similar license. Using statutory language to flout purpose I suggest that judges sometimes behave toward legislatures and toward litigants just as you behaved toward me in our hypothetical stories about Hannes, whether you took advantage of an inference that was available but not a fair interpretation of my words, or whether you took advantage of a mistake. They take advantage of linguistic indeterminacy to interpret language that undermines the communicative intent of the speaker, in this case, the legislature. Sometimes, the indeterminacy results from a presumed chain of inferences, as in our hypothetical. At other times, it results from vagueness or from ambiguity. In still other cases, judges take advantage of errors in drafting. Before I illustrate this point with examples from case law, I wish to make two points. First, these problems do not arise in every case. Typically, cases involve precisely the situation that the statute was enacted to address. Thus, as is often true with controversial statutory cases, we are dealing with situations that occur at the margins. Moreover, even when judges have the opportunity to flout the legislature’s intent or purpose, they usually do

13

In everyday interactions, we routinely compensate for grammatical errors of others and construe the utterances as they were intended to be construed. For recent studies, see Lyn Frazier & Charles Clifton, Jr., Quantifiers Undone: Reversing Predictable Speech Errors in Comprehension, 87 LANGUAGE 158 (2011). Frazier and Clifton note that such compensation for grammatical errors may be more prevalent in informal speech than in the construal of formal writings, a fact obviously relevant to the legal context. See id. at 167-68. This distinction among registers suggests that individuals confronted with obvious errors in formal settings are more likely to be consciously aware of the mistake, and then must decide how to construe the language: as literally written, or as an error. The fact that we so routinely compensate for speech errors in a manner respectful of communicative intent suggests that self-consciously doing the opposite in formal settings flouts social norms and is construed as such.

-341-

not do so because they regard their roles, at least in part, as furthering the will of the legislature.14 This essay, then, is truly about outliers. Second, the judicial practices that I criticize might be defended as the best practices even if I am right that they are laced with immoral linguistic games. That is because sticking closer to the text than to a set of expected inferences has its own systemic advantages, many of which are described in the literature defending textualism.15 One may argue that the risk of an occasional misreading of communicative intent is a small price to pay for the democratically salient principle of legislative primacy. This argument is convincing in some contexts, but not in the ones that form the subject of this essay. I return briefly to this question later. The cases that most clearly illustrate my point are ones in which the statute is susceptible to multiple interpretations, whether because of syntactic ambiguity, lexical ambiguity, or vagueness (i.e., the case involves a borderline case of a statutory word), and a court chooses to ignore the purpose of the legislation and to take advantage of the linguistic opening. Consider Ledbetter v. Goodyear Tire and Rubber Co., decided by the Supreme Court in 2007.16 It is an unfair employment practice under the Civil Rights Act to discriminate against “any individual with respect to his compensation . . . because of such individual’s . . . sex.”17 Ledbetter claimed that she was being paid less because of her sex.18 The statute has a 180-day statute of limitations. Although her claim was filed more than 180 days after the discrimination had allegedly begun, Ledbetter argued that her claim was timely as long as it applied only to those paychecks issued in the past 180 days. 19 In other words, her claim was that each time that Goodyear paid her less than they would have paid her if she were male, she had been subject to discrimination. In a 5-4 decision, the Court rejected this position, accepting instead Goodyear’s argument that the ordinary meaning of “discriminate” would focus on the decision to pay her less, made long ago, and not on the ministerial act of cutting a paycheck.20 Of course, the Court was right about that. And the Court does often employ the canon that statutory words should be construed in their ordinary sense.21 Justice Scalia has explained that the canon is

14

See, for example, Bruesewitz v. Wyeth LLC, 131 S. Ct. 1068 (2010), for a recent case in which the majority opinion, 131 S. Ct. at 1080-81 (Scalia, J.); the concurring opinion, 131 S. Ct. at 1085-86 (Breyer, J., concurring); and the dissenting opinion, 131 S. Ct. at 1086-87 (Sotomayor, J., dissenting), all make specific reference to the intent of the legislature. I discuss the debates about the propriety of referring to legislative intent, and the fact that judges of all political and philosophical stripes refer to such information in LAWRENCE M. SOLAN, THE LANGUAGE OF STATUTES: LAWS AND THEIR INTERPRETATION ch. 3-4 (2010). 15 See, e.g., ANTONIN SCALIA, A MATTER OF INTERPRETATION: FEDERAL COURTS AND THE LAW (1997) (arguing that since statutes are legislatively enacted and intentions are not, that the proper role of courts is to construe only the language of the statutes themselves); John Manning, The Absurdity Doctrine, 116 HARV. L. REV. 2387 (2003) (arguing that with an enhanced theory of word meaning, it becomes less necessary to look outside the statutory language itself, thus constraining judicial decision making). 16 550 U.S. 618 (2007). 17 42 U.S.C. § 2000e-(2)(a)(1) (2006). 18 Ledbetter, 550 U.S. at 621-22. 19 Id. at 624-25. 20 Id. at 628-29. 21 For discussion of the ordinary meaning canon, including linguistic justification, see SOLAN, supra note 14, at 53-81.

-342-

used as a surrogate for investigating intent: “The question, at bottom, is one of statutory intent, and we accordingly begin with the language employed by Congress and the assumption that the ordinary meaning of that language accurately expresses the legislative purpose.”22 Nonetheless, it defies common sense to think that the Congress intended to create a safe harbor for salary discrimination not discovered within six months. As Justice Ginsburg noted in dissent, unlike promotion decisions, compensation decisions are “hidden from sight,” and comparative information becomes known only after time, if at all.23 The decision literally gives an employer leave, after six months, to say to an employee, “I just want you to know that I am paying you less because you are a woman, and I have been doing it for long enough that I can do it forever.” The statutory language licenses the majority decision—it is surely not without basis. But the majority has taken advantage of a linguistic opening to flout the purpose of the statute. I suggest here that this practice is morally wrong in everyday life, and I can see no reason for it to be less so when judges engage in it. In 2009, Congress amended the law to make this interpretation no longer available, requiring that the statute of limitations be reset with the issuance of each paycheck.24 More difficult are cases in which it appears that the legislature erred. Unlike cases involving statutes whose literal meaning is ambiguous, these cases actually do pit the literal meaning of the statute against the goals that the legislature was attempting to accomplish. And not all errors are the same, as Jonathan Siegel points out in his important work on this issue.25 The easier of these cases involve scrivener’s errors in which the legislature seems to have suffered a written slip of the tongue. United States v. Locke26 illustrates this problem. A statute requires that a person claiming mineral rights on federal land file a statement with the Bureau of Land Management “prior to December 31 of each year.”27 Although it is possible that Congress intended to require that claimants file by December 30, the likelihood is that Congress meant to say “by” instead of “prior to.” Thinking that the statute required only that the claim be filed by year end, Locke filed his claim for his mineral mine in Nevada on the last day of the year.28 The Bureau of Land Management held him in default of the filing requirement and proceeded to take away the mineral rights.29 The situation was made worse

22 Morales v. Trans World Airlines, Inc., 504 U.S. 374, 383 (1992) (quoting FMC Corp. v. Holliday, 498 U.S. 52, 57 (1990); Park ‘N Fly, Inc. v. Dollar Park & Fly, Inc., 469 U.S. 189, 194 (1985)) (internal quotation marks omitted). 23 Ledbetter, 550 U.S. at 649 (Ginsburg, J., dissenting). 24 Lilly Ledbetter Fair Pay Act of 2009, Pub. L. 111-2, 123 Stat. 5 (2009). It is often the case that legislative overrides of judicial decisions occur in so-called “plain language” situations where a court argues that it has no choice but to rule in a manner that the legislature no doubt did not intend. See William N. Eskridge, Jr., Overriding Supreme Court Statutory Interpretation Decisions, 101 YALE L.J. 331, 347 (1991). For discussion of congressional overrides in the context of the civil rights laws, see Deborah A. Widiss, Shadow Precedents and the Separation of Powers: Statutory Interpretation of Congressional Overrides, 84 NOTRE DAME L. REV. 511 (2009). 25 Jonathan R. Siegel, What Statutory Drafting Errors Teach Us About Statutory Interpretation, 69 GEO. WASH. L. REV. 309 (2001). 26 471 U.S. 84 (1985). 27 43 U.S.C. § 1744 (2006). 28 Locke, 471 U.S. at 89-90. 29 Id. at 90.

-343-

by the fact that a member of Locke’s family had called the Bureau’s office and was told that the filing had to be made by the end of the year.30 However, the doctrine that the government may not be estopped as a result of its errors made this fact appear to be legally irrelevant.31 Justice Thurgood Marshall’s majority opinion showed little sympathy for the Lockes. On its face, such an opinion, although perhaps justifiable in its respect for the language that the legislature actually used, is morally questionable. The legal system is taking advantage in two different ways of a legitimate misunderstanding that the system itself caused: by enacting a statute that was “a trap for the unwary,” as Judge Posner has noted,32 and then by misinforming a member of the public about what the law said because the government workers themselves had fallen into the trap. I would hold Justice Marshall and those who voted with him morally blameworthy if it were not for a footnote in the opinion that threatened the government with abandonment of the “no estoppel” doctrine when the case was remanded.33 Lawyers for the government read the footnote and gave the mineral rights back to Locke.34 Thus the case successfully applauds language-based rule of law values without allowing one side or the other to take undue advantage of a communicative error. Most cases, however, do not accomplish these goals simultaneously.35 Less nuanced is Judge Bybee’s dissenting opinion in Amalgamated Transit Union Local 1309 v. Laidlaw Transit Service, Inc.36 The Class Action Fairness Act liberalizes removal to federal court of class actions filed in state court, and calls for removal decisions to be appealed as follows: [A] court of appeals may accept an appeal from an order of a district court granting or denying a motion to remand a class action to the State court from which it was removed if application is made to the court of appeals not less than 7 days after entry of the order.37 This is obviously a legislative error. Congress meant to say “not more than 7 days . . . .” What sense does it make to say that an unsuccessful party must wait a week, and then has until the end of time to appeal? Arguing that the language should be applied as written, Judge Bybee

30

Id. at 89 n.7. Id. 32 Richard A. Posner, The Problems of Jurisprudence 267-68 (1990). 33 Locke, 471 U.S. at 89 n.7. 34 I describe this history in more detail in SOLAN, supra note 14, at 109. 35 For one that does not, see Bowles v. Russell, 551 U.S. 205 (2007), in which a judge had misinformed a prisoner of the time that he had to file an appeal, and the Supreme Court ruled that the law should be interpreted literally, since the statute was jurisdictional in nature, id. at 206-07, an interpretation that is legitimate, but not necessary given the statutory language. 36 448 F.3d 1092, 1094 (9th Cir. 2006) (Bybee, J., dissenting). For further discussion of this case, see Jonathan R. Siegel, The Inexorable Radicalizaiton of Textualism, 158 U. PA. L. REV. 117 (2009), and my response to Professor Siegel, Lawrence M. Solan, Response, Opportunistic Textualism, 158 U. PA. L. REV. PENNUMBRA 225, 228-29 (2010), http:///www.pennumbra.com/response/1042010/solan.pdf. 37 28 U.S.C. § 1453(c)(1) (2006) (emphasis added). 31

-344-

adduced textualist rhetoric. Arguing that Congress did not make a scrivener’s error, which a court might have the right to correct, Bybee continued: Section 1453(c)(1) makes perfect sense; it is fully grammatical and can be understood by people of ordinary intelligence. That we think Congress might choose a different word if it decides to redraft the statute hardly means that someone made a “typographical error” that the court may blithely correct. “‘It is beyond [the Court’s] province to rescue Congress from its drafting errors, and to provide for what we might think . . . is the preferred result.’”38 Of course, the fact that the sentence is grammatical has nothing to do with whether the legislature committed an error in drafting. Bybee would have been on stronger moral grounds had he forthrightly admitted that the legislature made an error, but that in his view the legal system should trade some of its moral authority in an individual case for certainty, which brings credibility to the system. Instead, he engaged in fetishizing the assertion notwithstanding the legislature’s obvious communicative intent. Finally, let us look at a more difficult situation, one in which the legislature erred by writing a statute inconsistent with its purpose not because it used the wrong language, but because it based its legislative decision on erroneous facts. Consider United States v. Marshall,39 a case that is well known to law students. It pits Judges Easterbrook and Posner against each other in the Seventh Circuit. Perhaps for that reason, the circuit court opinions are studied more than is the Supreme Court’s affirmance.40 Marshall was convicted of distributing more than 10 grams of “a mixture or substance containing more than a detectable amount of LSD.”41 The LSD, which weighs next to nothing, was sold on blotter paper, which is more than 100 times as heavy as the drug itself.42 Nonetheless, both the Seventh Circuit and the Supreme Court affirmed the convictions. This, as Judge Posner wrote in dissent, resulted in the penalty per dose of LSD to exceed the per-dose penalty for other drugs to an extent that makes little sense.43 The best explanation is that Congress wrote a law in which LSD was treated like powder drugs, such as heroin, because those who wrote the law did not know that the various drugs covered by the statute were neither manufactured nor sold in a comparable way.44 Not all legislative errors are created equal. Judge Easterbrook’s majority opinion is not about taking advantage of linguistic accidence, but rather about taking advantage of

38 39 40 41 42 43 44

Amalgamated, 448 F.3d at 1098 (Bybee, J., dissenting) (internal citations omitted). 908 F.2d 1312 (7th Cir. 1990). Chapman v. United States, 500 U.S. 453 (1991). Marshall, 908 F.2d at 1314-15. Id. at 1315. Id. at 1332-34 (Posner, J., dissenting). See id. at 1333-34.

-345-

legislative ignorance. There is no doubt that Congress intentionally punished LSD as it did. Favoring consideration of purpose in deciding which of two available readings should be accepted does nothing to compromise respect for the constitutionally-mandated legislative process. Correcting linguistic errors in drafting intrudes further into the legislative process, but does no more than attempt to arrive at the legislature’s communicative intent. As noted above, we routinely compensate for grammatical errors in the speech of others, unselfconsciously drawing from the language the intended meaning.45 The correction of legislative errors that concern mistakes of fact resulting in the enactment of senseless laws are more difficult, however. When the mistake is a scrivener’s error, the statutory interpreter who corrects the error says in essence, “You said x, but you meant to say y.” When, in contrast, the mistake is one of basing a law on an erroneous set of facts, the interpreter who corrects the error instead must say the equivalent of, “You intended to accomplish goal g by virtue of enacting law x. But x does not accomplish g. Rather, y does. So I will change x to y to assist you in achieving g. Changing a law to better accomplish the legislature’s goal is more of an intrusion into the legislative process, and reasonable judges and scholars are likely to disagree about its propriety. Perhaps the appropriate solution of this case would have been to declare the law unconstitutional as applied, since the penalty, in the statutory scheme, lacks a rational basis. Statutory interpretation and judicial candor Judges must both decide disputes and explain the reasons for their decisions. Often, it is suggested, judges make decisions based on personal values, their own politics, or perhaps an unarticulated sense of the best decision under the law, which then must be justified post hoc. Gaps between reasonable inferences about what drove a judge to decide a case and the reasons for the decision that the judge articulates create the inference that judges are not being candid. David Shapiro describes the importance of judicial candor: A requirement that judges give reasons for their decisions—grounds of decision that can be debated, attacked, and defended—serves a vital function in constraining the judiciary’s exercise of power. In the absence of an obligation of candor, this constraint would be greatly diluted, since judges who regard themselves as free to distort or misstate the reasons for their actions can avoid the sanctions of criticism and condemnation that honest disclosure of their motivation may entail. In a sense, candor is the sine qua non of all other restraints on abuse of judicial power, for the limitations imposed by constitutions, statutes, and precedents count for little if judges feel free to believe one thing about them and

45

See Frazier & Clifton, supra note 13.

-346-

to say another. Moreover, lack of candor seldom goes undetected for long, and its detection only serves to increase the level of cynicism about the nature of judging and of judges.46 As Judge Posner points out in How Judges Think, the issue of candor arises more with the judges of high courts, the Supreme Court in particular, since those jurists have far more discretion than do lower court judges.47 Moreover, the cases most likely to generate published opinions are the more controversial ones, increasing the likelihood that a judge’s personal values will infiltrate the decision-making process. It is hard to believe it is an accident that the five most conservative justices voted as they did in Ledbetter (the employment discrimination statute of limitations case), and that the four more liberal justices opposed them in dissent. It appears that the justices had reason to vote as they did based on their political and personal values, and used the linguistic opening as an opportunity to further these values. Not all cases involving assertive fetishism involve a lack of candor, just as not all cases that concern candor raise the moral issues discussed here. Nonetheless, the relationship between the two issues creates a particularly troubling set of examples. Judges who simultaneously construe statutes in a way consistent with the language but inconsistent with the statute’s purpose, do so to further their own values, and hide the ball about all of this. As for whether this lack of sincerity is appropriate, I agree with Professor Shapiro that it should be kept to a minimum. A recent article by Micah Schwartzman argues that judicial sincerity is important to a democracy, for people are entitled to know the reasons behind the ways in which the state treats them.48 While the lawyer may be insincere, the judge plays a different role. Yet, as Judge Posner points out, the requirement that judges write only about legitimate legal arguments severely restricts their decision-making options, and preserves rule of law values at least in part.49 The practice essentially tells judges: “Whatever your actual motivations for making a decision, unless you can justify it in legal terms to the legal community, you should not go there.” I surely do not recommend that judges cease this practice. However, when other values are self-evidently driving the decision-making process, this practice most likely comes at some cost in credibility. A model for judicial sincerity in this context is Chief Justice John Marshall’s decision in United States v. Wiltberger.50 A statute that federalized crimes committed on American vessels in the high seas, defined “high seas” to include rivers in other countries for most of the crimes, but failed to do so for the crime of manslaughter, with which Wiltberger was

46

David L. Shapiro, In Defense of Judicial Candor, 100 HARV. L. REV. 731, 737 (1987) (footnotes omitted). RICHARD A. POSNER, HOW JUDGES THINK 269-323 (2008) (chapter entitled “The Supreme Court Is a Political Court”). 48 Micah Schwartzman, Judicial Sincerity, 94 VA. L. REV. 987, 990-91 (2008). 49 See POSNER, supra note 47. A recent article by Mathilde Cohen takes a similar position, arguing that judges need to be candid about the legitimacy of the arguments they make, but not about their belief in them. Mathilde Cohen, Sincerity and Reason-Giving: When May Legal Decision-Makers Lie?, 59 DEPAUL L. REV. 1091, 1098 (2010). 50 18 U.S. (5 Wheat.) 76 (1820). 47

-347-

charged.51 Marshall, in his opinion, admitted that the legislature most likely intended to include this crime, but decided that the rule of lenity, which at the very least prohibits courts from expanding criminal liability beyond any reasonable reading of the statute, was the more important principle to apply.52 Thus, Marshall placed other values above the intent of the legislature. But he did not do so by either ignoring and flouting that intent as an opportunity to impose his own values, nor by pretending that he was unable to discern the legislature’s intent in such an obvious case. Rather, he placed his own hierarchy of values on the table, an act of judicial candor and commitment to avoiding the immorality of disrespecting communicative intent while pretending not to be doing so. Reasonable minds can disagree with Justice Marshall’s ruling, but he cannot be accused of falling prey to assertive fetishism. Conclusion I have argued in this essay that the legal system loses some moral high ground when judges take advantage of linguistic indeterminacy to flout the intent of the legislature to accomplish its goals. How important this is depends in part upon how much legitimacy is lost when the law operates in a manner that is inconsistent with the moral intuitions of citizens.53 I assume that there is at least some loss of legitimacy when judges make arguments that are both transparently insincere and wrongful in their treatment of language. I conclude by addressing some objections that may be made in response to my analysis. First, the legislature’s purpose is not always obvious. When statutes contain compromise, the purpose of a statute is to accomplish legislative goals to the extent that the compromises have allowed these goals to go forward. As Justice Scalia describes it, “the purpose of a statute includes not only what it sets out to change, but also what it resolves to leave alone.”54 To some extent, Scalia is correct. Nonetheless, when the question is which of two readings of a statute licensed by the language is the better one, we are not dealing with a question of unexpressed purpose or intent. In Ledbetter, for example, it is difficult to come up with any reason for Congress to have written a statute consistent with the majority position. My point is not that looking at text should be replaced by looking at purpose; in fact, I cannot think of anyone who takes that position as a general matter. Rather, my point is that when language is either uncertain or clearly mistaken, it is simply wrong to use the uncertainty to interpret a law in a manner that thwarts the communicative intent of the law.

51

Id. at 98. Id. at 99. 53 See PAUL H. ROBINSON & JOHN M. DARLEY, JUSTICE, LIABILITY AND BLAME: COMMUNITY VIEWS AND THE CRIMINAL LAW (1995) for discussion about the loss suffered by a legal system with large gaps between the intuitions of the community and the dictates of the code. 54 W. Va. Univ. Hosps., Inc. v. Casey, 499 U.S. 83, 98 (1991). 52

-348-

Secondly, one reason for reliance on text is that the legislature wrote the text, so sticking to the language creates fewer opportunities for judicial activism. Again, this is true when the decision is between paying attention to the language on the one hand, or unexpressed purpose on the other. But paying attention to both ipso facto reduces judicial discretion. For the intersection of two considerations (in this case language and purpose) can be no greater than either of the considerations individually. To conclude otherwise reflects an error in reasoning that Kahneman and Tversky have called “the conjunction fallacy.” 55 In their most famous example, a feminist named Linda who takes a job as a bank teller cannot be more likely to be a “feminist bank teller” than to be either a feminist or a bank teller.56 Yet people engage in this fallacy as a result of the mental models they form about the intersection of these traits. By the same token, if linguistic indeterminacy leaves two sensible interpretations, requiring that judges examine these interpretations with respect to furthering the legislature’s purpose cannot, as a logical matter, expand the number of possible readings that the judge has the discretion to impose. Thirdly, there is one important difference between my stories about everyday deception on the one hand, and the undermining of legislative purpose on the other: the latter is transparent. When I deceive you about Hannes’s whereabouts, you cannot turn to a record, or to a dissenting opinion to learn the truth of the matter. On the contrary, if I am successful at deceiving you, you will never discover the truth. That is my very goal. Nonetheless, the presence of a public record, which includes opinions that run contrary to the offending one, does not excuse an individual judge from taking responsibility for a morally unjustified position. The fact that a bad act—even a bad interpretive act—is discoverable does not convert it into a good act. But the transparency does provide a partial vindication of the legal system as a whole, and is a positive attribute of the American legal order. Far more serious an objection to my position is the fact that the legal system might be doing the best it can whether or not I am right about the status of using language to flout purpose. If, for example, the kinds of cases that I describe rarely arise, but the value of focusing on linguistic nuance as a general matter enhances rule of law values, then perhaps it is best to tolerate small moral failures at the margins of a just and legitimate system. For example, a great deal of value is placed on requiring that statutory interpretation respect the legislative process that led to the law’s enactment in the first place. I personally do not accept the argument that such linguistic fidelity excuses the legal system from acting in a way that we would regard as immoral in our everyday lives. The loss of legitimacy is not worth any perceived gains, and when it comes to construing statutes whose language is susceptible to only a single interpretation, there is no gain. However, if those who defend the status quo recognize the obligation to defend this balance of insincerity at the margins against the

55 Amos Tversky & Daniel Kahneman, Extensional Versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment, 90 PSYCHOL. REV. 293 (1983). 56 Id. at 299.

-349-

benefits of practices that reinforce formal and systemic values, then progress will have been made.

-350-

Retiring President’s closing address: ethics, best practices, and standards Ronald R. Butters Forensic Linguistic Research & Duke University, Durham, NC, USA [email protected] The topics announced in my title are such that The International Association of Forensic Linguists, like most scholarly and professional organizations, rarely have time to talk about enough. Perhaps it is somewhat imprudent to take up such topics in the last long speech of the conference—after days in which the audience have heard numerous stimulating academic papers, have been exposed to new ideas and professional challenges, and have made new professional friends and strengthened old ones. But I believe that these are important matters, even for the brain-weary and ambitious, so please bear with me while I explore some aspects of the questions of just why we are here in the first place, and what we might do about the promises that we, as an Association, have made to the world. I’d like to begin by looking at our web page—specifically, a series of bullet points that list the ‘Purpose and Aims’ of the International Association of Forensic Linguists that we find there, and that have been there now for many years. These four items are so important that they are listed in the IAFL Constitution (2005, Article 2. ‘PURPOSE’; for convenience in discussion, I have added numbers where the original had bullets): The purpose of the Association is to improve the administration of the legal systems throughout the world by means of a better understanding of the interaction between language and the law. More specifically, the Association aims to promote: 1) The study of the language of the law, including the language of legal documents and the language of the courts, the police, and prisons 2) The alleviation of language-based inequality and disadvantage in the legal system 3) The interchange of ideas and information between the legal and linguistic communities 4) Research into the practice, improvement, and ethics of expert testimony and the presentation of linguistic evidence, as well as legal interpreting and translation 5) Better public understanding of the interaction between language and the law In addition, the IAFL makes additional commitments presented in another section of the web site, ‘Further Aims’ (); while these are not listed in the Constitution itself, and their author(s) are not listed, they take on a strong measure of authority from having been announced for many years as ‘Aims’ of the IAFL: 6) Furthering the interests of linguists engaged in research on the development and practice of forensic linguistics -351-

7) Disseminating knowledge about language analysis, and its forensic applications, among legal and other relevant professionals around the world 8) Drawing up a Code of Practice on matters such as giving evidence in court, writing official reports etc. 9) Collecting a computer corpus of statements, confessions, suicide notes, police language, etc., which could be used in comparative analysis of disputed texts On the whole, we have done rather well in fulfilling many of our stated ‘Purposes’ and ‘Aims’ in the twenty-odd years since our founding. Certainly, we have individually and as a group addressed and continue to address the ongoing projects described in (1), (6), and (7). There has been considerable progress in area (9) corpus collection, as well as (3) and (5). In addition, although the ‘International’ part of ‘IAFL’ still has a long way to go, progress on that score is obvious just from the broad spectrum of participants at this conference alone. And we have made considerable effort towards fulfilling our obligation (2) to ‘alleviate language-based inequality and disadvantage in the legal system’—even if we have by no means succeeded in completely—and I am sure these efforts will continue to be prominent in our undertakings. However, two items that I especially want to draw attention to are (4) and (8), the commitment to ‘research into the practice, improvement, and ethics of expert testimony and the presentation of linguistic evidence’, and ‘drawing up a Code of Practice’. In committing ourselves to these two announced goals of our Association, we have committed ourselves specifically to deepening and understanding the nature of the ethical implications of our testimony and writing, and then explicitly formalizing them. In these areas we still have work to do. That is, we still have to draw up a ‘Code of Practice’, and we still have work to do in the area of ‘research into the practice, improvement, and ethics of expert testimony.’ These are important goals. We owe it to ourselves to pursue them more vigorously than we have to date. I believe firmly that whatever failures we have had in this regard do not arise from the IAFL being an organization of ‘hired guns’ who merely pay cynical lip service to ethical considerations and best practices. Rather, we have been working so hard at the projects of IAFL that we have failed to step back and examine as fully as we might have why we do what we do, and how best to move ahead ethically, objectively, and with greater scientific certainty. There is a certain practical efficacy to turning our attention to these questions, precisely because we are not infrequently accused of the very sin of being cynical hired guns who pay lip service to ethical considerations and best practices; hypothetical courtroom interchange: Cross-examining attorney: And how much are you being paid for your testimony? Forensic linguist: Well, I am not paid for my testimony, I am paid for the work that I do for the court. Cross-examining attorney: Yet none of these organizations you belong to even has a statement of ethics? Not one has even the vaguest position on standards of practice or methodology in the area you have just testified about?

-352-

This is not to say that we never talk about ethical issues or debate standards and practices. In our individual books and articles, ethical issues are sometimes considered (see, for example, Shuy, 2006: 123—29). Our own journal recently published papers from a colloquium on ethics (Ainsworth, 2009; Butters, 2009; Finegan 2009; Nunberg, 2009; Shuy, 2009; Stygall 2009). And during the past four years, the IAFL Executive Committee has attempted to bring a draft of a ‘Code of Practice’ to the membership so as to fulfill the aim we announced years ago. But the issue is generally far from central—for example, the index of the impressive Routledge Handbook of Forensic Linguistics (2010) indicates that the word ethics appears on only two of its 614 pages of text. Happily, I believe we are ready to accomplish this goal in the next two years. Here is how. As a result of work that began with the colloquium papers just mentioned, the Linguistics Society of America (LSA) has amended their general ethics statement so as to set down principles and guidelines for forensic linguistic consulting in America. (Their statement formerly dealt chiefly only with the relationship between linguists and the people from whom they gather data.) I chaired the LSA subcommittee that drafted this proposal, based upon the papers of that same LSA colloquium. The draft was first circulated to the colloquium speakers and then to a number of LSA members who also do forensic linguistic consulting. It was then vetted on the LSA Ethics Committee blog (), where it was posted for comment from the LSA membership (which includes many IAFL members). The LSA Ethics Committee further revised the statement in the light of the comments received from the blog responders, and the Committee then sent it to the LSA Executive Committee for final approval. Perhaps I should note that all but one of the members of the original colloquium panel are also members of the IAFL. The new IAFL president, Maite Turell, has agreed to appoint me to chair the IAFL Committee to attempt to draw up an IAFL Code of Practice far enough in advance of the 2013 meeting that we can very well be ready to adopt one of our own at that time, using the LSA statement as a beginning documents for preparing our own. I have asked Tim Grant to be a member of the committee, and President Turell will shortly appoint additional members. I expect to post a copy of the revised LSA Ethics Statement on the IAFL web site as soon as possible, at which time we expect to set up an online mechanism by which members can post comments about how the LSA Ethics Statement might best be modified in creating our own IAFL ‘Code of Practice’. Are there needs specific to an International Organization that are not addressed by the LSA statement? Are there needs specific to LSA that should not appear in the Code of Practice of an international organizations? (I have also attached a copy of the LSA statement in its most recent form as Appendix A to this paper.) The LSA Ethics Statement is narrower than what the IAFL website bullets seem to indicate is the purpose of an IAFL ‘Code of Practice’. In my own opinion, the LSA Ethics Statement does not go far enough in the way of addressing specific issues related to what the IAFL’s ‘Aims’ asserts should be our goal with respect to ‘Research into the practice, improvement, and ethics of expert testimony and the presentation of linguistic evidence, as well as legal interpreting and translation’. We need to decide just how deeply the IAFL is committed to creating minimal standards that can be applied to those areas of forensic practice that concern us most. In the remainder of this paper, I will focus my attention on one such area of endeavor that is of such great interest to IAFL that it is mentioned specifically in our website presentation of ‘Further Aims’: How far are we pledged to go towards -353-

specific requirements with respect to conclusions concerning ‘comparative analysis of disputed texts’? I have derived some of the data for my discussion from McMenamin (2011), one of a very few recent forensic linguistic reports submitted to an American court that are also actually widely available in the public record; the author is one of the acknowledged leaders in what he calls ‘forensic stylistics’ (McMenamin, 2002), and the theoretical and methodological framework displayed in McMenamin (2011) is characteristic of the work of other forensic linguists (e.g., Leonard, 2010; see also Howald, 2008 who critiques another authorship analysis case, as does Butters, 2011 in rebuttal of Leonard, 2010). While the following questions (1-10) are just about authorship analysis, similar issues also arise in other areas of forensic linguistic endeavor. Authorship analysis is not my own primary concern in forensic linguistics, and I hope here simply to stimulate further questioning from those who engage more centrally in the ‘comparative analysis of disputed texts.’ To this end, I would like to suggest to the IAFL Executive Committee that, at the next meeting of IAFL in 2013, a half-day colloquium be organized that would discuss the issue of standards. As I envision it, such a colloquium would invite scholars from other disciplines—fingerprinting, DNA, voice analysis, to share with us how they approach the question of standards. 1. Should the IAFL take a specific position on the means and necessity of taking into account (when one is comparing two or more sets of texts) the genre differences between the known set (K) and the questioned set (Q)? For example, at least for purposes of testimony, shouldn’t we require that, all other things being equal, instant messages must be compared primarily with instant messages and business letters compared primarily with business letters? 2. Shouldn’t evidentiary standards be much higher for courtroom testimony than for investigatory enterprises? For example, shouldn’t every report written for the police about whether the forensic evidence suggests that a particular ‘person of interest’ should be considered a suspect contain some explicit notice to the effect that ‘this is for investigative purposes only’—and that further investigation would be needed for courtroom testimony? 3. Can we establish guidelines on the minimal number and relative strength of the variables that constitute the ‘idiolect’ or ‘unique set of variables’ needed to come to a conclusion about authorship? Certainly, such numbers are a function of various important factors, for example, (a) the strength or quality of the individual variables and (b) the number of tokens found in each document compared. For example, in his recent ‘expert report’ published online in connection with a high-profile civil case in the United States, McMenamin (2011) offered definitive conclusions based on a list of only nine variables, many of which appear to be as seemingly weak as, for example, noting that the K documents have one-out-of-one instance of the word internet (spelled with a lower-case intital ) and the Q documents have two-out-of-two instances of Internet (spelled with an upper-case initial )’ (see also my Question 10 below). 4. Can we not require that the validity of every putative marker that we introduce into evidence in authorship analysis be attested to in some significant way, such as comparison to a sociolinguistically valid or a statistically meaningful comparison corpus? Minimally, validity could conceivably be the sociolinguistic fact that a single Q-document word represents an obscure dialect or lexical grammatical usage that the K author is likely not to have known. Most cases, however, would be more likely to require a fairly substantial (and of course comparable) comparison corpus (as, for example, the appearance of metathesis-like inverted apostrophes in words such as don’t appearing deviantly as dont’). -354-

5. Could we not begin to establish an official list of variables that have been found, on the basis of solid linguistic research, to be reliable markers of authorship? As it stands, putatively valid ‘expert reports’ are based on ad hoc lists of little more than a handful of spelling and punctuation features about which no research has been done whatsoever, yet about which the authors simply declare that, on the basis of their appearing or not appearing in the Q and K texts, Q and K are or are not authored by the same person? Are we really doing ‘scientific’ and ‘linguistic’ analysis at all when we simply note instances or absences of this or that potentially superficial textual feature? Can we validly claim that a linguist who undertakes such a ‘methodology’ is professionally more qualified to draw scientific conclusions than would be a reasonably experienced high-school English teacher with a set of colorful marking pens and an agenda to find the ‘alikes’ and/or the ‘unalikes’? 6. Should we not require that reports state explicitly what are the principles of selection of data in cases in which the K set is so large that it is not practical to select all possible documents for analysis? 7. Can we not require that reports list a complete and meaningful set of data and sufficient sentential contexts from which the findings are drawn—not just table summaries and/or the bare examples? 8. When reports allude to scholarship, can we not require that the authors cite the specific works upon which the analyst is relying? For example, in his ‘expert report’, McMenamin (2011) writes as follows: Syntax: SINGLE-WORD SENTENCE OPENERS It has been shown that words introducing sentences (sentence openers) group as a habitually-used set for individual writers. The set of sentence openers present in the QUESTIONED writings is wholly distinct from that of the KNOWN-Zuckerberg writings. Questioned: 090203Z Further, / 090203Z Additionally, / 010104Z Thus, / 010604Z Again[,] / 020204Z First[,] / 020204Z Mostly though / 040604Z Paul, Known-Zuckerberg: Okay / And / Anyhow, (2X) / Also, / But / But regardless, / Then / However, The reader is not told where, by whom, and under what conditions ‘it has been shown’ that ‘sentence openers’ are valid markers of authorship. 9. When presenting reports in scholarly publications, ought forensic linguists, insofar as they are legally allowed to do so, make available the complete set of data and take all reasonable steps that will insure replicability? Even in cases wherein the linguist is not allowed to make data public and/or is simply preparing an opinion for the use of police investigators, the forensic linguist ought to make clear in his or her report the location (in the body of data available for examination) and text of specific examples cited, knowing that linguists retained by the contestant at law will need such access in order to prepare a scientifically accurate critique. 10. Can we not implore the forensic linguist who examines spelling and punctuation to take into account the fact that automatic features of word processors often automatically make corrections that the author may not even know were made, or was too hurried to change? For example, as noted above (my Question 3), one of the putatively meaningful variables used by McMenamin (2011) is the variable capitalization of internet ~ Internet. Yet many word-processing software automatically -355-

correct ‘internet’ to ‘Internet’. Does this fact not make authorship claims based on such a feature dubious at best? 11. Shouldn’t a set of standards also suggest criteria for evaluating how meaningful an aggregated small number of variables can truly be in indicating the authorship of Q and K documents? For example, McMenamin (2011) relies upon only three tokens involving the capitalization of internet. Thus a mere three bits of data make up one of the nine total ‘unique set of variables’ upon which his ‘expert report’ relies. Moreover, in the same report, McMenamin also reports as follows: Spelling: CANNOT The word ‘cannot’ appears as two words in the QUESTIONED writing but appears multiple times as a single word in KNOWN-Zuckerberg. Questioned: 020604Z can not [2 words] Known-Zuckerberg: cannot [1 word] (6x) When the seven tokens that make up the can not ~ cannot set are added to the three tokens that make up the internet ~ Internet set, it is apparent that McMenamin’s conclusion about the authorship of the Q and K documents is formed on the basis of 2/9 of his criterial set that in turn are comprised of only ten tokens (ones that, moreover, have not been demonstrated to be other than meaningless as discriminators). And this is not all; another variable is likewise dependent upon a very small number of tokens: Punctuation: SUSPENSION POINTS Suspension points appear in threes and are spaced in QUESTIONED. Three suspension points appear in KNOWN-Zuckerberg but are never spaced between each other or away from words. Questioned: 073003Z . . . I’ve been tweaking the search engine today / 010104Zb I’ll just get this site online as quickly as I can ...’ Known-Zuckerberg: So let me know... (3x) / boxes...there (3x) Thus one-third of the ‘unique set of variables’ upon which McMenamin (2011) bases his firm conclusion that the Q and K authors are very likely different are composed of exactly 18 reported tokens. A similarly small number of features make up the other members of the unique set. It seems reasonable to at least begin to question the scientific validity of a methodology that bases definitive conclusions on so little data— conclusions, moreover, that are intended to be probative in litigation that may involve hundreds of millions of dollars and in criminal proceedings where a death penalty may be at stake. One can argue that, at least in America and perhaps elsewhere, even if testimony is allowed on the basis of a report, the adversarial system of justice allows for rebuttal by other experts who question the methodology and results. Furthermore, the use of ‘expert reports’ in such circumstances at least allows expert testimony on both sides, which might conceivably be more in the interest of justice, since having no forensic linguistic reports at all would leave judges and juries to speculate about the likelihood of authorship on their own. That is, triers of fact left to their own lay knowledge might reach the same invalid and unscientific conclusions that putatively expert linguists might, with no expert testimony to set them right. However, as Solan and Tiersma point out (2005: 167—68), once an expert witness testifies—even if -356-

allowed only to present the questionable data without giving ultimate conclusions as to authorship (as is often the case in the United States)—a good deal of damage may be done, because even merely drawing a jury’s attention to data of dubious value can mislead and confuse them. Moreover, it is an unfair burden on a litigant who has few financial resources to have to hire a linguist to rebut the report of an opponent who has been engaged by a wealthy one. I want to make it clear that I am not indicating a general disbelief in the possibility of authorship analysis. I can point to the work of numerous cautious and conservative practitioners, most of whom are active members of IAFL (e.g., Chaski, 2001, 2005; Coulthard, 2004; Grant, 2007; Howald, 2008; Turell, 2010), and many papers presented at this conference are themselves contributions to exactly the sort of increased knowledge that is needed (Fitzgerald and Schilling, 2011; Kredens, 2011; MacLeod and Grant, 2012; Nini, 2011; Shuy, 2011; Tomblin, 2012; Wright, 2011). Rather, I am suggesting that we at least consider how we can—and that we should— forestall the use of putative authorship analysis that is in fact not up to the high scientific standards of the IAFL that have been represented in general by the papers that have been presented here in the past few days. I am not even going to go so far as to say that the IAFL ‘Code of Practice’ is the proper place for the enunciation of standards and guidelines. But I do think that our statement of ‘Goals, Aims, and Further Aims’ points in the direction of making some kind of summarizing statement on Best Practices with respect to Standards in Authorship Analysis. Appendix A: Linguistic Society of American Proposed Code of Ethics for Linguists in Forensic Linguistics Consulting [revised September 13, 2011] Preamble The following principles of ethical conduct are intended to guide those members of the Linguistic Society of America who engage in forensic linguistic research and legal consulting and testimony; other scholarly and professional associations (for example, the International Association of Forensic Linguists) may have additional ethical codes that members of those organizations should also consult. I. Integrity A. Mindful of their obligations to furnish valid, reliable, and accurate linguistic information and analyses to the justice system, consultants must recognize that their duty is to provide objective scientific evidence that will assist the court in arriving at its conclusions—a duty that overrides any obligation owed to the retaining attorney or litigants who have engaged them. Under no circumstances will consultants knowingly provide linguistic analyses or conclusions that are misleading to an accurate fact-finding process. B. Linguists who are engaged in forensic linguistic consulting will not enter into any arrangements in which compensation is dependent on the outcome of the case. C. In appropriate cases (usually where clients are unable to pay full fee), consulting linguists may provide their services at reduced-fee rates or without charge, not only for the sake of the advancement of science but also as a duty to linguistic science, society, and the judicial system. D. Testimony and reports must be based upon the linguist’s professional knowledge and expertise, and upon meticulous research that uses established and accepted scientific linguistic knowledge and methodology.

-357-

E. Consultants will not add to, delete from, or otherwise alter their reports at the request or suggestion of the retaining attorney or law-enforcement agents if doing so would materially affect the accuracy or reliability of their analyses or conclusions. If material within a report must be deleted because it is legally privileged or inadmissible, consultants should carefully consider whether such deletions materially affect the validity of their analyses or conclusions, and they will inform the retaining attorney or agent if the emended report would be inaccurate or misleading to the fact-finding process. II. Objectivity and Professional Competency A. Consultants undertaking forensic linguistic analyses will state in their reports the methods they have followed and provide all relevant details of the data, equipment, statistical-reliability tests, and computer programs used. B. In making their analyses, consultants will take due account of—and act diligently in accordance with—the technical and professional methods available at the time and their appropriateness to the purposes of the inquiry and the data under examination. C. In reporting on cases where judgments concerning the consultant’s level of certainty of an opinion or conclusion are scientifically possible, consultants may indicate levels of their certainty of opinion or conclusion, expressed either quantitatively (e.g., ‘80% certainty’) or on a discursive scale (e.g., ‘with the highest degree of scientific certainty’). D. Consultants will maintain awareness of the limits of forensic linguistic analysis and of their own knowledge and competencies when agreeing to carry out work, making certain that they possess or can with certainty acquire the specific professional knowledge and skills at the level necessary to ensure that their linguistic analysis is performed at the highest level of competency. III. Confidentiality and Conflict of Interest A. Consultants will not disclose confidential information acquired as a result of consulting relationships or negotiations leading towards the establishment of consulting relationships without proper and specific authority, unless there is a legal obligation to do so. Duties of confidentiality must be maintained even in a social or academic setting—both during the pendency of the case and thereafter—unless confidentiality is waived by the party to whom the duty is owed or the information that the linguist discloses is a matter of public record. B. In any publication or conference presentation that makes use of material or analysis generated by forensic consulting work, or of material that bears directly on the matters at issue in the work undertaken, linguists will reveal the nature of the consulting origins of the generated material. Moreover, when making such use of material generated by forensic linguistic consulting work, the consultant will, whenever feasible, obtain a waiver from any party to whom a duty of confidentiality is owed. If waiver is not feasible or is refused, and to the extent that the regulations of the government and the scholar’s Institutional Review Board allow, the consultant should consider whether anonymizing the factual presentation of the material will adequately protect confidentiality. When confidentiality cannot be protected and is not waived, the consultant must forego academic use of such material. C. To avoid appearances of impropriety, during the pendency of a case consultants should avoid unnecessary contacts with an opposing party’s expert witnesses or attorney outside the formal litigation process.

-358-

D. Consultants should not accept an engagement that would result in the consultant’s conflict of interest with respect to the potentially engaging party and another party on whose behalf the expert is currently a consultant. Before accepting an engagement that could present such a putative conflict, the consultant will reveal the nature of past, current, and proposed consulting engagements to the attorneys representing both parties so that the attorneys may determine whether a conflict of interest exists or potentially could arise. IV. Recommended Practices A. Consultants may wish in their written reports to state explicitly that they are aware of the contents of this Code of Ethics. A statement to this effect might read in whole or in part as follows: This report is based on my professional knowledge and expertise, and on my research using established and accepted scientific linguistic knowledge and methodology. The data and sources that I considered in forming the professional opinions expressed here are referenced where relevant throughout the report. If sworn as a witness, I could testify competently to the matters stated herein. I understand that my duty in providing written reports and giving evidence is to assist the court—and that this duty overrides any obligation to the party by whom I am engaged or the person who has paid or is liable to pay me. I confirm that I have complied and will continue to comply with my duty. My compensation is not contingent in any way on the outcome of this case. B. Consultants should also be mindful that their forensic work may be of utility and scientific importance to other linguists; in this respect, consultants are encouraged to share their results through active participation in the meetings and publications of appropriate professional organizations and related societies. References Ainsworth, J. (2009) A lawyer’s perspective: ethical, technical and practical considerations in the use of linguistic expert witnesses. International Journal of Speech, Language, and the Law 16(2): 279—291. Butters, R. R. (2009) The forensic linguist’s professional credentials. International Journal of Speech, Language, and the Law 16(2): 237—252 Butters, R. R. (2011) Preliminary analysis of Dr. Robert A. Leonard's [1910] report … concerning attempts at determination of authorship in Illinois v. Christopher E. Coleman. [Circuit Court for the 20th Judicial Circuit, Monroe County, Illinois, Case No. 09-CF-50]. Unpublished expert witness report and summary of testimony for the defense in a capital murder trial. Chaski, C. E. (2001) Empirical evaluations of language-based author identification techniques. International Journal of Speech, Language, and the Law 8(1): 1— 65. Chaski, C. E. (2005) Who’s at the keyboard? Authorship attribution in digital Evidence investigations. International Journal of Digital Evidence 4(1): 1—13. -359-

Coulthard, R. M. (2004) Author identification, idiolect, and linguistic uniqueness. Applied Linguistics 25(4): 431—447. Finegan, E. (2009) Expert witnesses and the whole truth. International Journal of Speech, Language, and the Law 16(2): 257—278. Fitzgerald, J. R, and N. Schilling (2011) Uncovering linguistic disguise: forensic linguistic analysis in three 2007 staged suicide attempts. Paper presented at the 10th meeting of the International Association of Forensic Linguists, Birmingham, England, July 2011. Grant, T. (2007) Quantifying evidence in forensic authorship analysis. International Journal of Speech, Language, and the Law 14(1): 1—25. Howald, B. S. (2008) Authorship attribution under the rules of evidence: empirical Approaches—a layperson’s legal system. International Journal of Speech, Language, and the Law 15(2): 219—248. IAFL Constitution. (2005) Kredens, K. (2011) ‘The worst thing is that you killed him’—Pragmatic meanings in forensic contexts. Paper presented at the 10th meeting of the International Association of Forensic Linguists, Birmingham, England, July 2011. Leonard, R. A. (2010) Report on determination of authorship of evidentiary documents in People of Illinois v. Christopher E. Coleman [Circuit Court for the 20th Judicial Circuit, Monroe County, Illinois, Case No. 09-CF-50]. Unpublished police report and summary of testimony for the prosecution in a capital murder trial. MacLeod, N. and T. Grant. (2012) Whose Tweet? Authorship analysis of micro-blogs and other short form messages. In S. Tomblin, N. MacLeod, R. Sousa-Silva & M. Coulthard (eds) Proceedings of the Tenth International Association of Forensic Linguists' Biennial Conference, Aston University, Birmingham [Ebook] www.forensiclinguistics.net McMenamin, G. R. (2002) Forensic Linguistics: Advances in Forensic Stylistics. Boca Raton, Florida: CRC Press. McMenamin, G. R. (2011) Declaration of Gerald R. McMenamin in Support of Defendants’ Motion for Expedited Discovery, Paul D. Ceglia v. Mark Elliott Zuckerberg and Facebook, Inc., United States District Court, Western District of New York, filed June 2, 2011; Exhibit B, p2 (http://www.scribd.com/doc/56976903/W-D-N-Y-1960merged-linguist)

Nini, A. (2011) Style, systems and genre: a theoretical base for stylistic approaches to authorship analysis. Paper presented at the 10th meeting of the International Association of Forensic Linguists, Birmingham, England, July 2011.

-360-

Nunberg, G. (2009) Is it ever OK not to disclose work for hire? International Journal of Speech, Language, and the Law 16(2): 227—236. Routledge Handbook of Forensic Linguistics. (2010) Ed. by M. Coulthard and A. Johnson. New York: Routledge. Shuy, R.W. (2006) Linguistics in the Courtroom: A Practical Guide. New York: Oxford University Press. Shuy R. (2009) Ethical questions in forensic linguistics: introduction to papers from a Linguistic Society of America panel presentation, San Francisco, California, 9 January 2009. International Journal of Speech, Language, and the Law 16(2): 219—226. Shuy, R. (2011) Analyzing large amounts of language in criminal and civil cases. Paper presented at the 10th meeting of the International Association of forensic Linguists, Birmingham, England, July 2011. Solan, L. M., and P. Tiersma. (2005) Speaking of Crime: The Language of Criminal Justice. Chicago: University of Chicago Press. Stygall, G. (2009) Guiding principles: forensic linguistics and codes of ethics in other fields 66d professions. International Journal of Speech, Language, and the Law 16(2): 253—248. Tomblin, S. (2012) Investigating formulaic language as a marker of authorship. In S. Tomblin, N. MacLeod, R. Sousa-Silva & M. Coulthard (eds) Proceedings of the Tenth International Association of Forensic Linguists' Biennial Conference, Aston University, Birmingham [E-book] www.forensiclinguistics.net Turell, M. T. (2010) The use of textual, grammatical and sociolinguistic evidence in forensic text comparison. International Journal of Speech, Language, and the Law 17(2): 211—250. Wright, D. (2011) Stylistic variation within genre conventions in the Enron email corpus: Developing a text-sensitive methodology for authorship research. Paper presented at the 10th meeting of the International Association of Forensic Linguists, Birmingham, England, July 2011.

-361-

Ethics and method in forensic linguistics1 Lawrence M. Solan Brooklyn Law School, USA [email protected] With insight and candor Ronald Butters (this issue) reminds us that forensic linguists, like practitioners in most areas of forensic science, have done more to advance their field substantively, than they have to advance it ethically. The program he suggests is an ambitious one. Butters complains that forensic authorship identification lacks not only a set of agreed understandings about methodology, but lacks also and is in need of standards, so that bogus conclusions based on inadequate data and poor analysis do not pass as valid conclusions based on adequate data and good analysis. In this regard, Butters places methodology beyond mere practice and elevates it to the realm of the ethical: It is simply wrong for a profession to go about its business without some verification that it is doing a good job. And what could be more important than making sure that those academics, whose “day jobs” are to seek the truth, do more good than harm when they enter the courtroom with the express task of presenting analysis that will affect the lives of others in profound ways? I applaud Butters for raising these important issues in such a public and salient context. I raise two points here in response and, to some extent, in elaboration. One explains the ethical problem a bit differently, the other suggests reason for optimism. My first point is that the conflict of interest inherent in expert forensic testimony—especially by those who make their livings, or at least a significant part of their livings as consulting experts—can indeed best be remedied by the development of methods that are demonstrably both diagnostic and replicable. For those who rely upon judgments of co-authorship based on their knowledge of linguistic features and upon a sense that a large cluster of differences or commonalities in a particular case cannot be a matter of accident, research into methodology should be a top priority. Proficiency testing may take the place of the development of replicable methods in the short run, but the best direction for the field is to demonstrate that methods work and are not highly dependent on the skill of the practitioner alone. My second point is that work in computer science and computational linguistics is moving toward answering many of the specific questions that Butters raises about particular standards in the field. Questions about such matters as how much data of a particular kind are needed for valid conclusions to be drawn are commonplace in statistics and modeling, and can easily enough be transported to forensic linguistic application. I end this essay with some brief conjecture about why the field does not appear to have moved ahead quickly with respect to some of these questions, and what it might do to adjust its course. The conflict of interest that inheres in forensic consultation Everyone likes making fun of lawyers. Even lawyers like a good lawyer joke. One thing so galling about lawyers that makes them the butt of these jokes is that they appear to take even the simplest real life situations and convert them into some kind of complicated pageant in which only they can act, and for which they are well paid:

1

My thanks to Lorna Fadden and Roger Shuy for valuable comments on an earlier draft.

-362-

Said a lady to her friend, “When we got our divorce we divided everything we had equally between us. Two children stayed with me, two went to my ex-husband.” “What happened to the property?” asked a friend. “That was shared equally between his lawyer and mine.” Galanter (2005: 65) Those who engage regularly in expert consultation, and especially in expert testimony, have their own version of this conflict. It is sometimes referred to as the “hired gun” syndrome, and it comes from the fact that testifying experts are in a compromised position when it comes to their own methods. Consider the following vignette about expert witness Lucy: Lucy is a professor of computational linguistics and currently has a grant-funded project on authorship identification, which she hopes will have practical application eventually. Last month, a lawyer phoned Lucy, saying he had been told of her work, and asked her if she would be willing to apply it to a legal case, and possibly testify as an expert. Lucy was intrigued. She took the case, analyzed it according to the methods that she had developed, and concluded, by virtue of applying her algorithm, that the questioned document the lawyer presented was very unlikely to have been written by the person to whose known writings she had compared it. In her lab, Lucy was correct 90 per cent of the time when she conducted this kind of analysis this way. She told the lawyer that she would be happy to testify to all of this, as she continues to work in her lab to improve the 90 per cent rate of accurate rejection of authorship. Now compare Lucy to Lacy: Lacy is a forensic linguistic consultant. From time to time she takes authorship attribution cases. Lacy does not conduct her work computationally. Rather, she has a set of 36 stylistic markers by which she analyzes all documents that come to her. She has found from past experience that when the documents are long enough for comparison, some of these 36 markers will tend either to co-occur between a questioned document and a reference set, or be noticeably different between them. There is sometimes controversy about whether her testimony will be permitted, but when she is allowed to testify, her testimony is generally convincing: She is what lawyers call “a good witness.” At first glance, we might prefer Lucy. After all, we know how good her methods are, making it less likely that she is a hired gun. With Lacy, in contrast, we must rely on her persuasive rhetoric, and the intuitive appeal of the data she presents. Here, however, is the problem: We have no idea which expert does a better job. It may well be that Lacy limits herself to the kinds of problems that she is certain to get right, and that her success rate exceeds Lucy’s not unimpressive 90 per cent. Or, it may be that Lacy gets a lot of slack from the intuitive appeal of her analyses, and from her charisma, and that her success rate is far lower than Lucy’s. Psychologists have devoted a great deal of study to the question of expert intuition, in areas as diverse as the thinking of chess masters, medical diagnosis and treatment by physicians, neonatal intensive care nursing, and decision making about firefighting. It is clear that at least in some settings, people are able to form sophisticated mental models of situations about which they are experts, and to weigh relevant factors with great accuracy, notwithstanding that they are unable to describe how they did it.

-363-

(See Kahneman, 2011; Kahneman and Klein, 2009 for debate over the relative usefulness of intuitive versus algorithmic expertise.) Thus, while Lacy may rely upon intuitive judgment more than Lucy, there may well be nothing unscientific about these judgments. I therefore wish to make it clear that I do not use the word “intuitive” in a pejorative sense. Nonetheless, when given a choice, the legal system typically opts for Lucy-like expertise, not because we know that she is more likely to be correct, but because we at least know how likely she is to be right, reducing the probability that cognitive biases or a witness’s compelling personality will play too great a role in the outcome of a case. (See Sanders, 2001. See also Solan and Tiersma, 2005 for discussion of evidentiary standards in the Daubert age). Of course, the expert who offers an algorithmic approach must be sufficiently transparent in what methods were actually used, or the algorithmic expert is presenting no less a black box than the intuitive expert. Over the past two decades, the field of forensic linguistics, I believe, has developed as a field with more Lacys than Lucys, and this has led to some of the problems that Butters observes. Many involved in the field—especially authorship attribution specialists who rely on stylistic markers—conduct little or no laboratory work. This is true both of independent consultants and of academics who self-identify as forensic linguists, sometimes training students to employ in casework the techniques that they currently use, and which are accepted, at least some of the time, by the courts. The result is a dearth of serious research, provoking reasonable questions about the legitimacy of the conclusions reached. In creating these characters, I do not mean to say that one has more integrity than the other: I don’t know them personally well enough to judge, and would have come to the same conclusion if Lucy and Lacy had switched roles. I do mean to say that, as a half century of social psychology has taught us, how we act depends in significant part on the situations in which we find ourselves. So while we should not make the mistake of associating a person who acts normally in a conflicted situation as an individual with bad character (Gilbert and Malone, 1995), neither should we ignore the situation that creates the conflict (see Solan, 2010 for further discussion). Thus, while Butters attributes the lack of attention to ethical and methodological issues to the field’s having chosen other priorities, I see it more as reflecting the reality of the conflict between promoting one’s career and promoting scholarly research. There is a serious tension between two of the aims of the forensic linguistics community that Butters discusses: 4) Research into the practice, improvement, and ethics of expert testimony and the presentation of linguistic evidence, as well as legal interpreting and translation; 6) Furthering the interests of linguists engaged in research on the development and practice of forensic linguistics. Of course, one can define “the interests of linguists …” as including high methodological and ethical standards, but simply saying that this should be the case begs the question. The problem is that promoting the interests of a practice that does not include adequate methodological prowess is at odds with developing best practices that might lead to the abandonment of some of those practices at the expense of the careers of current practitioners. This tension has not always been so pronounced. The history of ‘voiceprint’ analysis provides quite a different story. During the 1960’s, an employee of Bell Labs, which invented the sound spectrograph, began to make extravagant claims about the ability of such devices to distinguish one voice from another, much the way fingerprints were (and still largely are) seen as distinguishable from one person to the next. Police laboratories received training in the use of the new technology, about which there was considerable excitement. Then, in -364-

1979, the National Research Council issued a devastating report, pointing out that there has not been adequate testing to determine how well spectrography can be used to distinguish one voice from the other in forensic settings (NRC, 1979). The report noted: The degree of accuracy, and the corresponding error rates, of aural-visual voice identification vary widely from case to case, depending upon several conditions including the properties of the voices involved, the conditions under which the voice samples were made, the characteristics of the equipment used, the skill of the examiner making the judgments, and the examiner's knowledge about the case. Estimates of error rates now available pertain to only a few of the many combinations of conditions in real-life situations. These estimates do not constitute a generally adequate basis for a judicial or legislative body to use in making judgments concerning the reliability and acceptability of aural-visual voice identification in forensic applications. (NRC, 1979: 60) The leaders in the effort to make sure that linguistic science, if used in law enforcement efforts, would meet the high standards of science itself were chiefly academic linguists, with special efforts by the distinguished phonetician, Peter Ladefoged. (For fuller discussion of this history, see Solan and Tiersma, 2003; 2005). Fast-forward to 2009, when the National Research Council came out with another devastating report, this one looking at forensic identification science in the United States more generally. The report decries the absence of scientific integrity in forensic identification procedures, much as the earlier report did with respect to speaker identification technology used at the time: A body of research is required to establish the limits and measures of performance and to address the impact of sources of variability and potential bias. Such research is sorely needed, but it seems to be lacking in most of the forensic disciplines that rely on subjective assessments of matching characteristics. These disciplines need to develop rigorous protocols to guide these subjective interpretations and pursue equally rigorous research and evaluation programs. (NRC, 2009: S-6) Yet now, it is not the linguistic academic community taking the lead in remedying this situation on behalf of linguists who do not want to see the legal system making excessive claims about the forensic application of the language sciences. Rather, it is chiefly legal academics with expertise in the area of scientific evidence taking the lead, with focus not on linguistics in particular, but on the forensic identification sciences generally. (See, e.g., Koehler, 2011; Sanders, 2007). What has happened between 1979 and 2009? In 1979, there was no field of forensic linguistics, or at least not much of one. Linguists were occasionally called to testify as experts in court, but they did so because their academic expertise serendipitously crossed paths with a legal issue, much the way an academic physicist or engineer might be called upon. Linguists were in the business of being linguists, and the legal system was a novelty for them. Not so now. With undergraduate and graduate programs in forensic linguistics, many academics devote much of their time to applying linguistic knowledge in legal settings, as do consulting linguists without academic affiliations. The gap between the academic community -365-

that once policed its field for abuse by the legal system, and practitioners within the legal system has shrunk considerably. Some in the language and law community have recognized this gap between theory and practice to be a healthy one, and have attempted to maintain it in their description of how linguists engaged in courtroom testimony should view their work. Tiersma (1993: 122) suggests that the field is at its best when the reluctant scholar is asked to share her expertise with the court for the benefit of reaching a proper result in a dispute in which the expert has no interest, either intellectual or otherwise. Shuy (2006: 3) states the forensic linguist’s ideal role more fully: [Forensic linguistics] has become a useful way to refer to the use of linguistics knowledge where there are data that serve as evidence. But I have some concerns about the term itself, because it seems that when one does “forensic linguistics” one is simply doing linguistics, a type of applied linguistics, in fact. These views are attractive. To the extent that they are descriptively accurate views of the field, they explain how it was that the academic phoneticians were the ones who shut down the voiceprint craze of the 1960s and ‘70s. Now, however, expert testimony on questions of language goes beyond ordinary research in linguistics into areas developed by those interested in forensic linguistics as its own discipline (see Butters, 2009). Individuals, sometimes without a great deal of training in linguistics generally (contra Shuy’s ideal description) become skilled in areas of language analysis developed particularly for consumption by the legal community. For reasons stated earlier, many practitioners have little motivation to police their methodologies along the conventional scientific criteria of validity and reliability, apart from disagreement within a particular case in which they have taken opposing positions. The result of this development is that it is not always possible to judge forensic testimony against ordinary practices among linguists, because linguists do not ordinarily engage in the activities that generate the expert testimony. All of this means that the conflict of interest that I have described is institutionally entrenched in at least part of the testifying forensic linguistic community. Does this amount to an ethical issue? It probably does if neither methodological testing nor proven individual proficiency forms a sufficient scientific basis to accept some of what passes for linguistic expertise. Moreover, to the extent that this lack of foundation results from the dearth of research that is itself a product of this conflict, then it is the fruit of a conflict of interest, and is an ethical issue for this reason. But I am reluctant to make too strong a statement in that regard because, as I noted earlier, there is no reason to conclude—one way or the other—that intuitive expertise based on experience and insight fares any better or worse than does algorithmic expertise. Current trends in forensic linguistic authorship attribution Here I will comment briefly on how I think the field appears to be developing to bring a healthy balance between theory and practice in forensic linguistic identification. As I have noted, there seems to be a methodological, and for that matter, cultural division between computational and stylistic analysts, with the former approaching authorship attribution problems in a manner more consistent with conventional views of expert testimony than the latter, but with no evidence that either approach does a better job. Two things are likely to bring about a convergence. First, I expect that if the computational approaches reach some level of maturity and consensus, they will be far more -366-

often accepted by the courts than those approaches based largely on intuition—even excellent intuition based on experience and learning. Computational linguistics is a well-established subfield of linguistic inquiry. Thus, it really is possible to apply ordinary linguistic techniques to a forensic problem, and to ask whether the application in the forensic setting meets the standards of the field itself. (For a recent example, see Rico-Sulayes, 2011. For summaries of the state of the art, see Chaski, forthcoming; Juola, 2006). By the same token, computational linguists and computer scientists are accustomed to testing their algorithms to see how well they work, and reporting the rate of error. One conference, for example, requires the submission of an answer to an authorship attribution task as a criterion for participation.2 These procedures are consistent with contemporary views of acceptable scientific evidence (see Solan, 2010). If judges, at least in the United States, begin to accept authorship identification as a routine matter precisely because it is transparently algorithmic with identifiable rates of error, then such work will become the standard. Second, it will be incumbent upon those whose work is more intuitively stylistic to demonstrate its scientific underpinnings. This can be accomplished by incorporating stylistic features into the computational algorithms being developed by computational linguists and computer scientists. In fact, this is already occurring. It can also happen through the creation of models that demonstrate the efficacy of stylistic analysis as its own approach. This will, no doubt, require far more collaboration between the forensic linguistics community and those adept at statistical modeling. Such collaboration has not been adequate to date. Many of the questions that Butters raises (such as how large an effect must be for it to merit evidentiary weight) are answered automatically in computational settings by virtue of the models employed. Similarly, for those engaged in the promising methods of using linguistic corpora as reference sets in authorship attribution (see Kredens and Coulthard, forthcoming), collaboration with modelers will become a necessity. Whether I am right or wrong in predicting the subsequent history of the field, I firmly believe that far more collaboration among scholars with different areas of expertise is absolutely essential. And I would like nothing more than to see a significant increase in collaboration notwithstanding any conflicts experienced by the players, and without a precipitating crisis. Let’s all hold our breath. References Butters, R.R. (2009) The forensic linguist’s professional credentials. International Journal of Speech, Langauge and the Law 16: 237—252. Butters, R.R. (This volume) Retiring President’s closing address: ethics, best practices, and standards. In S. Tomblin, N. MacLeod, R. Sousa-Silva & M. Coulthard (eds) Proceedings of the Tenth International Association of Forensic Linguists' Biennial Conference, Aston University, Birmingham [E-book] www.forensiclinguistics.net

Chaski, C. (forthcoming) Authorship Identification in the Forensic Setting. In P.M. Tiersma and L. M. Solan (eds) The Oxford Handbook of Language and Law. Oxford: Oxford University Press. Galanter, M. (2005) Lowering the Bar: Lawyer Jokes & Legal Culture. Madison, Wisconsin: University of Wisconsin Press. 2

The PAN Lab held in conjunction with the CLEF conference. See http://pan.webis.de/.

-367-

Gilbert, D.T. and P.S. Malone (1995) The correspondence bias. Psychological Bulletin 117: 21—38. Juola, P. (2006) Authorship attribution. Foundations and Trends in Information Retrieval 1: 233—334. Kahneman, D. (2011) Thinking Fast and Slow. New York: Farrar, Straus and Giroux. Kahneman, D. and G. Klein (2009) Conditions for intuitive expertise: a failure to disagree. American Psychologist 64: 515—526. Koehler, J.J. (2011) If the shoe fits they might acquit: the value of forensic science testimony. Journal of Empirical Legal Studies 8: 21—48. Kredens, K. and M. Coulthard (forthcoming) Corpus linguistics in authorship identification. In P.T. Tiersma and L.M. Solan (eds), The Oxford Handbook of Language and Law. Oxford: Oxford University Press. National Research Council of the National Academies (NRC) (1979) On the Theory and Practice of Voice Identification. Washington, D.C.: The National Academy Press. National Research Council of the National Academies (NRC) (2009). Strengthening Forensic Science in the United States: A Path Forward. Washington, D.C.: The National Academy Press. Rico-Sulayes, A. (2011) Statistical authorship attribution of Mexican drug trafficking online forum posts. Journal of Speech, Language and the Law 18: 53—74. Sanders, J. (2001) Complex litigation at the millennium: Kumho and how we know. Law & Contemporary Problems 64: 373—415. Sanders, J. (2007) Expert witness ethics. Fordham Law Review 76: 1539—1584. Shuy, R. (2006) Linguistics in the Courtroom: A Practical Guide. Oxford: Oxford University Press. Solan, L. (2010) The expert witness meets the adversarial system. In A. Johnson and M. Coulthard (eds), Routledge Handbook of Forensic Linguistics. Oxford: Routledge, 395—410. Solan, L. and P. Tiersma (2003) Hearing voices: speaker identification in court. Hastings Law Journal 54: 373—435. Solan, L.M. and P.M. Tiersma (2005) Speaking of Crime: The Language of Criminal Justice. Chicago: University of Chicago Press. Tiersma, P. (1993) Linguistic issues in the law. Language 69: 113—137.

-368-

Smile Life

When life gives you a hundred reasons to cry, show life that you have a thousand reasons to smile

Get in touch

© Copyright 2015 - 2024 PDFFOX.COM - All rights reserved.