THE APPLICATION OF NONLINEAR PHONOLOGICAL THEORY TO [PDF]

THE APPLICATION OF NONLINEAR PHONOLOGICAL THEORY TO INTERVENTION WITH PHONOLOGICALLY DELAYED TWINS By MARIA VERENA VON BREMEN B.A., The University of British Columbia, 1985 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (School of Audiology and Speech Sciences)

We accept this thesis as conforming to the required standard

THE UNIVERSITY OF BRITISH COLUMBIA September 1990 © Maria Verena von Bremen, 1990

ii

ABSTRACT Despite the fact that speech-language pathologists do not develop the theories underlying the principles and procedures used in the clinic, speech and language clinicians are ethically obliged to apply the "best" possible theory in their practice. Recently phonologists have been developing a theory of nonlinear phonology. Application of this theory to cross-linguistic data and to child language data has shown that what appeared to be idiosyncratic or difficult to account for using previous theoretical formulations can be neatly explained using a nonlinear phonological explanation. The purpose of this study was to investigate the possibility of further extending the application of nonlinear phonological principles to the field of speech-language pathology. Two levels defined by the theory, segmental and prosodic, were investigated in a six-month-long intervention programme with a set of phonologically delayed twins (aged 5;6 at the outset of the project). Each twin was assigned to an experimental condition motivated by one of these levels, or tiers. In each condition, intervention goals were determined by parameters of the theory; the segmental condition contrasted features "higher" versus "lower" in the feature hierarchy, while the prosodic condition contrasted moraic with onset-rime descriptions of syllable/word shape. Using twins as subjects also allowed the twin aspect of language acquisition and speech-language intervention to be explored. Results of the phonological intervention study revealed that nonlinear phonology provides a viable framework for assessing and determining goals for phonological remediation. A comparison of progress in therapy indicated that one twin acquired therapy goals faster than the other. An investigation of the differential progress of the twins allowed conclusions to be drawn regarding social awareness and success in phonological therapy.

iii

TABLE OF CONTENTS

ABSTRACT

ii

LIST OF TABLES

v

LIST OF FIGURES

vi

ACKNOWLEDGEMENTS

viii

CHAPTER 1. INTRODUCTION

1

CHAPTER 2. THE THEORY

6

Autosegmental Phonology

6

The Feature Geometry

11

Markedness

15

Underspecification

16

Prosodic Phonology

19

Syllables and Segments Combined

25

The Acquisition Process

27

CHAPTER 3. METHOD

29

Initial Assessment

29

Segmental Condition

31

Syllabic Condition

32

Therapy Schedule

33

Therapy Procedure

36

CHAPTER 4. THE SEGMENTAL CONDITION The Starting Point Determining Targets for Remediation

38 '.

39 45

iv Results

48

Conclusions

53

CHAPTER 5. THE SYLLABIC CONDITION

56

The Starting Point

58

Determining Targets for Remediation

60

Treatment ,

63

Results

65

Conclusions

78

CHAPTER 6. TWINS AS RESEARCH SUBJECTS

81

Twins: Handicapped or Not?

81

Heredity versus Environment

83

Twin versus Co-Twin

84

Twins and Psychology

88

Speech and Language Therapy with Twins

90

Autonomous Language

92

Summary and Conclusions

93

CHAPTER 7. CONCLUSIONS

96

Comparison of Nonlinear Phonology and Phonological Process Theory for Intervention

98

Limitations of the Study

100

REFERENCES

102

APPEND DC 1: MINIMAL SPECIFICATIONS OF ENGLISH CONSONANTS

108

APPENDIX 2: GRAND PROBE WORD LIST

112

APPENDIX 3: GRAND PROBE DATA FOR BOTH SUBJECTS

115

APPENDIX 4: THERAPY SESSION OBJECTIVES FOR BOTH SUBJECTS

136

V

LIST OF TABLES Table 1: Percentage accuracy achieved by Subject 2 for clusters targeted in therapy based on the moraic theory of syllable organization

65

Table 2: Percentage accuracy achieved by Subject 2 for clusters targeted in therapy based on onset-rime theory

65

Table 3: Percentage accuracy figures for the initial cycle of targeting /zf and /!/

69

vi

LIST OF FIGURES Figure 1: An example of feature columns for the word toe. based on Chomsky & Halle (1968)

7

Figure 2: SPE characterization of a nasal assimilation rule

9

Figure 3: Nonlinear representation of the Place Assimilation Rule

10

Figure 4: Example of the spreading of the PLACE of /k/ to the preceding nasal consonant

10

Figure 5: Example of a possible rule according to linear phonology which is impossible under nonlinear phonology (as described in McCarthy, 1988)

10

Figure 6: Adaptation of Sagey's (1986) feature geometry incorporating McCarthy's (1988) refinements and Clements' (1985) schematic representation

12

Figure 7: The hierarchical arrangement of the syllable as proposed by onset-rime theory (based on Pike & Pike, 1947)

20

Figure 8: The internal constituent structure of the syllable as proposed by moraic theory

23

Figure 9: Moraic representation of the syllable hierarchy for the word pin

24

Figure 10: Summary of the design of one therapy block

34

Figure 11: Word-initial phonetic inventory of both subjects of this study Figure 12: The syllable-initial-within-word phonetic inventory for both subjects of this study

40 40

Figure 13: The word-final phonetic inventory for both subjects of this study

40

Figure 14: The established feature hierarchy at the beginning of intervention for both subjects

42

vii Figure 15: The twins' realizations of word-final /f/ (left) and word-final /0/ (right) with unestablished nodes marked in parentheses

44

Figure 16: The nonlinear specification of the feature changes required to realize /r/ as [w]

44

Figure 17: Graph of the segmental targets based on mini probe results for Subject 1

50

Figure 18: Graph of the segmental targets based on mini probe results for Subject 2

51

Figure 19: Graph of the syllabic targets based on mini probe results for Subject 2

70

Figure 20: Graph of the /s/-cluster data based on grand probe results for Subject 2

71

Figure 21: Graph of the /Cr/-cluster data based on grand probe results for Subject 2

73

Figure 22: Graph of the /Cl/-cluster data based on grand probe results for Subject 2

74

Figure 23: Graph of the syllabic targets based on mini probe results for Subject 1

77

ACKNOWLEDGEMENTS

Many people have been involved with different stages of this project. I would like to thank the twins and their parents for taking part in this study. The Vancouver School Board was generous in making space and the subjects available to me. Many thanks are due to Meredith Land, V S B Speech-Language Pathologist for her support and suggestions during the therapy portion of the study. Carolyn Johnson deserves more thanks than can be expressed here for patiently reading and rereading drafts of this manuscript. Your encouragement throughout this study and my masters programme were greatly appreciated. I would also like to thank John Gilbert for volunteering to be a member of my examining committee and for first introducing me to the field of speech-language pathology. Many, many thanks go to my mentor, Barbara Bernhardt, for not only guiding me throughout this project, but also for challenging and inspiring me at every stage. I would like to thank my parents for their constant support throughout my university years. And most of all, I would like to thank my husband, Jonathan Oliphant. My appreciation of your understanding and support, especially during the final stages of writing this paper, cannot be adequately expressed here.

1

CHAPTER 1 INTRODUCTION

The assessment and remediation of communication disorders requires that the speech-language pathologist be able to draw on and integrate the facts and theories of several academic disciplines. Physiology, psychology and linguistics name only a few examples. It is impossible to make statements regarding the nature of a communication disorder, nor how it should best be treated, without referring to at least basic theoretical assumptions about language, its components, and how humans make use of it for communication. The assessment and treatment of phonological disorders presents no exception; clinicians rely on theories developed in the field of linguistics to guide them in sampling, analyzing and describing the phonological disability of their clients. Currently, one of the most widely used phonological assessment procedures evaluates a child's utterances for the number of phonemic contrasts which have, and have not, been established in their phonological system. The usual result of this evaluation is a description of the child's speech in terms of phonological processes. Such processes were first proposed by Stampe (1969), a phonologist interested in relationships between phonological theory, universals, historical change, and child language acquisition. Stampe defines processes as mental operations which "merge a potential phonological opposition into that member of the opposition which least tries the restrictions of the human speech capacity" (Stampe, 1969:VII). According to this view, processes are considered to be innate and, without any conscious input from the child, just "happen" to simplify the adult form to one that the child's output mechanism

2

can manage. The role of phonological acquisition then, is to learn to suppress, limit or reorder the application of these processes so that the child's output matches the sound system of her native language. Of note here is the implicit assumption that the child's underlying representation matches the adult form of a word. A more recent view of phonological processes does not agree with Stampe's initial assumptions. Specifically, processes are no longer considered to be entities which apply to the utterances of a passive child. Instead, the child is viewed as actively, though still not consciously, using them to simplify an underlying representation in some relatively consistent way (Ingram, 1976). In addition, our knowledge of the child's perceptual system is incomplete and therefore does not allow us to assume that the underlying representation is adultlike as is implied by Stampe's interpretation. The rethinking of the theoretical assumptions of phonological processes, as demonstrated by Ingram (1976), has attempted to align the language learner's task in phonological acquisition with the more general task of language acquisition, in which it is thought the child actively develops and tests hypotheses regarding the patterns of his linguistic environment. The problem with such a reanalysis is that the theory remains incomplete. The difficulty is that processes, at least in their present form, do not have any psychological reality because we do not know the nature of the underlying representations on which they are acting. The process "intervocalic voicing," for example, can only be accepted as a descriptive statement rather than an explanation of the surface representation, since we cannot know if constraints are affecting the adult input or the child's output. A further difficulty presents itself when the speech-language pathologist uses a phonological process analysis in the clinic. There is nothing inherent in the theory underlying the analysis, nor in the processes themselves, that guides the speechlanguage pathologist in making decisions regarding which observed processes should

3

be targeted first in therapy.

1

Several researchers have suggested that those

processes which have the most detrimental effect on the child's intelligibility should be a priority for remediation (Grunwell, 1985; Leonard, 1985; Leinonen-Davies, 1988), while others recommend that processes which disappear early in the acquisition pattern of normal children should be eliminated first (Ingram, 1976; Grunwell, 1985). Suggestions such as these provide the clinician with only general guidelines for making decisions regarding treatment goals. It is often difficult to decide just which processes are having the most detrimental effect on a child's communication. In addition, decisions based on the patterns of normal children do not necessarily target those segments whose acquisition will lead to the greatest communicative success. Consequently, treatment decisions are ultimately left to the preferences of the individual clinician rather than being motivated by theoretical principles of the phonological process analysis. Thus far, only the assumptions regarding the role of processes in phonological acquisition have been considered. Inherent in any phonological assessment are theoretical assumptions defining what constitutes a sound segment in the first place and what its representation should look like. In this sense, processes are consistent with the view first developed in the Sound Pattern of English (Chomsky & Halle, 1968), which holds that sounds are represented by bundles of features. These feature bundles were developed to account for the sound changes and patterns observed in the languages of the world. Recent work in phonological theory, however, has shifted from a language rule-oriented approach to one more concerned with providing an account of phonological representations. These representations are considered to be "hypotheses about the structure of linguistic knowledge in the human brain" and, as

According to Stampe's (1969) view phonological processes are unordered in their application to an underlying representation. 1

4

such, are intended to explain how the mind represents the sound change rather than just to describe it (Sagey, 1986:10). Instead of organizing linguistic utterances into sequences of feature bundles, each representing a sound, current theory proposes that the phonological components making up a word are arranged hierarchically. The advantage of such an arrangement is that nonadjacent segments or features of segments, whose interaction was formerly accounted for by postulating relatively ad hoc transformation rules, are now considered to be adjacent because they fall on adjacent limbs or nodes of a hierarchy. As a consequence, a language rule, or the mind's representation of that rule, is considerably simplified, since the effect of one segment need only transfer to the neighbouring segment. The "nonlinear" view of phonology proposes that features are arranged into hierarchies to make up a segment, and that segments in turn are organized into hierarchies to create syllables. There is no question that speech-language pathologists should adopt the best phonological theory available and apply it to their work. In this way, the best theory— that is, the one that must come closest to being right—is the theory which accounts for the greatest amount of language data in the most constrained way.2 Where synchronic data (rather than historical change) is accounted for, the best theory should also represent the workings of the human mind in arriving at the surface output. This being the case, drawing on the best theory in order to determine goals and therapy activities in the clinical setting should lead to the most rapid improvement of speech and language because the procedures utilized present the material to be acquired in a manner which mirrors the organization of language within the brain.

The arguments supporting this "best" theory, however, come from linguistics and not from speech-language pathology. Although clinical data does not provide the primary motivation for the theory, it may provide independent evidence supporting it. 2

5

The purpose of the research study described in the remainder of this paper, then, was to evaluate the possibility and the advantages of applying a nonlinear theory of phonology to the assessment and remediation of phonological disorders. The second purpose focussed on the possibility of clinical data providing independent support for this theory. This study was closely modelled after one by Bernhardt (1990), who first evaluated nonlinear phonology from the clinical point of view. A set of identical male twins with a moderately-severe phonological disorder served as the subjects for the current study. It is hoped that their data will add to Bernhardt's data and results, as well as to the limited body of phonological twin data. The following chapter of this thesis presents the development and details of nonlinear phonology. Chapter 3 outlines the methodology of the seven-month long intervention programme. Chapters 4 and 5 then discuss the results of treatment from a segmental and syllabic point of view. Chapter 6 compares the treatment results of the two subjects and outlines possible explanations for the findings. Finally, chapter 7 presents my conclusions regarding the use of nonlinear phonology in the clinic, as well as possible limitations of the study.

6

CHAPTER 2 THE THEORY

Current phonological theories find their starting point in the Sound Pattern of English (Chomsky & Halle, 1968), henceforth SPE. Assumptions regarding the ordering of rules, and the distinction between underlying and surface representations, for example, have been maintained in present-day theories. Changes between current nonlinear theory and SPE lie mainly in how the theory represents the relationship between phonological components. SPE presented a linear view of phonology where segments, composed of bundles of features, could be lined up one next to the other. Nonlinear phonology presents a more constrained modular theory made up of distinct but interacting subcomponents that account for stress, syllabification, and segmental phonology (McCarthy, 1988). This current modular theory combines the research findings of two independent theories which were being developed at about the same time: autosegmental phonology (Leben, 1971, 1973; Goldsmith, 1976; Clements, 1976; McCarthy, 1979) introduced the notion of multilinear representations, while metrical phonology (Liberman, 1975; Liberman & Prince, 1977; Selkirk, 1980a; Hayes, 1981) developed the idea of hierarchical representations.

Autosegmental Phonology Chomsky & Halle (1968), along with their predecessors, postulated that each sound segment could be decomposed into a feature matrix or column which essentially listed each feature with the plus or minus value defining that segment (see figure 1). The difficulty with this view is that there is no internal organization to these columns.

7 As a result, no distinction can be made between features that are distinctive—that is, those features that are necessary to distinguish that segment from the other phonemes in the language—and those that are redundant. In the feature specification for IxJ in figure 1, the features [high], [back], and [low], traditionally needed to characterize the vowels, are unnecessary because of the specification of [+consonantal] as well as [+coronal]. A second difficulty presented by the SPE model is that feature matrices are unable to capture traditionally recognized groupings of features. Because segments are represented by a feature column that has no internal organization, the co-occurrence of features must be considered as coincidental or arbitrary. There is nothing inherent in the feature specification of hi, for example, that captures the generalization that all [+back] vowels in English are necessarily [+round]. As exemplified in figure 1, Chomsky & Halle (1968) claimed that any utterance could be divided into discrete slices of features where each slice represented one segment. Researchers who were attempting to extend the claims of SPE beyond English found that this absolute slicing hypothesis could not adequately account for N -son +cons -syll +cor +ant -high -low -back -cont -round -nas -lat etc.

hi +son -cons +syll -cor -ant -high -low +back +cont +round -nas -lat etc.

Figure 1: An example of feature columns for the word toe based on Chomsky & Halle (1968).

8

suprasegmental phenomena such as tone. Specifically, Leben (1973) observed that in some instances a segment is deleted, but its tone remains behind and attaches to an adjacent segment. In addition, he found that some languages have morphemes that have no segmental value and are represented only by tone. A combination of a high tone followed by a low one, for example, could change an adjective to an adverb in some hypothetical language. Leben (1973) also noted that some languages demonstrate only a small inventory of tone combinations which reoccur on words of different syllable length. Finally, he observed that single segments, usually vowels, are able to carry contour tones (which are commonly analyzed as combinations of level tones). Observations such as these indicated that tones operate independently of, and are more stable than, segments. The discovery that tones could operate independently of segments led to the hypothesis that these components of phonology occupy independent levels of the representation. Nonlinear phonology commonly refers to these levels as "tiers" and accounts for the interaction of tiers using association lines. Leben's observations indicated that the association of the "tonal tier" to the "segmental tier" is not necessarily one-to-one. Mapping relationships can be many-to-one, as with contour tones where more than one tone maps onto a single segment, or the relationship may be one-to-many, in which a single tone maps onto several segments. The idea that tone needed to be represented independently because its domain did not match that of the segment led researchers to consider the possibility that other features may also function independently of the segmental tier. Clements (1976b), for example, postulating that all features are represented on separate tiers, extended the hypotheses of autosegmental theory to vowel harmony. According to his view, it is unnecessary to specify each vowel of a word for the harmonic features. Instead, only one specification is needed which spreads to the vowel slots over the domain of the entire word. Following Clements' (1976) lead, several other researchers found that

9

they were able to account for phonological phenomena in a simple and constrained way if features were extracted out of the matrix and allowed to function independently of the segment. The term "autosegment" stems from the idea, then, that features and tones operate independently of the segment in a phonological rule, and in this sense must be considered to be segments on their own. The conclusion that a feature column or matrix representation of the segment was inadequate led researchers to query just what sort of phonological representation would be adequate. It was noted that hierarchical organizations are common in linguistics. In syntax, for example, hierarchical phrase structure trees allow groups of words to function as a unit in syntactic transformations. In a similar fashion, various features can be naturally divided into groups and a hierarchical arrangement may also be applied. It should be noted that the evidence for dividing features into groups comes not from any articulatory or acoustic similarity, but from the common functioning of features in rule systems (McCarthy, 1988). Many languages, for example, have a rule which assimilates the place of articulation of a nasal to that of the following obstruent (as in English bank [baerjk], bent [bent], and bump [bAmp]). Segmental phonology (following the SPE model) would represent such a rule as in figure 2.

[+nas] ->

a cor

P ant Y back

/_

a cor [3 ant

Y back

Figure 2: SPE characterization of a nasal assimilation rule As illustrated, the linear framework must call on three features, [coronal], [anterior] and [back], as well as use variables, to indicate that the nasal simply matches the obstruent's place of articulation. A hierarchical arrangement of features, however, has the [coronal], [anterior] and [back] features dominated by a PLACE node. The

10

Nasal

Obstruent

i

0 PLACE node Figure 3: Nonlinear representation of the Place Assimilation Rule nonlinear representation of the nasal assimilation rule then, simply requires the PLACE node of the obstruent to "spread" to the PLACE node of the preceding nasal as in figure 3. Figure 4 illustrates how the Place Assimilation Rule would apply in a word such as bank. The fact that nonlinear phonology uses slightly different features from those of Chomsky and Halle is unimportant here. What should be noted is that the nonlinear rule refers to only one piece of information, the place of the obstruent, while the linear rule must call on three features.

Figure 4: Example of the spreading of the PLACE of /k/ to the preceding nasal consonant.

[+nasal]

a back (3 cor y ant

/ _

a cor (3 ant y back

Figure 5: Example of a possible rule according to linear phonology which is impossible under nonlinear phonology (as described in McCarthy, 1988).

A second advantage presented by the nonlinear representation is that the features which refer to place of articulation are all grouped under a place node, making the spread of that node a logical occurrence. As previously stated, a linear framework

11

cannot account for such relationships or groupings of features. Thus a rule such as the one presented in figure 5 is just as possible as that of figure 2, but is impossible according to nonlinear phonology. Phonologists have traditionally recognized four groups of features: a) the major class features [sonorant], [consonantal], [syllabic] b) the place features [coronal], [anterior] c) the tongue-body features [high], [low], [back] d) the manner features [continuant], [nasal], [lateral] (SPE, 1968)

3

Several researchers have designed feature hierarchies which capture these four groupings of features (Clements, 1985, 1989; Sagey, 1986; McCarthy, 1988). The differences among the various proposals stem from the respective authors' hypotheses about how the features function in language rules. It is generally agreed, for example, that a MANNER node never operates in a language rule—that is, the features [continuant], [nasal] and [lateral] never function as a class of segments—and therefore, is unnecessary in the hierarchy.

The Feature Geometry The particular feature hierarchy adopted in this paper was initially proposed by Sagey (1986), and subsequently refined by McCarthy (1988) (see figure 6). The feature hierarchy depicted in figure 6 can be divided into class nodes and terminal or

SPE treats the listed features as maximally binary. See discussion under Underspecification for a description of the nonlinear "privative" and "binary" treatment of features. 3

12 SKELETAL TIER

[anterior]

[distributed]

Figure 6: Adaptation of Sagey's (1986) feature geometry incorporating McCarthy's (1988) refinements and Clements'(1985) schematic representation.

feature nodes. The nonterminal class nodes express claims about how features group together in phonological rules. The highest of these class nodes is the ROOT node, which is the structural representation of the segment itself. "Delinking" a ROOT node implies that that segment is deleted. The features or terminal nodes [consonantal], [sonorant], [continuant] and [nasal] attach directly to the ROOT node. As previously noted, phonologists agree that postulating a MANNER node to dominate [nasal] and [continuant] is artificial because these features do not function as a class of "manner" features in any attested language rule. It is hypothesized that the best representation attaches these features directly to the ROOT node because they group together with all of the place features to create segments and must, therefore, appear higher than the PLACE node..

13

McCarthy (1988) points out that the major class features, [consonantal] and [sonorant], function differently from [nasal] and [continuant] in that they delete, assimilate or dissimilate only when the entire segment is affected or, stated differently, when the rule makes reference to the ROOT node. As a consequence of this observation, McCarthy (1988) argues that [consonantal] and [sonorant] should literally form the ROOT node so that they can be distinguished from those features which operate in phonological rules. I would agree with McCarthy (1988) that the major class features function differently than the manner features but have chosen to depict them as sister nodes (following Sagey, 1986) for ease of representation. The two class nodes dominated by the ROOT tier are the LARYNGEAL node and the PLACE node. The LARYNGEAL node dominates the features that characterize the states of the glottis. Certain researchers (McCarthy, 1988; Clements, 1985; Sagey, 1986) propose the terminal nodes [slack folds] and [stiff folds] instead of [voice], as shown here, in order to account for breathy and laryngealized segments, which occur in some languages. The feature [voice], however, is the only terminal node required under LARYNGEAL to account for English data. The node [+spread glottis] is not necessary because voiceless segments are unspecified for laryngeal status (see Underspecification in this chapter for further explanation). Glottal stops do not function phonemically in English, meaning that [+constricted glottis] is not required in the representation. The class node PLACE does not directly dominate feature nodes, but instead dominates the nonterminal nodes 'labial', 'coronal' and 'dorsal'. It should be noted that researchers disagree as to how the nodes dominated by PLACE should be defined. Some (Chomsky & Halle, 1968; Clements, 1985, 1989; Archangeli & Pulleyblank, 1986) argue that the nodes under PLACE should be defined by their place of articulation, meaning that PLACE would be expressed by combinations of the features [coronal] and [anterior]. Sagey (1986) and McCarthy (1985) define the nodes under

14

PLACE according to the active articulator producing the constriction in the vocal tract. Segments produced using the lips are characterized as labial', those produced with the tongue blade or tip as 'coronal', and those made by the tongue body are termed 'dorsal'. Both points of view agree that 'coronal' needs to be further specified in order to account for the various segments produced with the front of the tongue; Sagey (1986) and McCarthy (1988) propose the features [distributed] and [anterior].

4

[Distributed] separates the "crown" of the tongue into two parts: the blade is characterized by [+distributed] while the tip is [-distributed]. Similarly, the feature [anterior] provides a finer definition of coronal according to where the constriction is made on the passive articulator: palato-alveolar segments are [-anterior] and dentals are [+anterior]. It should be noted here, that there is considerable controversy regarding the characterization of English affricate segments which are [-anterior] as well as a combination of [-continuous] and [+continuous]. Bernhardt (1990), proposes the mnemonic 'complex' in order to characterize the branching structure of the [continuant] feature. Bernhardt's (1990) formulation will be followed in this paper. The features dominated by the dorsal and labial nodes are primarily needed to characterize the vowels in English. Clements (1989) argues that vowels and consonants should have different feature specifications. He proposes that the PLACE node should bifurcate into CONSONANT-PLACE and VOWEL-PLACE nodes. Although I would agree that the features [high], [low], [back] and [round] primarily refer to vowel specifications, the interaction of consonantal and vowel features, especially apparent during the acquisition process, supports the representation

Keating (1990) defines several more 'coronal' places of articulation which are required to describe cross-linguistic data. The features [distributed] and [anterior], however, are the only 'coronal' features required to describe English data. 4

15

provided by Sagey (1986). Using different features to specify vowels from those used to specify consonants, as Clements (1989) suggests, would not allow for this interaction. Following this definition of the feature hierarchy, it should be obvious that the nonlinear view has not adopted the strictly binary view of features postulated in the SPE framework. Although the terminal feature nodes may carry either a plus or minus value ([-anterior] for example, is needed to account for the alveo-palatals in English), the higher nodes are held to be privative in the sense that they are either present or not present in a representation; [-coronal], for example, is not a possibility according to the view of nonlinear phonology adopted in this paper.

Markedness Inherent in the feature geometry described above is the notion of markedness first posited by the Prague School phonologists. Hyman (1975) outlines four currently accepted definitions of the term "marked." One view refers to the complexity of segments, denoting those sounds which add features to their specification as marked. In this sense, Pol is considered more marked than /p/ because it carries the additional feature of [+voice]. The second possible definition considers unmarked segments to be those which occur frequently cross-linguistically. According to this view, the vowels IM and /u/ are unmarked, while /u/ and /tq/ are the marked members of the high front/high back opposition because they are less commonly attested in phonological inventories. A third definition incorporates the Prague School notion of neutrality, which stipulates that the unmarked member of an opposition is the one that surfaces phonetically when two phonemes neutralize in a given position (Trubetskoy, 1939). Because both voiced and voiceless obstruents surface as voiceless in word-final position in German, for example, the voiceless member is considered unmarked. The final view of markedness, outlined by Hyman (1975), stipulates the unmarked member

16

as being the "productive" or "regular" member of an opposition (p. 146). Hyman points out that the regular stress pattern for disyllabic nouns in English is for stress to fall on the first syllable. A noun such as cement, which carries stress on the second syllable, is therefore denoted as marked according to this view. The nonlinear view of markedness embodies all four definitions outlined above. Originally, markedness was a language-specific notion whose definition was based on the phonological rules operating within that language. Current phonological theories, however, place heavy emphasis on cross-linguistic phenomena, stipulating that universally attested phonemes are unmarked and form part of an innate Universal Grammar. According to this view unmarked sounds are acquired earlier than marked ones because the marked values represent more unusual options which require positive evidence from the linguistic environment. The feature hierarchy incorporates the notion of markedness in two ways. First, unmarked segments are those whose salient or most characteristic feature is at a higher level in the hierarchy. Thus, /s/ is less marked than III because it does not require the more deeply embedded feature [-anterior] as part of its specification. Second, the feature hierarchy encodes markedness in terms of complexity of the feature specification. In this view, /s/ is considered more marked than III because it requires the added feature tier [+continuant]. As pointed out by Bernhardt (1990), the concept of markedness is inherently hierarchical because one feature value is usually viewed as being more likely than the opposite value.

Underspecification As previously outlined, markedness values distinguish universal properties from language-specific properties. Since the most desirable representation is also the simplest one, it has been postulated that it is not necessary to include any universal properties in a phonological representation because these properties are predictable

17

from Universal Grammar. Researchers hypothesize that universal 'redundancy rules' fill in missing features at the surface level of representation (Archangeli, 1984). Archangeli (1984) points out that it is desirable to further simplify the model by removing all predictable distinctive features from the phonological representation which then results in "underspecified" segments.

5

Currently there are two accepted models of underspecification. 'Contrastive underspecification' (Clements, 1987; Christdas, 1988) requires that only those features required to distinguish a phoneme in a given context need to be specified in underlying representations. Acquisition of this system requires that the child discover the features which function contrastively in the language of her environment. The second view, 'radical underspecification' (Kiparsky, 1982; Archangeli, 1984; Pulleyblank, 1986b), includes only unpredictable values for features in the underlying representation. In this view, not only the universally predicted features are extracted from the feature specification, but also those features generated by language-specific rules. When acquiring a radically underspecified phonological system, the child's first attempts at a sound will agree with the feature specification provided by Universal Grammar. If a language utilizes a more marked option in predictable contexts, then its feature specification will be filled in by a language-specific redundancy rule. If the more marked feature value is used contrastively in the language it becomes part of the underlying representation.

The initial proposals of SPE did not include the notion of underspecification. However, Chomsky and Halle introduce a form of underspecification in the final chapter of SPE. They propose that universal markedness rules, or redundancy rules, apply to the underlying representation of an utterance before any language specific phonological rules apply. According to this formulation, then, language specific phonological rules act on a fully specified underlying representation. The current view of underspecification holds that redundancy rules apply as late as possible in the derivation.

5

18

Radical underspecification is the view which has been adopted in this paper. A maximally underspecified model is regarded as advantageous because it requires less information to be stored by the language learner and allows language properties to be derived using rules. The contrastive underspecification model places an undesirable emphasis on the storage of phonological representations.

6

The second advantage of

radical underspecification is that it incorporates the notion of markedness, in that redundancy rules can be considered as statements defining the unmarked values of features. A universal rule such as [+sonorant] -> [+voice] interprets [+voice] as the unmarked value for [voice] on sonorants. It is noted that certain segments appear to be "transparent"; that is, they do not interfere or block the occurrence of some phonological rules, despite the fact that their feature specifications match the input requirements of these rules. This transparency effect is accounted for by postulating that these segments are not specified for those features involved in the phonological rule. Magnusson (1983), for example, reports that assimilation to labial and velar places of articulation was more frequently attested than alveolar assimilations in her child language data. This observation is accounted for in nonlinear phonology by stipulating that coronal (alveolar) segments are unspecified for place features and are therefore more vulnerable to the spread of dorsal and labial features once these are specified, or

It should be noted that some theories of phonological acquisition emphasize the storage of lexical items before the morphological or phonological segment-level analysis takes place (e.g. Macken, 1979; Ferguson & Farwell, 1975). The current author's view holds that while phrases or lexical items may initially be acquired as unanalyzed units, as soon as sufficient input is provided by the environment, the child replaces these with minimal representations allowing redundancy rules to compute the surface realization. This view is in conflict with Ferguson & Farwell (1975) who "assume the primacy of lexical learning in phonological development" and argue that phonetic words continue to be stored over the life span (p. 437). However, it is consistent with generative grammar accounts of syntax acquisition (e.g. Chomsky, 1964). 6

19

marked, in a child's system. If these features are not yet marked (acquired), the "default" redundancy rule inserts the coronal specification, as is commonly noted when children "front" velar segments (e.g. dog is produced as [dad]). The minimal specifications of the English consonants are presented in Appendix 1.

Prosodic Phonology In addition to developing more precise feature representations, current research is devoting much time to the development of theories that account for how segments combine to form higher level prosodic units. The only hierarchical structure imposed by SPE on a row of segments was that required by English syntax. That is, SPE assumed that the morpheme was the basic unit into which phonemes were grouped. Morphemes then combined to form words, which could be strung together to make phrases and sentences. Researchers pointed out, however, that syntactic bracketing of this nature was inadequate to account for the placement of stress in languages. In addition, the morphological level of representation provided no indication of which sequences of phonemes are considered well-formed in a particular language. Both of these shortcomings could be overcome by referring to the syllable as being intermediate between the segmental and the morphological levels of representation. Hyman (1985) outlines two basic arguments supporting the existence of the syllable. First, stress is assigned to a syllable rather than to a phoneme or a word. This claim is demonstrated by the fact that certain syllable shapes attract stress more readily than others in a particular language (see below for more detailed discussion). Second, by making reference to the syllable, it is possible to account for how segments group into well-formed constituents. Languages only allow certain consonants or combinations of consonants in the syllable onset position, for example. Van der Hulst & Smith (1985) assert that accounting for these two points—stress placement and

20

providing a definition of the notion of well-formed syllable—are the requirements of a satisfactory syllable theory. The reality of the syllable as the basic unit grouping segments into higher order constituents is readily accepted by phonologists. Considerable controversy exists, however, with regard to the internal structure of the syllable. Although several theories have been proposed by different researchers based on their observations of various languages, these theories may be divided into two basic types: those which support the traditional view of dividing syllables into 'onset' and 'rhyme' constituents, and those which divide the syllable into 'weight units' or 'morae'. The traditional onset-rime syllable theory was first introduced by Pike & Pike (1947) and is currently supported in works such as Halle & Vergnaud (1980), Kaye, Lowenstamm & Vergnaud (1987), and Steriade (1988). According to this theory the most basic opposition within the syllable hierarchy is that of onset versus rime (see figure 7 below). The onset directly dominates a syllable-initial consonant or consonants, as with clusters. The rime further branches to a nucleus, usually dominating a single vowel or two vowels in the case of diphthongs, and a coda or margin constituent, which dominates the remaining consonants of the syllable. According to onset-rime theory, the basic syllable shape is CV, as would be indicated by thefirstlevel of branching. Complexity is increased by adding more

Onset

Rime Coda

Nucleus C

C

V

V

C

C

Figure 7: The hierarchical arrangement of the syllable as proposed by onset-rime theory (based on Pike & Pike, 1947).

21

deeply embedded constituents to the syllable shape. The next stage of complexity, therefore, would entail branching within either the onset or rime nodes as the shapes C C V (branching onset) or C V C (branching rime) require. A shape such as C V V is considered to be even more complex because it requires branching at the even finer nucleus level of representation. Figure 7 illustrates that the rime node dominates the nucleus and coda constituents.

This configuration indicates that these sister nodes often operate

together in phonological rules. Two observations illustrate this fact. First, it has been observed in many languages that when the vowel under the nucleus node is deleted, the coda constituent, and never the onset, is resyllabified under the nucleus node. The English word b u t t o n

[bAtsn]

when

pronounced

rapidly becomes

[bAtn]

in which the /n/,

originally dominated by the coda, fills the nucleus position. The second observation supporting the sister relationship of the nucleus and coda within the rime was first pointed out by Jakobson (1931, 1937) and Trubetskoy (1939). These phonologists noticed that the assignment of stress is often sensitive to the structure within the rime. Specifically, they observed that "heavy" syllables—that is, closed syllables or those with long vowels —attract stress, while open or "light" syllables with short vowels do not. This generalization is captured by the statement that stress is assigned to syllables with branching rhymes (where branching can occur either at the level of the rime or within the nucleus). It should be noted that the status of the onset is irrelevant when assigning stress. As previously noted, van der Hulst & Smith (1985) state that a satisfactory syllable theory should account for the assignment of stress in a particular language. In order for onset-rime theory to fulfill this requirement, the formal representation must indicate that only the rime distinguishes between heavy and light syllables and that onsets are "weightless."

Hyman (1984) argues that resorting to the geometry of the

rime, as onset-rime theory does, is not sufficient for determining syllable weight. He

22

points out that some languages only regard syllables with long vowels (branching nucleus) as heavy but treat closed syllables (branching rime) as light. Still other languages, though counting CVC syllables as heavy, treat CV? sequences as light despite the fact that a glottal stop appears to be filling the coda position. Thus, it is sometimes necessary to look below the syllable hierarchy to the feature content in order to determine phonological weight. In addition, Hyman points out that there is nothing in the formal representation of onset-rime theory which indicates that the syllable onset is weightless and thus plays no role in stress assignment. In order to capture the generalizations regarding stress placement, individual languages must add definitional statements to the syllabic representation, a move which adds a considerable burden to the learnability of the system. Recently, phonologists such as Hyman (1984), McCarthy & Prince (1986), and Hayes (1989) have proposed that generalizations regarding syllable structure are better captured using 'weight units' or 'morae' rather than an onset-rime hierarchy. These morae constitute a weight tier which is intermediate in the hierarchy between the segmental feature geometry and the syllable tier. According to Hyman (1985), each mora "defines a beat or peak of sonority" within the syllable (p. 20). Supporters of the weight unit theory of syllabification agree that the moraic structure of languages can vary. Essentially languages differ in the rules required for assigning moraic structure depending on how that language distinguishes heavy syllables from light ones, as discussed above. Recall that some languages regard CVV and CVC syllables as heavy while only CV shapes are light. Other languages, however, only consider CVV syllables as heavy and hold that both CVC and CV syllables are light. Once this heavy/light syllable distinction has been established for a certain language, morae may be assigned: light syllables contain only one mora and heavy syllables contain two. In this sense, morae reflect the weight of a syllable.

23

Figure 8: The internal constituent structure of the syllable as proposed by moraic theory (elements in parentheses are optional).

According to moraic theory, syllable-initial consonants attach directly and individually to the syllable node. This view, therefore, unlike the traditional syllable theory, does not maintain any opposition between the onset and the rest of the syllable (see figure 8). Controversy exists over the representation of post-vocalic (coda) consonants.

Some researchers (Hyman, 1984; Hayes, 1989) claim that

syllable-final consonants should be linked to the mora of the preceding vowel (or consonant if the language considers closed syllables to be heavy), while others (Clements & Keyser, 1983; Kiparsky, 1980) argue that they should, like onsets, attach to the syllable node. The position adopted in this paper is that coda consonants are adjoined to the mora of the preceding consonant (see, however, chapter 5 and Hayes (1982) for discussion of extrametrical consonants). Based on observation of many languages, it has been suggested that the maximum number of morae allowed in a syllable is two (McCarthy & Prince, 1986). All languages, then, must be either monomoraic, meaning they do not distinguish between heavy and light syllables (e.g. Japanese), or bimoraic and therefore sensitive to syllable weight. English falls into the latter bimoraic category and allows any sonorant (vowel or consonant) to fill the first mora position in the syllable (recall the button example in which the word final syllabic /n/ moves into the position of the first mora in the syllable during rapid speech). English allows the second mora position in the syllable to be filled either by a vowel if the syllable contains a diphthong, or by a

24

Figure 9: Moraic representation of the syllable hierarchy for the word pin.

[+sonorant] consonant (see figure 9). Obstruents may never fill this position in English. The basic syllable according to moraic theory is a monomoraic CV syllable where the vowel is dominated by a mora and the onset consonant is directly attached to the syllable node. The next level of difficulty would entail establishing a bimoraic syllable as is required by diphthongs (CVV) and closed syllables (CVC, where the final consonant is [+sonorant]). Moraic theory disregards onsets in the establishment of syllable structure; it is difficult to discern, therefore, whether a CCV syllable (monomoraic) is actually considered to be less complex (and thus easier to acquire from the child language acquisition point of view) than a CVC (bimoraic) syllable. Bernhardt (1990) suggests that the development of multi-consonant sequences could occur at any time, but perhaps could be delayed until the critical parameters for syllable structure (i.e. bimoraic syllables) have been set. Contrasting the internal structure of syllables according to both moraic and onset-rime theory, several empirical differences become obvious. First, the distinction between heavy and light syllables is more defined in the moraic theory representation; onset-rime theory must make reference to language-specific rules in order to make this distinction clear. Second, moraic theory does not place any importance on onset segments while onset-rime theory maintains the opposition between these syllableinitial constituents and the rime. According to the onset-rime view, then, consonant

25

clusters in onset position form a unit; the moraic theory provides no representation of such a unit and simply appends each consonant individually to the syllable node. A third difference arises from the way each theory treats vowel constituents. Onsetrime theory places all vowels (short, long and diphthongized vowels) under the nucleus node. Moraic theory, however, represents long vowels and diphthongs using two morae which attach individually to the syllable node. In this sense, moraic theory treats the second vowel of a diphthong as equivalent to a consonant in establishing heavy syllables. Recall that onset-rime theory considers a CVV syllable to be more complex than a CVC syllable because the former requires a deeper level of branching within the nucleus node. Finally, the representation of onset-rime theory presents a richer hierarchy using branching at each level, while moraic theory utilizes a more linear representation of the syllable.

Syllables and Segments Combined 1

An outline of possible syllable hierarchies, as well as a feature hierarchy, leads

to the question of how these two levels of phonology may be combined. McCarthy (1979) originally proposed that a 'skeletal tier', made up of C's and Vs, mediates between the syllabic and segmental hierarchies. Clements & Keyser (1983) and Christdas (1988), who currently support the existence of a CV tier, point out that this tier provides the input to the syllable hierarchy which, in their view, builds not on segments but on consonant and vowel constituents. According to these phonologists, the CV tier "can be considered to contain quantitative information about a phonological string" (Christdas 1988: 23). In other words, the skeletal tier provides a count of the number of segments making up a syllable unit; such a count is not provided elsewhere in the phonological derivation. Levin (1985) and Lowenstamm & Kaye (1986) conclude that the use of Cs and Vs in the skeletal tier is unnecessary because the consonant/vowel distinction may be distinguished by the higher level syllable

26

structure. These researchers support the onset-rime syllable theory and point out that vowels are always dominated by nucleus nodes, while consonants fall under onset and coda nodes. As a result, Levin (1985) and Lowenstamm & Kaye (1986) have replaced the C and V symbols with a sequence of X's. It should be noted that the difference between a C V tier and an X tier is minimal; both views provide a measure of the number of segments present. Hyman (1984), McCarthy & Prince (forthcoming) and Hayes (1989) support the view that no skeletal tier is needed to mediate between the syllable and segmental hierarchies. Hayes (1989) points out that phonological rules have been observed to count morae and syllables, but no known phonological process exists that counts segments. Thus, a measure of the number of segments making up a syllable is not needed. What is required, however, is a measure of syllable weight, and this may be provided by the moraic theory supported by these phonologists. According to Hyman (1984), McCarthy & Prince (forthcoming) and Hayes (1989), then, the ROOT node of the segmental hierarchy is directly dominated by the syllable node, in the case of onset consonants, or a mora, in the case of all other constituents. Just which of these views—the C V tier, the X tier or no skeletal tier—best accounts for child language acquisition data remains unclear. Reports based on the presence of morphophonemic alternations indicate that some level of mediating tier may be necessary. A child, for example, who produces [da] for dog and [dagi] for doggy, and who recognizes the relationship between these two words, appears to realize that a /g/ is underlyingly present in dog but is simply unable to produce syllable final consonants (Dinnsen et al., 1980). This realization could be represented by an empty X-slot-that is, a node with no feature values-if a skeletal tier is present in the representation. In addition, intuitively it makes sense that phonological rules and derivations should have access to phoneme-sized units at some level of the representation.

27

The Acquisition Process Theories of nonlinear phonology as described above appear to present an overwhelming task for the language learner. In reality, however, the acquisition process can be described by making reference to syllable and segmental hierarchies. Based on the predictions of Universal Grammar, the child comes to the language learning task with the established basic C V syllable. From this level, two possibilities for further acquisition exist. The child may either add branches to the syllable hierarchy, by developing a C V C syllable for example, or may develop the feature hierarchy and increase the inventory of segments which may fill the C or V slot of the established C V syllable shape. Based on her work with phonologically disordered children, Nettelbladt (1983) generalizes that severely disordered children have undeveloped syllable structure, whereas children with mild-to-moderate disorders have difficulty at the segmental level. Bernhardt (1990) suggests that this fact is accounted for if the syllable level of representation is considered a more basic (meaning "not as embedded" in this case) level than the segmental level. This suggestion is exemplified by Pollock's (1983) description of a severely disordered child who was able to increase his segmental inventory only after the acquisition of closed syllables. It is possible, therefore, that certain levels of complexity in the syllable hierarchy coincide with levels within the segmental hierarchy and that these levels must be attained before the feature hierarchy can continue to develop. Further research is required to substantiate such a claim. A concluding remark addresses the nature of the acquisition process described above. Phonological acquisition, in the view of nonlinear phonology, can be described as a process of building up the syllable and segmental hierarchies. At no point is the child viewed as deleting features or segments from a more complex underlying representation. A syllable shape or segment missing from a child's phonological

28

inventory is, therefore, not considered to be "simplification" of the output. Instead, the child has not yet acquired certain branches or levels of the hierarchy. To date very few researchers have attempted applying the recent nonlinear advances in phonological theory to child language acquisition data or data taken from the speech-language pathology literature. The first attempt, by Gandour (1981), used Kahn's (1978) syllabic theory of phonology to present a reanalysis of the data of a phonologically delayed subject. Following this lead, Spencer (1984) reanalyzed the data presented in Grunwell & Pletts (1974) using a nonlinear framework for description. Similarly, Spencer (1986) used nonlinear principles to present a reanalysis of phonological acquisition data of the typically developing child first outlined in Smith (1973). More recently, Stemberger (1988) presented nonlinear analyses of processes operating across word-boundaries in the phonology of his daughter. A l l of the cited research show that when phonological acquisition data is examined from the point of view of nonlinear phonology, the data no longer appears to be deviant or idiosyncratic, but instead, can be seen to follow the general rules and principles governing the languages of the world. If nonlinear phonology provides a superior means of accounting for language acquisition data, then it should follow that the principles of this theory should have validity for the remediation of phonological delays. The only research to date which has focussed on the application of nonlinear principles in the clinical setting is that recently completed by Bernhardt (1990) and the research findings presented here.

29

CHAPTER 3 METHOD In order to explore the application of nonlinear phonology in the clinical setting, a longitudinal intervention study was carried out with a set of identical male twins, aged 5;7 at the onset of the project. The boys are from a middle-class family where English is the only language spoken in the home. There are no other siblings in the family. The twins had had no speech or language therapy prior to that received in the study. This point is important because it meant that no "biases" from previous intervention needed to be accounted for or overcome in therapy. Instead, each twin's phonological inventory at the commencement of the study was one which had evolved spontaneously. The methodology and design of this study was adapted from that of Bernhardt (1990), with a few changes introduced in order to accommodate the twin subjects of the current project. Bernhardt's study employed both a multiple baseline and an alternating treatment design. The details of these will be developed below.

Initial Assessment Standardized language tests were administered at the beginning of the study by the speech-language pathologist who first identified the twins as possible subjects for the project. These measures included: a) The Peabody Picture Vocabulary Test - Revised (Dunn & Dunn, 1981)

30 b) The Expressive One-Word Picture Vocabulary Test (Gardner, 1981) c) The Structured Photographic Expressive Language Test-II (O'Hara Werner & Dawson Kresheck, 1983) d) The word discrimination subtest of the Test of Language Development - Primary (Newcomer & Hammill, 1982) e) The Test of Auditory Comprehension of Language Revised (Carrow-Woolfolk, 1985) These tests were readministered by the same examiner at the end of the project and made up part of the multiple baseline design. Although syntactic and pragmatic goals were never the focus of therapy sessions, the possibility existed that activities which emphasized the use of phonological targets in sentences and natural conversation would also have a positive effect on general language ability. The results of the poststudy language measures are discussed in chapter 6. In addition to the above, the parents were asked to fill out a detailed case history form, which provided further information regarding the medical background of each of the twins, as well as the acquisition of major motor and speech and language milestones. The initial phonological assessment was recorded on Ampex 631 tapes using a Nagra IV-D reel-to-reel tape recorder with an A K G D-202 E l microphone. This assessment was administered to each of the twins by the author of the current study. The assessment procedure, resembling that of Hodson (1980), involved eliciting single words using objects and pictures. The word list used in this study, developed by Bernhardt (1990), included words which assessed the production of "all consonants in syllable-initial word-initial position, syllable-final word-final position, intervocalic medial position and, where possible, syllable-final within-word position." In order to see whether the complexity of the word in which it appeared affected the production of a consonant, segments were elicited in C V , V C and C . V C . shapes when such words existed, as well as within clusters and polysyllabic word shapes. Finally, selected

31

nouns with [i] added to the end (e.g. gum -> gummy ), and the present progressive 1

form (-ing) of several verbs (e.g. run. -> running') were elicited in order to assess whether morphophonemic alternation had any effect on the production of word-final consonants (which became word-medial with the addition of the endings). A list of these words is presented in Appendix 2. The phonological assessment was supplemented with an oral mechanism examination based on suggestions presented in Robbins & Klee (1987) and an evaluation of the stimulability for production of those phonemes not evidenced in the twins' productive inventory. The recordings of the phonological assessments were transcribed noting any spontaneous and elicited repetitions, as well as self-corrections.

A Revox reel-to-reel

tape recorder and Sennheiser H D 540 headphones were used for this procedure. The data was then analyzed using the phoneme and cluster realization worksheets and the phonotactic analysis worksheet of Grunwell's Phonological Assessment of Child Speech (1985). Based on the results of these analyses, targets for remediation were chosen. These targets fell within one of two conditions (segmental versus syllabic condition, see below), to which the twins had been randomly assigned.

Segmental Condition The main question that remediation within the segmental condition was to answer was whether the feature geometry of nonlinear phonology can predict which segments should be targeted first in therapy. As outlined in the previous chapter, the feature hierarchy predicts that segments which have their "salient" or most characteristic feature at a higher level in the hierarchy should be more quickly acquired than segments with lower or more deeply embedded salient features. As a result, two segments—one representing a higher level, and one representing a lower level target of the feature hierarchy—were chosen as goals for the first block of therapy. Alternate training of two segments not only allowed us to test the predictive power of nonlinear

32

phonology, but it also allowed for intensive training of one segment followed by a period where the child could concentrate on that speech sound and generalize it to conversation on his own, while therapy continued with the second target. Such a cyclic approach was first introduced by Hodson & Paden (1983). Therapy did not involve targeting all segments within a sound class. It was assumed, and hoped, that the acquisition of one segment would generalize to segments with similar feature geometries serving as baseline measures. The acquisition of /z/, for example, was expected to generalize to the production of /s/, since the two segments differ only by the laryngeal feature [voice]. It should be noted that the traditional notion of sound class—that is, segments which function similarly in a language rule—is captured in nonlinear phonology in terms of similarity of feature geometry. Words used in therapy activities did not include any of the test items of the major and minor probes (see below). In addition, training words consisted only of syllable shapes already present in the twins' productive inventory. A l l segments other than the target segments of these words were also present in the twins' production repertoire. Therapy focussed on the acquisition of the target in all word positions.

Syllabic Condition The second twin was assigned to the syllabic condition, which focussed on establishing new syllable shapes in the productive inventory. For the twins this entailed targeting word-initial consonant clusters. A l l clusters targeted were made up of segments the twins were producing consistently in at least one word position. As with sound classes in the segmental condition, not all clusters were targeted in therapy. Some served as baseline measures and remained untrained to see if generalization from targeted clusters would occur.

33

As described in the preceding chapter, there is considerable controversy over how the syllable hierarchy should best be represented. The main purpose of training within the syllabic condition, then, was to see which approach resulted in faster acquisition of therapy goals, either an approach reflecting the moraic theory or one reflecting the onset-rime theory. Recall that faster acquisition time is thought to occur because the syllabic theory used in training more closely resembles the way in which the mind organizes the syllable. A second test for the "correctness" of this theory is the amount of generalization that occurs across untrained clusters. Details regarding the training differences between these approaches and the results will be discussed in chapter 5. An alternating treatment design was also used in the syllabic condition. The difference here, however, is that the goals of each condition were of equal difficulty so that the difference in acquisition time could only be attributed to the training approach. In the segmental condition, it was hypothesized that the two segments were of unequal difficulty because of their respective feature geometries, and that acquisition time would reflect this difference. Consonant clusters were considered to be of equal difficulty if their differing segment reflected similar features. With s-clusters, for example, /sn/ could be trained under one theory and compared with /si/ trained under the opposite theory. In this case the features of /n/ and l\J are considered to be similar and, therefore, cluster acquisition should reflect equal difficulty. It was not always possible, however, to keep both conditions exactly the same; thus it was occasionally necessary to resort to balancing an easier and a more difficult cluster against two clusters of medium difficulty (see chapter 5 for details).

Therapy Schedule The twins in this study were seen three times per week for individual therapy sessions which lasted thirty to forty-five minutes each. Twice a week therapy took

34

place at their elementary school, and once each week they were seen at their daycare centre. The twins alternated being the first to receive therapy. The intervention programme was divided into three blocks of therapy, and each block was divided into two cycles (see figure 10 below). Half of the cycle (three and one half sessions) was spent targeting one feature or training clusters using one syllabic approach, while the second half of the cycle (another three and one half sessions) targeted the second feature or syllabic approach. A review of the cycle's targets took place on the eighth session. The ninth and final session of the cycle was a "minor probe" which assessed the acquisition progress of treatment targets by eliciting them in single words not trained in therapy (as before, pictures and objects were used).

Treatment Block (19 sessions) Two cycles (9 sessions each) session 1 session 2 session 3

therapy targeting segment 1 (higher feature in hierarchy) or word-initial clusters using moraic approach

session 4 session 5 session 6

— therapy targeting segment 2 (lower feature in hierarchy) or word-initial clusters using onset-rime approach

session 7 session 8 review of all goals of cycle session 9

minor probe one week vacation from therapy

i

Major Probe (the 19th session of the block) Figure 10: Summary of the design of one therapy block

35

The production of therapy targets in connected speech was assessed by having the twins retell stories in which the target segment occurred frequently (Clinical Probes of Articulation Consistency [Secord,1981] stories were used for segmental goals, while cluster goals were targeted using stories created by Bernhardt). Because retelling the stories proved to be a difficult task for the twins (their auditory recall was below age level), this procedure often resorted to conversations about the accompanying pictures. This change allowed the targets to be produced in a natural way and provided reliable results. Where generalization to specific segments or clusters was expected, these were also tested using the same single-word elicitation and connected speech procedures. The third purpose of the current study was to assess the possibility of one twin influencing the other's learning without direct training from the speech-language pathologist. That is, would one twin acquire the other's targets simply because of the amount of time they spend playing and conversing with each other? In order to evaluate whether any "cross-talk" was occurring, the minor probe for each twin included the targets of both the segmental and the syllabic conditions. If the minor probe results indicated that a target segment or cluster had been acquired and generalized to conversational usage, then new targets within the same condition were chosen as goals for the second cycle of the block. If there was no evidence of acquisition, then the cycle was repeated using different activities to maintain interest. The second minor probe was followed by a week's vacation whose purpose was to ensure that the major probe, administered immediately after, reflected the spontaneous productive inventory of each twin. The major probe was a repetition of the phonological assessment administered at the beginning of the project. The first two blocks of therapy targeted goals within the conditions to which each twin had been assigned. After the first block an environmental reinforcement programme was initiated. This addition consisted of the twins' daycare workers

36

providing verbal reinforcement for correct spontaneous productions of therapy targets. At the same time each twin was awarded a check mark which could be turned in to the examiner for a prize once a certain number of checks had been earned. Goals for the third block of therapy were chosen based on gaps in the twins' phonemic inventories. In most cases this meant training targets from the opposite twin. The parents became involved in the study at this point by playing games and completing therapy activities with the twins at home. The involvement of the parents had not been sought before this time in order not to confound any natural cross-talk that might occur between the twins.

Therapy Procedure The design of individual therapy sessions was relatively consistent throughout the project and followed the order: 1) Auditory awareness training consisted of reading stories and poems or singing action songs and poems that focussed the child on the segment or cluster goal of the session. This activity was usually done under amplification. 2) Auditory discrimination training contrasted the target segment and the twin's production using minimal pairs. 3) A series of production tasks whose level of difficulty depended on the twin's production ability followed the auditory tasks. Initially, these activities focussed on imitation in a game or role-play situation, but moved on to production in short phrases and sentences in conversation as quickly as possible. recorded on Fuji C r 0

2

A l l treatment sessions were

(FR-II) tapes using a Marantz P M D 220 cassette recorder and

an A K G SA 41/1 microphone. The data reported in the following chapters is drawn from the single-word elicitation tasks of the minor and major probes, from the story-retell conversations of the minor probe and, where specified, from the production data of individual treatment

37

sessions. Language samples were recorded at the time of each of the major probes. In addition, connected speech occurred (and was encouraged) during minor probes and therapy sessions. Where conversational use of segments differed from single-word use, it will be specified. In general however, it was found that the probes, especially the major probes, accurately reflected the subjects' spontaneous use of all segments.

38

CHAPTER 4 T H E S E G M E N T A L CONDITION

When speech-language pathologists are faced with assessing a phonologically delayed child, they typically begin by administering an assessment tool, usually requiring the child to label pictures or objects, in order to determine appropriate intervention goals. As previously noted, the bases for choosing one sound or process for remediation over another are relatively ad hoc and depend on suggestions such as following the sequence of acquisition of typical children or remediating those sounds which would gain the child the most communicative success. Utilizing a nonlinear framework to determine intervention goals requires the clinician to first assess the productive phoneme inventory of the child. This inventory is then analyzed according to the minimal specifications of each phoneme outlined in Appendix 1, and the child's established feature hierarchy is determined. From here, undeveloped nodes and branches can be outlined and decisions regarding remediation are made. Adopting a nonlinear framework for determining the segmental targets for remediation in the current phonological intervention project led to the analysis of three questions. First, does the feature geometry lead to the accurate prediction of treatment targets? Recall that segments having their most salient feature at a higher level in the hierarchy are predicted to be acquired more quickly than segments with lower or more deeply embedded salient features. The second question investigated whether establishing a feature required for the acquisition of one segment would generalize to the spontaneous acquisition of targets requiring the same feature. Would the acquisition of [+continuant] for /f/, for example, generalize to the acquisition

39 of hi or Isl, which also require [+continuant] to be established? Spontaneous generalization of this nature would provide independent evidence in favor of targeting one level of features over another. Finally, the third question resulted from using twins as subjects for this study. Specifically, it was asked whether the twin placed on the syllabic condition would acquire the segmental targets without any direct intervention. The spontaneous acquisition of one target before another would provide further evidence in favor of targeting higher versus lower features. Before dealing with the responses to these three questions in detail, the segmental inventory of the twins at the commencement of the project will be described.

The Starting Point Initial goals for treatment were determined based on the analysis of the data elicited at the first grand probe. The transcribed single-word data is presented in Appendix 3. At the outset of the study the twins had moderately-severe phonological delays. Based on parent report and observation of them playing together, the twins were completely intelligible to each other. The parents reported that they could understand the twins 100% of the time, but they estimated that strangers could comprehend them only 75% of the time. This latter estimate was somewhat generous. The twins entered the study with identical phonetic inventories. Because they were not discriminating between their production and the adult target form, these also represent their phonological inventories. The twins provided identical productions for almost all target words, with only one significant exception which will be addressed in the next chapter. It is worth mentioning that none of the twins' surface realizations of adult targets was unusual; all substitutions are commonly cited in the phonological acquisition and disorders literature. Since neither of the twins had difficulty producing the vowels of English, these sounds are ignored in the present study and only consonantal data is analyzed. The consonantal segment inventory for word-initial,

40

p

t

k

b

d

g

m

n

w

j

Figure 11: Word-initial phonetic inventory of both subjects of this study. p

t

k

b

d

g

m

n s

w

rj

Figure 12: The syllable-initial-within-word phonetic inventory for both subjects of this study. p

t

k

b

d

g

m

n

n

s r Figure 13: The word-final phonetic inventory for both subjects of this study. syllable-initial-within-word, and word-final positions are outlined in figures 11, 12 and 13. The segment [1] in figure 13 is entered in angle brackets because of its marginal status for both the twins. The status of HI was difficult to discern in word-final position from the recordings; for both of the subjects it was judged to be an [1] rather

41

than a vowel only 50% of the time. The rest of the segments were well established for each twin. In descriptive terms it can be seen that the twins had not yet established fricatives, affricates, liquids and what shall be described here as a laryngeal fricative (/h/) in their word-initial inventory. The syllable-initial-within- word and word-final inventories showed essentially the same productive segments as for word-initial position, with the addition of the liquid [r] and the fricative [s]. This latter segment was very productive since voicing distinctions were not being marked in the fricative class (i.e. fricatives all surfaced as [s] in medial and final position regardless of the laryngeal status of the adult target). It should be noted that the twins were attempting to mark the [-continuant] [+continuant] status of affricates. In word-initial position these segments surfaced primarily as [t] and [d], but in medial and final word positions they were more consistently produced as [ts] and [ds]. Finally, note should be made of the subjects' 'default' segment, [j], which was produced for both / l / and /h/ in initial and medial word position. The term 'default' is used here to refer to a robust segment which the language learner substitutes for complex unestablished segments. When the nonlinear minimal specifications of each of the productive segments are combined, the established feature hierarchy of the twins is determined (see figure 14). In figure 14, the terminal node [+voice] is required under the L A R Y N G E A L class node to characterize the voiced obstruents. The 'labial' node accounts for the bilabial stops and nasal, while the [+round] terminal feature is required in the specification of the glide, /w/. The 'dorsal' node is the salient feature of the velar consonant specifications and [+nasal] is the same for the nasal consonants. The [consonantal] feature is required for the specification of all consonants.

42

SKELETAL TIER ROOT NODE LARYNGEAL NODE

[consonantal] [sonorant] [+nasal] [+continuant]* x

[+voice] PLACE NODE labial

coronal

dorsal

[+round] Figure 14: The established feature hierarchy at the beginning of intervention for both subjects. (* The 'complex' aspect of affricates was partially established.)

The status of the glides, Av/ and /j/, with respect to the feature [consonantal] is uncertain. The position adopted here is in agreement with Hyman (1985) who argues that glides are really vowels filling consonantal slots (p.77). This position is reflected by the lack of [consonantal] in the minimal feature specification and the marking [sonorant] which is required in the characterization of all vowels, glides and liquids. The segments /w/ and /j/ remain unmarked for the [consonantal] feature because this status is redundant by the position these segments occupy in the syllable. Since 1)1 is not marked for [consonantal], its minimal specification consists solely of [sonorant] which is also the minimal specification of the neutral shwa vowel in English. Sharing this representation should not confuse the language learner or 7

Elsewhere it has been argued that [-anterior] under the 'coronal' place node is required in the minimal specification of / j / and that /w/ is the most underspecified of the glides (Shaw, personal communication; Bernhardt, 1990). Based on the fact that / j / surfaced as the twins' default segment, it is argued here that / j / should be the least specified of the glides. It should be noted that [-anterior] is not required to differentiate / j / from other English segments provided that the default 'coronal' place node is supplied by redundancy rules. 7

43 the application of phonological rules, because the position the segment occupies in the syllable defines whether it surfaces as shwa or /j/. In other words, syllable onsets, which must be filled by consonants, will surface as /j/, while syllable nuclei or slots dominated by morae (depending on the preferred theory) will always be realized as shwa. This account of 1)1 provides a neat analysis of the twins' favorite segment. Essentially, the twins' underlying representation of l\J and fhj in syllable-initial position (both word initially and within a word) correctly identified a consonantal slot. However, the features of the adult target were not yet established in their productive inventory. Their solution to realizing this empty slot was to fill in the [sonorant] specification by default which necessarily surfaces as /j/ because of its syllable position. Alternatively, it could be hypothesized that the [sonorant] feature spreads from the following vowel to the preceding empty slot. However, the default segment hypothesis is more appealing in this case since children often have a frequent segment which is robust in their phonological inventories for a time. The minimal feature specifications required by these default segments cannot always be provided by the spreading of features from adjacent segments. The feature [-(-continuant] in figure 14 is appended to the root node using a dashed line to indicate that it was not established in all word positions. Recall from the segment inventories presented above, that [+continuant] was only established in syllable-initial-within-word and final word position. This branch had not yet been established in initial word position. The same notation is used for '+complex' which was not established in initial position. Up to this point, it has been possible to describe the twins' phonological inventories at the outset of this project by simply stating the established branches of their productive hierarchy. That is, in every case thus far, the twins' surface realizations of the adult targets result from the fact that they had not yet established

44

ROOT

ROOT [consonantal]

[consonantal]

([•fcontinuant])

[+continuant]

PLACE /

(PLACE)

labial

(coronal) ([+distributed])

Figure 15

The twins' realizations of word-initial HI (left) and word-final /9/ (right) with unestablished nodes marked in parentheses.

certain branches of the adult feature geometry. This finding is illustrated in figure 15. The features in parentheses are those which needed to be established in order for the twins' surface realizations to match the adult target phoneme. The exception to the above description arises in the case of Ihl and HI, where the default segment surfaces (although the surfacing of [j] for HI could also be accounted for by saying that [consonantal] has not yet been established, together with the specification [sonorant]). The case of /r/ surfacing as [w], however, does not have as simple, or as satisfying, an explanation. ROOT [sonorant] ([consonantal])

PLACE

Figure 16: The nonlinear specification of the feature changes required to realize /r/ as [w]. (Underlined features are those which surface in the twins' realization.)

45

In order for the twins to realize /r/ as [w] one needs to state that the [consonantal] feature is not yet established in the environment of [sonorant]. Thus far, the same argument applies to the adult target HI. In the case of /r/, however, it is not [j] that surfaces but [w]. The minimal specification required for/w/ includes the terminal feature [+round] dominated by the 'labial' node. In essence, the twins were not realizing the [+distributed] feature of the adult target and instead were acknowledging the redundant secondary features of /r/ (see figure 16). The difficulty with this argument is that redundancy rules do not fill in the full specification of phonemes until very late in the derivation (in the twins' case the default segment is filled in before redundancy rules are applied). Hypothesizing that full specification is complete earlier in the derivation is undesirable because of the loss of simplicity in the explanation and the increased burden on learnability. A better explanation for the /r/~>[w] phenomenon does not present itself at this time (see, however, the concluding chapter of Bernhardt, 1990).

Determining Targets for Remediation Goals for the segmental condition of this project were determined based on a comparison of the adult feature geometry for English, illustrated in figure 6 (see p. 16), and the twins' established hierarchy described above. As previously mentioned, the [+continuant] feature needed to be established in initial position for all fricatives. The terminal features [-anterior] and [+distributed] under the 'coronal' place node were both undeveloped at the outset of this project. One could say that the twins' production of [ds] and [ts] for the affricates in syllable-initial-within-word and wordfinal positions is evidence of the designation 'complex'. Recall that Bernhardt (1990) inserted this mnemonic simply as a way of working around the controversy that exists in the phonological theory literature regarding the feature specification of affricate

46

segments. The twins had not yet established this 'complex' specification in wordinitial position. Finally, the subjects of this project had not established the feature [consonantal] in the presence of [sonorant] in word-initial position as is required for the realization of both PJ and /r/. Based on the undeveloped branches of the twins' established feature hierarchy, segmental pairs for remediation were chosen which allowed for the comparison of targeting lower features of the adult geometry with higher ones. Individual session objectives are listed in Appendix 4. The segments /h/ and HI were initially targeted with Subject 1 who was randomly assigned to the segmental condition. The minimal specification of /h/ is [consonantal] [+continuant]. A l l [sonorant] segments are redundantly [+continuant]. Since the twins were producing [j] for /h/, the feature [•fcontinuant] was already established and only [consonantal] remained unestablished for the specification of /h/. The feature that the twins needed to establish in order to acquire /h/, then, was [consonantal]. This feature was well established for consonants whose place of articulation is in the oral cavity, but still needed to be acquired for this laryngeal fricative. The teaching of the segments /h/ and

HI allowed the acquisition of the feature [consonantal], the salient feature for /h/, to be compared with [-fcontinuant], required for HI in word-initial position. Notice that in medial and final word positions the comparison is between the acquisition of [consonantal] and the 'labial' place feature since [+continuant] is already established here. Despite the fact that different nodes are being compared depending on word position, the prediction remains the same. The feature [consonantal] occupies a higher position in the feature hierarchy than [+continuant] and is, therefore, expected to emerge earlier. Recall McCarthy's (1988) claim that [consonantal] and [sonorant] are the highest features in the geometry because they actually form the R O O T node

47 while the manner nodes, [+continuant] and [+nasal] are only appended to it. In segmental terms, then, fhf was expected to emerge before Ifl. In syllable-initial-within-word position training the production of Ihl and Ifl allows for the acquisition of [consonantal] and 'labial' to be compared. As with wordinitial position, Ihl is expected to emerge before Ifl because [consonantal] is less embedded than 'labial'. The difference in the acquisition of Ifl due to word position leads to interesting speculations regarding the difficulty of the learning task. One wonders if it is easier to establish the feature [+continuant] while maintaining (essentially) the same place of articulation, or whether it is easier to develop the 'labial' place node once [+continuant] has already been established. It should be noted that 'labial' was 8

already established in medial and final word positions as an obstruent; the difficulty is in establishing 'labial' together with [+continuant] to create Ifl. Based on the "embeddedness" of these features, the acquisition of Ifl in word-initial position should be easier, since [+continuant] is directly attached to the root node. Conclusions regarding the acquisition of /h/ and Ifl will be discussed below. The second cycle of Block 2 compared the acquisition of /dg/ and /J/ with that of hi and If/. A valid criticism of this choice of targets points out that Ifl has the unfair advantage over the other segments because it had already received three cycles of therapy. Since essentially no progress on Ifl was made during this training time, it was decided that pairing the labio-dental fricatives with the alveo-palatal segments might provide some interesting data. The voiced cognate of Ifl was added to the training task with the hope that the increased vibration of the lower lip against the

Note that a separate 'labio-dental' place of articulation is not required, since English has no phonemic bilabial fricative. The 'labial' node is sufficient for the minimal specifications of both bilabial and labio-dental places of articulation. 8

48

teeth because of voicing would make this pair of targets more salient for Subject 1. Since all four segments have a continuant component in their specification (affricates are considered here as sequences of [-continuant] [+continuant]) and this is the only branch required for the twins' realization to match the adult target in word-initial position, no comparison can be made between lower and higher features. In medial and final word positions, however, where the subjects realize HI as [s], establishing the 'labial' place node can be compared with developing [-anterior]. In this case, the feature hierarchy predicts that 'labial' would be easier to acquire because it is less embedded than the [-anterior] feature of the 'coronal' place of articulation.

Results.

The first mini probe was administered after eight therapy sessions, which targeted first /h/ and then /f/. This mini probe, which assessed any generalization of treatment targets using a single word and connected speech production task, found that no acquisition of targets had occurred by this time. The compiled mini probe data for segmental targets is presented in graphed form for both subjects in figures 17 and 18 (see pages 49 and 50). Percentage accuracy figures are based on the combined results of imitated forms and spontaneous forms. Recall that mini probes took place every three weeks (or twice each block) throughout the project. By the second mini probe at the end of Block 1, a minimal increase in the accuracy of producing /h/ and /f/ is recorded. Given that the percentage accuracy of pronunciation in subsequent mini probes for HI drops off, the slight increase in accuracy at the end of Block 1 is considered coincidental. As is illustrated in figure 17, however, the accuracy of /h/ production continued to increase until it reached 100% at mini probe #5. The acquisition of HI clearly followed that of Ihl for both Subject 1 and Subject 2. As illustrated in figure 17, essentially no generalization to novel single words

49 containing HI, nor to the story retell activity was noted until mini probe #6, at the end of Block 3 of the study. Just five days prior to this last mini probe, Subject 1 greeted the author with the question "Can I be first today?" This author suspects that a breakthrough had occurred a day or two before and Subject 1 had planned this greeting with a perfectly articulated first, to show that he could finally produce the target that had been the focus of therapy for so long. Results on the sixth mini probe revealed 87% accuracy on single word targets and 67% accuracy on the story retell task for HI (hence the 72% combined result plotted in figure 17). It should be noted that the first cycle of Block 3 did not target HI or /v/, but trained the syllabic targets of Subject 2 and the alveo-palatal targets (see session objectives in Appendix 4). Based on her results with singleton subjects, Bernhardt (1990) suggests that alternating syllabic and segmental goals is beneficial for acquiring both types of targets. Other explanations for this late breakthrough of HI present themselves. It is possible that the one-and-one-half cycle break from targeting the labio-dental fricatives allowed Subject 1 to pay attention to the occurrence of these segments in natural conversation and subsequently generalize their production to conversation himself (Hodson & Paden, 1983). Alternatively, success with the syllabic /s/-cluster targets effectively established the [-fcontinuant] feature in word-initial position, which may then have transferred to the acquisition of HI in this position. Just which of these explanations holds in Subject l's case cannot be concluded from the data here; quite possibly the sum of all three explanations was required before HI could generalize to spontaneous conversation. As previously outlined, it was speculated that HI in word-initial position would be more easily acquired because the needed [-fcontinuant] feature was less embedded than the 'labial' place node required for HI in medial and final word positions. Although HI was acquired in spontaneous conversation in all word positions simultaneously, its production in initial training sessions was superior in medial and final word positions.

0

1

2

3

4

5

6

mini probe

Figure 17: Graph of the segmental targets based on mini probe results for Subject 1.

0

1

2

3

4

5

6

mini probe

Figure 18: Graph of the segmental targets based on mini probe results for Subject 2.

52 Subject 1 had no difficulty approximating the correct production of Ifl during imitation tasks when the target was not word initial. He tended to produce a labio-dental stop following a word-initial /f/ before articulating the remainder of the word. At the end of Block 2, Subject 1 changed strategies for a single session and produced [sp] sequences for each /f/-initial word. Whether this production was influenced by the syllabic goals Subject 2 was targeting at the time, or whether Subject 1 had decided that [+continuant] could only be established with a 'coronal' place of articulation followed by a stopped 'labial', is unknown. Interestingly, Subject 2 produced the same [sp] for all /f/-initial words on the mini probe administered two days after Subject 1 had attempted this strategy. Based on the finding that the production of If/ in word-final and medial positions was easier to train than in word-initial, we would have to conclude that establishing the more embedded 'labial' place node in the presence of established [+continuant] is easier than developing [+continuant] itself. This finding appears to counter the predictions made by the feature geometry. Bernhardt (1990) discusses similar findings with her subjects and speculates that the [+continuant] feature must be learned separately for each of the PLACE nodes and the LARYNGEAL node. Continuancy, therefore, is not acquired "across the board," but must be established individually for labial, coronal and dorsal segments. The data here supports the claim that continuancy is most easily established with a dorsal consonant place of articulation. It can be concluded, based on the findings that Ihl was acquired before Ifl, that the feature geometry—as described in chapter 2—is supported by clinical data. As predicted, the segment whose salient feature occupied a higher position in the feature geometry was acquired before the segment whose salient feature filled a more embedded position. Training the production of the targets /J/ and /dg/ and comparing acquisition results with the results of intervention targeting Ifl and /v/ allowed the

53 development of the 'labial' place node to be compared with the development of the terminal feature [+anterior]. The feature geometry predicts that 'labial' will be acquired first. Recall that this comparison could only be made for syllable-initialwithin-word and word-final data because the same feature, [-fcontinuant], was being established when training all these targets in word-initial position. The results for Subject 1 clearly support the predictions of the feature geometry. At mini probe #6, he achieved 75% accuracy on /f/ productions and 58% accuracy on /v/ productions. Neither /J/ nor /dg/ had shown any generalization in novel single-word targets or in spontaneous conversation at this time. It should be noted, however, that the affricate, /tj/, whose acquisition was expected to generalize from that of HI and /dg/, was produced with 60% accuracy for single-word stimuli only (three correct productions out of a possible five). Conversation level for /tj/ could not be tested at this mini probe because of the abundance of targets which needed to be assessed. Figure 18 illustrates that the described findings are supported by Subject 2's data. At both mini probe #5 and #6 Subject 2's percentage accuracy for producing /f/ exceeded that of /JV and /dg/. The fricative /v/ was never targeted with this subject; however, at the final grand probe following Block 3 of therapy, Subject 2 was producing /v/ with approximately 50% accuracy in all word positions. Based on the acquisition of labio-dental fricatives as compared with alveo-palatal fricatives and affricates in this study, the predictions of the feature geometry are borne out. For both subjects the 'labial' place node was established before the more deeply embedded [anterior] feature.

Conclusions Returning to the three questions outlined at the beginning of this chapter, it can be concluded that the results of the segmental condition of this study provide evidence for choosing targets for remediation based on the nonlinear feature geometry. The

54

feature [consonantal] was acquired before the more embedded [+continuant] feature. Since /f/ was more easily acquired in word-final position, one could conclude that the 'labial' node is easier to acquire than [+continuant]. Such a conclusion would not support the feature geometry because the 'labial' place node is more embedded than [+continuant].

The hypothesis was presented, however, that continuancy .must be

separately established for each place of articulation. Comparing the acquisition of manner nodes with place of articulation nodes, then, is not in accordance with this hypothesis. It can be concluded however, that the acquisition of the 'labial' place node for HI occurred before [+anterior] of the alveo-palatal segments was developed. This conclusion provides evidence in favor of choosing targets based on the feature hierarchy. The second question addressed the issue of generalization of a targeted segment to other segments with similar minimal feature specifications.

The data

provides some evidence that generalization occurred. As described above, targeting

III and /dg/ generalized to the emergence of AJ/ for both subjects. Training the production of /f/ and /v/ with Subject 1 did not lead to the acquisition of any other segments (/s/ in initial position would have been a possibility). However, /v/ was beginning to emerge at the end of the study for Subject 2, who had only received training on HI. Generalization, then, was only evident in voiced or voiceless cognates of trained targets; acquisition of targets beyond this immediate sound class did not occur (within the time frame of this project). The third question of investigation focussed on the possibility of Subject 2, who was placed on the syllabic condition, spontaneously acquiring the segmental goals of Subject 1. Interesting results were found in this area. Subject 2 acquired both /h/ and

HI without any direct training. Surprisingly, this acquisition occurred before Subject 1 had acquired either of these targets! As illustrated in figure 17, Subject 1 made slow but steady progress on establishing Ihl until mini probe #5, when he reached 100%

55 accuracy at the conversational level. Subject 2 achieved 64% accuracy for the production of /h/ at mini probe #3 when Subject 1 achieved only 38% accuracy. Similar results were obtained for /f/. The day before mini probe #5 Subject 2 began producing HI in spontaneous conversation; Subject l's request to be first did not occur until one week later. These results are reflected in the data of mini probes #5 and #6. Whether Subject 2's acquisition of HI occurred through generalization from his /s/-cluster syllabic targets, or whether HI was acquired because Subject 1 was attempting this target, cannot be conclusively determined from the data here. In summary, then, the data described in this chapter provides evidence supporting feature hierarchies, some generalization of trained targets to related segments occurred, and spontaneous acquisition of segmental targets by the twin placed on the syllabic condition occurred.

56

CHAPTER 5 THE SYLLABIC CONDITION

Initial assessments of phonologically delayed children often describe deficits that affect more than a single segment. Clinicians using a phonological process analysis often include terms such as 'cluster reduction', 'weak syllable deletion', and 'reduplication' to describe these types of phonological errors. In addition, phonological process descriptions such as 'final consonant deletion', 'assimilation' and 'stopping of fricatives in syllable- initial position' capture difficulties with the production of single segments in certain word positions. Both of these kinds of processes refer to difficulties affecting phonological development at not the segmental but the syllabic level of representation. Recall from chapter 2 that the basic syllable is the CV unit. The universality of this syllable shape is supported by both cross-linguistic data, and the predominance of CV syllables in the earliest utterances of typical language learners and in the utterances of phonologically disordered children. In order for development to proceed beyond this "default" syllable shape, the language learner is required to mark those more complex syllable shapes evident in her linguistic environment. Prior to developing these advanced syllable units, the child must fit complex lexical items into her more basic existing syllable inventory. This stage of development is typically described using the processes and rules cited above. If, for example, a child attempts saying the word truck but only has CV and CVC syllable shapes in her inventory, then truck will surface as [L\k] assuming no other phonological difficulties affect the production of this target. A process analysis of such an error would describe the error

57

as 'cluster reduction'. The view supported in this paper, however, holds that the language learner has not "reduced" anything, she simply has not yet developed the branching required at the syllabic level to correctly realize the adult target. The second kind of syllabic process typically referred to in phonological process assessments accounts for the acquisition, or lack thereof, of segments or groups of segments in certain word positions. The subjects of this study, for example, realized HI in word-final position as [s], but in word-initial position HI surfaced as [p]. In actual fact, at the outset of the project, the twins had no fricatives in their word-initial inventory and a phonological process such as 'stopping of word-initial fricatives' would have adequately described their production. A better explanation (as opposed to the process description) holds that the [+continuant] node had not yet been established in -

their word-initial inventory. Phonological difficulties such as these provide evidence for the interaction of the segmental and syllabic tiers during phonological acquisition. As with intervention under the segmental condition, the syllabic condition of this project led to the investigation of three questions. Recall from chapter 2 that two main theories are currently being developed in the field of linguistics to account for the organization of phonemes into syllables. The primary question in the syllabic condition investigated whether training following the moraic approach or the onsetrime approach would lead to the faster acquisition of therapy targets. This faster acquisition could reflect a "match" between the theoretical hierarchy and the child's mental representation of the syllable's organization. As with the segmental condition, this central inquiry led to two further questions. The second question, then, addressed what spontaneous generalization to other syllabic goals would occur, and if such generalization could provide independent support for either moraic or onset-rime theory. And finally, the third question focussed on whether Subject 1, placed on the segmental condition, would acquire any of the syllabic goals of Subject 2 without direct intervention. Depending on which targets were acquired, such spontaneous

58

development could provide support for either the moraic or the onset-rime syllable theories. Before discussing the results of intervention within the syllabic condition, the twins' syllabic inventories at the outset of the current study will be described. The Starting Point It was hypothesized in chapter 2 that the syllabic level of phonological development could be more basic—that is, less embedded—than the segmental level. In agreement with this hypothesis, phonologically disordered children with restricted syllabic inventories are usually considered to be more severely disordered than those showing deficits at only the segmental level. The twin subjects of this study were not severely disordered and, as expected, they did not have severely restricted syllabic inventories. At the time of the initial assessment, (please see the grand probe #1 data presented in Appendix 3) the CVC syllable was firmly established and it was found that the twins had no difficulty stringing these CVC units together to match mono-, di-, and polysyllabic adult targets. In syllable- and word-final positions both twins were able to produce clusters of two and three consonants in words such as ajlc and ants. Syllabic deficits were evident, however, in word- and syllable-initial positions where the twins' productions only allowed, at most, a single consonant to surface. At the time of the first grand probe, sequences of a consonant followed by either /!/, /r/ or Av/ (as in plum, truck, or quick) predictably surfaced as a singleton obstruent consonant for both of the twins. Producing truck as [tAk], as described above, was typical for the twins. Consonant clusters with /r/, IM and /w/ in which the initial consonant was a fricative, were realized as a stop consonant, since continuancy was not yet established for the twins in word-initial position. Hence, throw was pronounced [po] and slide as [taid]. Clusters with /s/ followed by a non-liquid or glide consonant surfaced without the /s/. For example, snake was produced as [neik], spot

59 as [pat], stop as [tap] and skinny as [kmi]. The /sw/ cluster provided the only noted difference between the phonological inventories of the two twins. Subject 2 treated /sw/ as a /Cw/ cluster rather than a /sC/ cluster. As such, the /w/ was not realized and the /s/ surfaced as the noncontinuant [t]. Subject 1, however, grouped /sw/ with the /sC/ clusters, which resulted in the /s/ not being realized and the /w/ surfacing. The grand probe word sweater was pronounced [tedr] by Subject 2 but [wedr] by Subject 1. This difference in syllabic constraints led to interesting differences in the twins' acquisition of the /sw/ cluster, which will be further discussed below. Similar to /sw/, the cluster /si/ was not treated as an /sC/ combination. If the twins had grouped /si/ with the other /s/-clusters, one would expect [j] to surface, since HI was pronounced [j] and /s/ was not produced for these clusters. The fact that [t] was the surface realization of /si/ indicates that this cluster was grouped with the /CI/ clusters where HI was not realized and only the consonant [C] surfaces (in this case [t] for the target /s/). The second type of syllabic restriction—where positional constraints are evident in the production of segments or groups of segments-were also evident in the phonological inventories of the twin subjects of this study. As already mentioned, the [+continuant] node was only established in syllable-final position. The only productive fricative in this position was /s/. It was difficult to determine whether the voiced cognate /z/ was productive in the twins' inventories because of the natural occurrence of word-final devoicing. Based on auditory judgment, vowels and consonants preceding a targeted word final /z/ were sufficiently lengthened to mark a contrast between /s/ and Izl. In syllable-initial position all fricatives surfaced as stop consonants. Production of the liquids HI and hi reflected syllabic constraints since they were established only in word-final position at the outset of this project. The segment /r/ was firmly established in word-final position but surfaced as [w] at the beginning of a

60 word. The correct production of /r/ in syllable-initial position within a word supports the ambisyllabic nature of liquids within words. At the time of the initial assessment, HI was emerging in word-final position but surfaced as [j] in both word- and syllableinitial position.

Determining Targets for Remediation Determining therapy goals within the syllabic condition of this project was a relatively straightforward task given the limited number of syllabic deficits evident in the subjects' phonologies. The choice of targets was limited to syllable- and wordinitial consonant clusters and certain segments showing positional constraints. A further restriction on choosing goals for remediation within the syllabic condition required target clusters and segments to be made up of segments established in the twins' inventory. Since /s/, /r/ and HI were all at least emerging in word-final position, any of the /sC/, /Cr/ or /Cl/ clusters presented options for therapy. In addition, all segments (except for /s/ and /9/) that cluster with /w/ were well established. Hence /Cw/ clusters were also a possibility for therapy. Of the options, /sC/ and /Cw/ clusters were chosen as intervention targets for this study. Four reasons supported the choice of /sC/ clusters for remediation. First, the abundance of words with initial /s/-clusters in the English language meant that the twins' communicative success was greatly diminished by not being able to produce these clusters. This effect appeared to be greater in the case of /s/-clusters than with either /CV or /Cr/ combinations. Second, /s/-clusters tend to emerge earlier than HI- or /r/-clusters in the inventories of typical children, which suggests that they may be easier to acquire. Third, during the period of initial assessment it was found that the twins were more stimulable for the correct production of clusters with I si than those with HI or /r/. Finally, the possibility existed that establishing /s/,-clusters in the twins' productive inventories would generalize to the acquisition of other segments—

61 specifically, that establishing the [+continuant] feature for /s/ in initial clusters would enhance the emergence of the class of singleton fricatives. Of the seven /sC/ combinations, four were chosen for direct intervention and three remained untreated to see if spontaneous acquisition would occur. The clusters /sn/ and /sp/ were targeted in therapy using the moraic approach, while /st/ and /sk/ followed the onset-rime approach. The details of each training approach will be described in the next section. In order to keep the two approaches approximately equal in difficulty, clusters with similar features were assigned to opposite conditions. In the case of /s/-clusters, /sn/ and /st/ are similar because /s/ is paired with a coronal segment in each combination. These two clusters are unequal, however, because /s/ is part of the syllable in the case of /sn/ but is considered to be 'extrametrical' in the case of /st/. An explanation of the notion 'extrametricar is required. Saussure (1915) was the first to notice the general tendency for syllable constituents to be arranged according to their sonority. Kiparsky (1979) claims that the least sonorous elements of a syllable fall at the margins while the most sonorous make up the syllable centre or peak. The sonority hierarchy from least sonorous to most sonorous, as defined by , Kiparsky (1979), takes the form: stops~>fricatives—>nasals—>liquids—> glides—>vowels. When the arrangement of segments does not follow this hierarchy, violating segments are termed extrametrical, provided they fall at the margin of the syllable (Hayes, 1982). Designating certain elements as extrametrical redefines the 9

edge of the syllable, allowing the application of phonological rules to "overlook" edge segments. Rules requiring reference to extrametrical constituents are widely

It should be noted that extrametricality is not restricted to single segments but may refer to any single edge constituent including segments, morae, syllables and feet (McCarthy & Prince, 1988). y

62 documented cross-linguistically (e.g. Kiparsky, 1979; Cairns & Feinstein, 1982; Hayes, 1982; Harris, 1983). Returning to /s/-clusters, it can be seen that when compared to the sonority hierarchy, /s/ followed by the stop consonants /p/, /t/, or fkf would be considered as extrametrical. The remaining /s/-clusters, /sm/, /sn/, /si/, and /sw/, follow the sonority hierarchy by having the least sonorous element at the syllable edge; /s/, therefore, is not designated as being extrametrical. In addition to determining whether the moraic or the onset-rime approach in therapy would lead to the faster acquisition of targets, the training of /s/-clusters would determine if extrametricality affected their acquisition. Since the sonority hierarchy is considered to make up part of universal grammar, the theory predicts that /sn/ would be easier to acquire than /sp/, /st/ or /sk/, where /s/ is extrametrical. In this sense, the moraic condition, which included /sn/ and /sp/, could be considered easier than the onset-rime condition, which included /st/ and /sk/. In addition to /s/-clusters, /Cw/-clusters were also targeted in the syllabic condition. All the consonants that combine with /w/ were well established in the twins' inventories, with the exceptions of /sw/ and /6w/ (the latter cluster does not occur in words commonly used by 5 year-olds!). The targets /kw/ and /tw/ were chosen for direct intervention while /sw/ served as a baseline for generalization (the acquisition of /dw/ was not monitored since the only word likely to be part of a kindergarten lexicon in which it occurs is dwarf). The cluster /tw/ was assigned to the moraic condition while /kw/ was targeted under the onset-rime condition. After three cycles, direct therapy of /s/-clusters was discontinued and the segments /z/ and HI, along with /tw/ and /kw/, were targeted. Recall that /z/ and HI were established in word-final position only and, therefore, reflect positional constraints. Differentiating the training of these singleton targets according to the moraic or the onset-rime approach proved to be more difficult than for cluster targets.

63

Details of these training differences, however, will be described in the following section. As previously mentioned, targeting singleton consonants within the syllabic approach reflects the interaction between the syllable and segmental tiers of the phonological hierarchy. The training and acquisition of these segments allows not only the two syllable approaches to be compared, but also acquisition results to be analyzed based on the feature content of the targets. Both /z/ and HI are characterized by two features appended to the root node in their specifications and, therefore, would appear to be equivalent in difficulty. In actual fact, the [consonantal] [sonorant] combination of /l/ is more marked (fewer segments require this specification) than the [consonantal] [+continuant] combination of /z/. This marked status of HI is borne out when cross-linguistic data is analyzed. Based on features, then, it was predicted that M would be acquired before HI. The /sC/- and /Cw/-clusters made up the goals for the first three cycles (or first one-and-one-half blocks) of the intervention project, while Izl, HI and /Cw/-clusters were targeted in the fourth cycle. The individual session objectives for Subject 2, who was assigned to the syllabic condition, are outlined in Appendix 4. Treatment Before discussing the results of treatment within the syllabic condition, therapy following the moraic approach will be contrasted with therapy following the onset-rime approach. As outlined in Chapter 2, the moraic theory of syllable structure holds that onset cluster constituents are appended individually to the syllable node. Consequently, the cluster was treated as two units and minimal pairs such as napsnap and pot-spot served as the basis of training. Treatment under this condition allowed the clinician to emphasize the initial /s/ by lengthening it considerably when modelling target forms. Subject 2 responded well when instructed to "put the magic S on." Similarly, training of /tw/, which also followed a moraic approach, contrasted

64

pairs such as wig with twig. Subject 2 could easily produce the /w/-initial words of the minimal pairs and, when instructed to produce A/ before these, he was easily able to pronounce the consonant cluster. The onset-rime theory of syllable organization holds that constituents of an onset cluster form a unit in opposition to the rest of the syllable (the rime). In this case /s/ and the following obstruent could not be separated from each other in therapy and needed to be contrasted with the remainder of the word. The /sk/-cluster in the word skate, for example, was presented in two ways: either ale. was contrasted with skate or the /sk/ unit was repeated several times before saying the target word (e.g. [sk-sk-sk-skeit]). For some words production was facilitated by moving a word-final /sC/-cluster to wOrd-initial position. For example, saying ask followed by al several times simulates /sk/ in word-initial position and eventually sounds like a scat. After . several repetitions of a scat Subject 2 was encouraged to drop the a and produce scat on its own. The teaching of /kw/ followed similar principles. In this case /kw/ was pronounced [ku] followed by the remainder of the word (e.g. quick was presented as [ku + ik]). The discussion under Results provides further rationale for this teaching approach for the /kw/ cluster. The presentation of the singleton consonant targets /z/ and IM according to moraic or onset-rime approaches, differed only at the initial stimulation or elicitation stages. Both theories allow the onset target consonant to be taught in isolation and opposed to the remainder of the syllable, the rime in the case of onset-rime theory or the mora or weight unit in moraic theory. In order to maximize the differences between approaches, however, the targeting of /z/ in initial position, which followed the moraic approach, did not allow the consonant to be separated from the following weight unit. Following this approach, /z/ was initially elicited in CV syllable units before moving on to CVC words (e.g. zip., zap, zoom, etc.). Eliciting the production of IM in initial position followed an onset-rime approach. Recall that IM in word-final

65 position was emerging at the outset of the current project. By the third cycle of therapy, its production in this position was well established. As a consequence, training /l/ in word-initial position involved moving a word-final A7 first to an ambisyllabic position and then to a syllable onset as in pill-->pillow->low. A basic premise of the onset-rime theory is that syllable onsets should be maximized or created wherever possible, and it is this principle which is being emphasized when requiring Subject 2 to move an established coda consonant to a syllable onset position (Hyman, 1985). The results of these teaching methods for /z/ and /V will be discussed in the next section.

Results Target

Presentation Session

Review Session

Review Session

Combined Results

/sn/

78.0

63.6

22.2

63.8

/sp/

60.7

34.5

43.3

46.0

Aw/

64.2

10.0

38.5

52.6

Table 1: Percentage accuracy achieved by Subject 2 for clusters targeted in therapy based on the moraic theory of syllable organization.

Target

Presentation Session 1

Presentation Session 2

Review Session

Combined Results

/st/

0.0

4.4

6.7

4.4

/sk/

0.0

2.2

0.0

1.9

/kw/

95.2

33.0

45.7

57.8

Table 2: Percentage accuracy achieved by Subject 2 for clusters targeted in therapy based on onset-rime theory.

66 Tables 1 and 2 present the combined imitation and spontaneous data of the first cycle of therapy for Subject 2. The percentages achieved during the initial cycle of therapy most accurately reflect the effectiveness of the training or stimulation approaches used in therapy. In general, Subject 2 was able to produce all of the targets quite accurately during subsequent cycles. He needed, however, to generalize this accuracy to spontaneous conversation. Cycles 2 and 3, therefore, focused on providing opportunities for Subject 2 to produce targets during relatively natural conversation and the differences in therapy which were evident during the initial training of targets were no longer present. More than one percentage is provided for each target cluster because a review of moraic goals took place on the fourth and eighth sessions of the cycle. Onset-rime goals, however, were targeted over two sessions and reviewed on the eighth session of the cycle. Appendix 4 presents a schematic representation of individual session objectives. Comparing the percentage accuracy for the initial attempts at producing /s/-clusters, Subject 2 was far more successful in therapy when the moraic approach I

was followed. Three main reasons present themselves as explanations. First, it is possible, as suggested earlier, that the clusters taught using the moraic approach were actually easier to produce than those taught using the onset-rime approach. Since /s/ in /sn/ is not extrametrical and the cluster instead follows the sonority hierarchy, its production may be easier than either /st/ or /sk/. Bernhardt (1990) found that all of her singleton subjects had difficulty acquiring the /st/ cluster. She speculates that this is because the segments of the /st/ cluster are homorganic and because /t/—Subject 2's surface realization of this cluster—is the unspecified (in terms of features) default segment in English (please see chapters 2 and 4 for definitions of underspecification). In essence, the difficulty lies, first, in allowing something other than /t/ to surface and, second, in establishing [+continuant] in word-initial position. In other words, realizing a target segment or cluster with the default coronal is more

67

difficult to correct than if some other segment is substituted. Comparing the learnability of /sp/ with /sk/, it is possible that /sk/ is a more difficult target because it requires two places of articulation on the single tongue articulator, while /sp/ is produced using two separate articulators. A second possible explanation of the higher percentages acquired while intervention followed the moraic theory is that the moraic approach is simply easier to apply in therapy than the onset-rime approach. During the therapy sessions, the clinician sensed that the presentation and modelling allowed by the onset-rime approach did not reflect the normal usage of target words, but simply provided an ad hoc method of eliciting the target cluster. Subject 2 did not appear to transfer this stimulation technique to the production of target words in subsequent therapy tasks. In addition, the onset-rime approach did not allow the correct pronunciation of /s/clusters to be contrasted with Subject 2's usual production of /st/ and /sk/ words. The moraic approach, for example, allowed the clinician to elicit a correction by asking: "Did you say go! or spot?" The moraic approach also allowed the clinician to emphasize the /s/ when modelling /s/-clusters by lengthening it and by increasing the audibility of the frication. Since /s/ could only be produced in combination with the following consonant when teaching clusters using the onset-rime approach, the same elicitation and modelling techniques could not be used. Unfortunately, the current project only attempted targeting clusters using the onset-rime approach with one subject. Had more subjects been used, it is possible that more innovative and facilitating techniques following this approach would have been discovered. Bernhardt (1990) presents a comparison of the moraic and onset-rime approaches using six subjects and concludes that, based on quantitative data, there is no significant advantage in using one approach over the other. She does, however, report qualitative differences between using the two approaches for certain subjects.

68

A third explanation for the advantage observed when teaching was based on the moraic approach reflects the distribution of segments over therapy sessions. Appendix 4 outlines the cluster objectives for individual therapy sessions. Each cycle consisted of nine treatment sessions. Of these, three-and-one-half sessions targeted clusters following the moraic theory and three-and-one-half session followed the onset-rime theory. The eighth session of the cycle reviewed all the targets following designated training approaches, and the ninth session was a mini probe which assessed the acquisition of targets in novel stimuli. As shown in Appendix 4, this distribution meant that the goals of the onset-rime approach were targeted in two half sessions. However, Subject 2 had great difficulty changing goals in the middle of a session. This distribution, therefore, may have reduced the percentage accuracy figures and unfairly placed the onset-rime condition at a disadvantage where the acquisition of s-clusters is concerned. As noted earlier, if more subjects had been trained on the same syllabic goals, it would have been possible for cluster targets to be assigned to the opposite condition, and more definite conclusions regarding the advantage of one treatment approach as opposed to the other could be drawn. The data for /Cw/ clusters from the first cycle of therapy reflects slightly higher percentages when the elicitation of targets followed the onset-rime approach, especially when the individual session data, as opposed to the combined data, is compared. Therapy in the onset-rime approach presents /Cw/ clusters using an initial /Cu/ syllable. Such a technique encourages the use of epenthesis, a facilitating step that many typical children use when acquiring these clusters. According to onset-rime theory, maximizing CV or onset-rime units is desirable (recall the default status of the CV syllable); hence, separating a cluster with an epenthetic shwa to create a new onset-rime unit is a logical operation for a language learner following the principles of this theory. Adding a weight unit, however, as would be required to account for epenthesis using a moraic organization for the syllable, is not a logical explanation

69

Presentation Session 1

Presentation Session 2

Review Session

Combined Results

/z/(m)

28.1

62.1

41.7

43.5

HI (o-r)

34.6

30.6

38.9

33.0

Target

Table 3: Percentage accuracy figures for the initial cycle of targeting /z/ (moraic approach) and HI (onset-rime approach). within moraic theory. In summary, then, by training the use of epenthesis, the accurate production of /kw/ was facilitated. Table 3 provides percentage accuracy figures for the initial cycle of therapy targeting the singleton segments /z/ and /l/. Recall that IzJ was targeted using a moraic approach while the production of HI was trained following the onset-rime approach in therapy. Based on the percentages presented above, it would appear that using a moraic approach in initial elicitation stages is more facilitating than using an onset-rime approach. Although this may be correct, there is insufficient data in the current study to draw this conclusion. Further research that compares the two training approaches over several segment pairs is required to verify this finding. It is the impression of this author that the two syllabic theories simply allowed the target structures to be elicited in different ways and that their differences were not significant to the acquisition of the segments. Instead, the differences in the percentages above may be better explained by referring to the feature content. From this point of view, the earlier prediction that /z/ would be more easily acquired than HI is borne out. Figure 19 illustrates Subject 2's acquisition of syllabic goals based on mini probe results. It can be seen that /z/ was established after two cycles of therapy, while HI was still emerging after three cycles of training. Whether the training approach used with IzJ, or its feature content, or a combination of the two accounts for these results can not be concluded based on the single subject data presented here.

Figure 20: Graph of the /s/-cluster data based on grand probe results for Subject 2.

72 The acquisition data (as opposed to the initial elicitation or training data discussed above) presented in Figure 19 does not provide any further evidence that would favor using one syllabic approach in therapy over the other. After three cycles of therapy, all the /s/-clusters were beginning to emerge, including the untrained targets /sm/, /sw/ and /si/, although the latter surfaced as [st]. It should be noted that, the 11% and 20% correct for /st/ and /sk/ achieved at the second mini probe are based on only nine attempts at /st/ and four attempts at /sk/. Because of the small number of tokens, it cannot be concluded that these clusters were emerging at this time. Figure 20 illustrates the acquisition of the group of /s/-clusters as demonstrated in the grand probes, which occurred after every two cycles of therapy (every six weeks). The grand probe results accurately reflect the spontaneous usage of therapy targets because they were taken after a one week break from therapy (which allowed any recency effects from therapy to be removed) and because the grand probe sessions bore no relation to the usual structure of therapy. The elicitation method used was natural and appeared to the twins to amount to a free play session. By the fourth grand probe, /s/-clusters were emerging steadily, and the number of these targets that surfaced as a singleton obstruent was decreasing. The single word data presented in Appendix 3 shows that the first cluster to emerge for Subject 2 was /sn/, which was accurately produced at the time of the second grand probe. All other /s/-clusters were emerging by the third grand probe, and their accuracy increased by the fourth grand probe. The non-extrametrical status of /s/ in /sn/ clusters, together with the fact that /sn/ was one of the four /sC/ combinations targeted in therapy, probably accounts for the earlier emergence of this cluster. Based on the /tw/ and /kw/ data at the time of the fourth mini probe, presented in figure 19, one would think that the conclusion reached earlier—that the onset-rime training approach used with /kw/ facilitated acquisition because it emphasized the use of epenthesis—is inaccurate. However, it should be noted that any attempt at

—

reduced epenthesis correct

1

2

3

4

grand probe

Figure 21: Graph of the /Cr/-cluster data based on grand probe results for Subject 2.

120

—

reduced epenthesis correct

1

2

3

4

grand probe

Figure 22: Graph of the /Cl/-cluster data based on grand probe results for Subject 2.

75

producing /tw/ which realized two consonants was counted as being correct. At the time of the fourth mini probe, every attempt at /tw/ was produced with epenthesis and was judged as correct. As a result, the /tw/ data at mini probe #4 in figure 19 actually provides support for using the onset-rime approach in therapy. Figures 21 and 22 illustrate the grand probe data of Subject 2 for /Cr/ and /CI/ clusters. Before beginning the intervention project, both subjects were producing only the initial consonant of these clusters (e.g. dress was pronounced [des]). At the time of the fourth grand probe, however, these clusters were being produced as /Cw/ combinations since neither /r/ nor /l/ were firmly established in initial or cluster position. The acquisition of clusters with /I/ and /r/, then, was doubly encouraged b y targeting /Cw/-clusters. First, the use of epenthesis, as was trained when producing /Cw/-clusters, generalized to the production of /CI/- and /Cr/-clusters (e.g. blue was pronounced [bslu] and

truck

as [ t u w A k ] ) . The percentage of tokens produced with

epenthesis b y Subject 2 for these clusters is presented in figures 21 and 22. Second, targeting

/Cw/ encouraged the [w] to surface for /r/ and

DJ

(see chapter 4 for an

explanation of the feature nodes involved in this production). This s u b s t i t u t i o n is commonly f o u n d in children with developing phonologies and, therefore, allowed the listener to better "guess" at what the twins were saying. To this point no mention has been made of any acquisition of syllabic goals b y Subject 1. All of the results outlined above are based on the results of therapy with Subject 2, who was assigned to the syllabic condition. Recall from chapter 4 that Subject 2 had spontaneously acquired all the segmental goals of Subject 1. The reverse, however, d i d not occur. By the end of the second block of therapy (four cycles), Subject 1 had only demonstrated four correct productions of /s/-clusters (three correct productions of /sn/ at grand probe #3). As a result, the decision was made to target /sC/- and /Cw/-clusters during the third block of therapy. The specific clusters were /sp/, /st/, /sk/, /sw/ and /kw/ (see Appendix 4 for session objectives).

76 The cluster /sn/ was not targeted because there was evidence of spontaneous acquisition. Figure 23 illustrates Subject l's acquisition of /sC/- and /Cw/-clusters based on the mini probes administered in the middle and at the end of the third block of therapy. The data at mini probe #5 indicates that of the /s/-clusters targeted in therapy, /sw/ shows the highest percentage accuracy. It should be noted that /s/ in /sw/, as in /sn/, is not extrametrical. Although Subject 2 had no difficulty acquiring /sw/, based on mini probe results this cluster did not appear to be "easier" than the other /s/-clusters. For Subject 1, however, /sw/ did appear to be an easy cluster to learn. This ease of acquisition may have resulted from Subject l's grouping of /sw/ with the /Cw/clusters. This organization meant that therapy focusing on both /s/-clusters and on /Cw/ clusters was facilitating the acquisition of /sw/. In contrast with /sw/, the clusters /st/ and /si/ appeared to the most difficult for Subject 1 to acquire. These findings are in agreement with the acquisition results of Subject 2 and the same explanation—that it is difficult for the language learner to realize something other than the underspecified default /t/—holds here as well. The results from mini probe #6 show that /kw/ was beginning to emerge in Subject l's speech at this time. The data from grand probe #4, administered one week after the sixth mini probe, showed no generalization to novel /kw/ words. However, at this time 12.5% of the /Cr/-cluster words surfaced with two adjacent consonants (with /r/ surfacing as [w]) while 25% of these targets were produced using epenthesis. As with Subject 2, the teaching of /Cw/-clusters facilitated the acquisition of /Cr/-clusters. The same effect had not spread to /Cl/-clusters by the time of the fourth grand probe.

Figure 23: Graph of the syllabic targets based on mini probe results for Subject 1.

78

Conclusions The principle question in the syllabic condition of this project investigated whether therapy based on the moraic theory or the onset-rime theory of syllable organization would result in the faster acquisition of targets. The data presented in this chapter provides support for both theories depending on the phonological goals of therapy. In the case of singleton consonants, the limited data of this study could not provide evidence favoring one theory over the other. In the case of consonant clusters, however, the remediation of /s/-clusters favored using a moraic approach, which allowed therapy to emphasize the two consonants as separate phonemes. In contrast, the intervention targeting /Cw/-clusters provided support for following an onset-rime approach in therapy. In this study, encouraging the use of epenthesis by presenting /Cw/-clusters as a /Cu/ syllable appended to the remainder of the word was a facilitating strategy for both subjects. Unfortunately, finding support for each of the theories being investigated does not provide independent evidence for either an onsetrime or a moraic syllable hierarchy. From a clinical point of view, however, what facilitated the acquisition of cluster goals was not following one or the other of the syllabic theories. Instead, the facilitating strategy involved emphasizing that two phonemes make up the cluster. This statement may appear to provide support for the moraic theory, where separating the constituents of a cluster is acceptable because cluster consonants are appended individually to the syllable node. However, when a /Cw/-cluster is pronounced as /Cu/, a /w/ consonant occurs spontaneously as the speaker makes the transition from /u/ to the following vowel (e.g. /ku/ + /ik/ is produced as [kuwik]). Essentially, separating the cluster consonants by an epenthetic vowel emphasizes the two constituents present and allows them to be more easily produced. What appears to facilitate the acquisition of consonant clusters, then, is emphasizing that two phonemes make up the cluster unit. The

79 difference between the two theories investigated here is that the moraic theory is better able to explain separating the cluster constituents in the case of /s/-clusters, while onset-rime theory is better able to account for the use of epenthesis and, hence, the separating of constituents in the acquisition of /Cw/- clusters. A secondary question which resulted from the choice of /s/-clusters as intervention goals in this study involved the notion of extrametricality. Recall the earlier prediction that clusters where the initial /s/ was more sonorous than the following consonant and, hence, was not extrametrical, were expected to be acquired before clusters where the initial /s/ was extrametrical. The fact that both subjects acquired /sn/ clusters first (where /s/ is not extrametrical), and that Subject 1 acquired /sw/ before other /s/-clusters, provides support for this prediction. It should be noted that Bernhardt (1990) reports that /sn/ was the easiest of the /s/-clusters for her subjects to learn. Similar findings with other subjects and for other non-extrametrical clusters (e.g. /sw/, /sm/ and /si/) are required before it can be concluded that these combinations are actually easier to learn. Before beginning therapy within the syllabic condition, it was asked whether the acquisition of certain syllabic goals would generalize to other syllabic targets. Results indicate that training the production of the clusters /sn/, /sp/, /st/ and /sk/ generalized to the acquisition of /sm/, /sw/ and /si/. Similarly, therapy focusing on /Cw/-clusters generalized to the acquisition of /Cr/- and /Cl/-clusters. Whether this generalization supports one or the other syllabic theory under investigation will not be concluded here based on the rationale outlined above. An interesting question raised here, is whether simply establishing the CCVC syllable shape (regardless of the cluster chosen) facilitates the acquisition of all consonant clusters. In other words, does the learning task involve the acquisition of two consecutive consonantal slots or does it involve the acquisition of specific feature content which happens to fill consecutive slots. The most conclusive data answering this question would be

80 provided by a child who has acquired all the phonemes of English in single segment positions but who has not yet acquired any consonant clusters. Intervention could then focus on one type of cluster (/s/-clusters for example) while all others served as baselines for spontaneous acquisition. The data provided by this study is inconclusive. It can be said, however, that at the outset of the project no clusters were evident in the twins' inventories but, by the end, all clusters were emerging (including three-consonant clusters, as in string'). Bernhardt (1990) found that targeting one type of consonant cluster with her singleton subjects influenced the acquisition of other consonant clusters. Unlike the results based on the segmental goals described in chapter 4, no spontaneous acquisition of syllabic goals by the twin assigned to the opposite condition occurred. Several possible explanations of this finding present themselves. These hypotheses, however, will not be discussed here but will be outlined in detail in the following chapter, which deals with using twins as the subjects of this study.

81

CHAPTER 6 TWINS AS R E S E A R C H SUBJECTS

To this point, the research results have only been discussed in terms of nonlinear phonology. The subjects' phonological deficits and progress in therapy have been described as established or unestablished branches of feature or syllable hierarchies. An entirely different, but equally interesting approach to the research findings analyzes the results from the point of view of using twins as the subjects of this project.

Twins: Handicapped or Not? The earliest twin studies compared the development of physical and cognitive development of twins to that of singletons.

These studies concluded that while motor

milestones were reached at similar ages, twins, especially monozygotic or identical twins, demonstrated slower development in the areas of language and social relationships than did their singleton peers (Day, 1932; Davis, 1937; Zazzo, 1977). Two complementary explanations account for this observed delay. First, it has been hypothesized that twins receive much more limited linguistic input than singleton children because the attention of the caregiver is necessarily divided (Lytton, Conway & Sauve, 1977). Each twin has only half the amount of language directed at him and, consequently, has only half the number of opportunities to respond verbally. While this hypothesis may be true to some extent, the twin situation itself can account for some of the linguistic delay observed. Matheny & Bruggemann (1972) hypothesize that the environment in which twins are reared presents a "unique psychosocial

82

condition." Since most, if not all, of their time is spent together, each member of the twin pair acts as a language model for the other. Because of the constant interaction between the members of the pair, twins do not use adult models as much as singletons do. The ease of communication within the partnership resulting from common experiences, then, may reduce the motivation to learn the adult language system. Several studies indicate that many of the delays exhibited by twins during the preschool years disappear at around 5 or 6 years of age (Davis, 1937; Wilson, 1974, 1975; Malmstrom & Silva, 1986). This finding is explained by the increased opportunity each twin member has to create his/her own experiences and friendships with age peers in preschool and kindergarten settings. Earlier research had found that I.Q. scores remained depressed after this age when compared with singleton scores (Zazzo, 1960). More recent and comprehensive research in this area has not been published. To date, no clinical data documenting twins' progress in speech or language therapy has been published. The current project replicated a study done by Bernhardt (1990) with singleton subjects and, therefore, general comparisons of therapy progress can be made. Bernhardt (1990) reports that her subjects were beginning to acquire syllabic and segmental goals after two cycles of therapy. However, it should be noted that Bernhardt (1990) alternated targeting segmental and syllabic goals with her subjects. This difference in methodology may have affected acquisition time.

As

outlined in the previous two chapters, the first acquisition of therapy targets by the twin subjects of this study did not occur until three cycles of treatment were completed. A possible explanation of this slower acquisition time (other than attributing it to different therapy targets) would claim that the twin situation, as

83

described above, continued to reinforce incorrect speech patterns because of the continued communicative success experienced between the twins.

10

Heredity versus Environment Although the earliest twin research compared twin development with that of singletons, the majority of twin studies were concerned with determining the relative contributions of environmental and genetic factors to development. Such studies typically compare data from identical (monozygotic) twins with that of fraternal (dizygotic) twins. Since monozygotic twins share the same genetic background and the same environment, those aspects of development under genetic control will develop at the same time. For dizygotic twins, who share only environmental factors and not a full set of genetic ones, development might follow quite different time lines. The general conclusion of these studies is that the language of monozygotic twins is much more similar than that of either dizygotic twins or singletons (Mittler, 1972; Munsinger & Douglas, 1976; Koch, 1977). This finding supports the view point that at least some of language acquisition is controlled by genetic make-up. The twins of the current study entered the project with almost identical phonologies, which supports the conclusion that phonological development has a genetic basis. The fact that these phonologies can also be influenced by the environment is demonstrated by the project results reported here as well as countless

Research by Sonnenschein & Whitehurst (1980) found that a child's communication improves when she is paired with a poor peer communicator. It is interesting to note that Waterman & Shatz (1982) quote Sonnenschein & Whitehurst (1980) as evidence that a twin's language may improve when paired with his language delayed co-twin. The application of these research finding to the twin situation, however, is incorrect, in that twins do not experience or even recognize communication breakdown as the subjects of the quoted research did. As a result, the same motivation to improve their communication does not exist. 1 0

84

reports in the disorders literature. Language acquisition, then, is not exclusively controlled by heredity.

Twin versus Co-twin

11

Relatively few researchers have followed the development of small numbers of twin pairs on a longitudinal basis in order to qualitatively compare one twin's development with that of the co-twin. In the area of language acquisition, one such study analyzed and compared the grammars of two sets of identical twins and found that one twin was in advance of the other in both cases (Bruggemann, 1970). A qualitative analysis of the productive structures revealed that the language of the more delayed twin formed a subset of the co-twin's syntactic inventory. To date only four studies have qualitatively described the phonologies of twins. The first of these, an unpublished manuscript by Edwards & Bernhardt (1973), details the phonologies of a set of opposite sex fraternal twins. Using a phonological process analysis, these authors conclude that the twins of their study shared an idiolect. In other words, the same processes were acting in each twin's phonology, but surface differences appeared because different limitations affected the application of those processes. In effect, the more advanced twin allowed a more restricted application of certain processes to be applied to the underlying representation, while the co-twin allowed a more general form to be applied. The advanced twin, for example, may have stopped fricatives in only word-initial position, while the co-twin stopped fricatives in all word positions. Edwards & Bernhardt's conclusions have been criticized by Ingram & Goad (1988) because the subjects of the study were fraternal twins who, according to other

Although it is understood that the term co-twin is semantically redundant, it will be used in this thesis as an abbreviation for the phrase "the other member of the twin pair." 1 1

85

research, are no more similar than singletons in their acquisition of language.

Since

other studies that qualitatively compare the phonologies of fraternal twins do not exist, it is not possible to disclaim the findings of Edwards & Bernhardt (1973). The existing fraternal twin studies have been large sample in nature and have only looked at quantitative differences in phonological acquisition. Thus, it is conceivable that such research has overlooked the possibility that one fraternal twin is in advance of the other. A more recent twin study by Leonard, Newhoff & Mesalam (1980) compared the phonologies of identical twin girls and found several qualitative differences between them. The authors concluded that the phonologies of their subjects were no more similar than those of singletons. Ingram & Goad's (1988) criticism of this study is based on the phone-tree method of analysis used, which they claim emphasizes phonological differences to the point of masking any similarities. It should be noted, however, that the findings of Leonard et al. (1980) are in agreement with the quantitative measures of Matheny & Bruggemann (1972) who found a high degree of phonological similarity for monozygotic male twins only. The phonologies of monozygotic female twins were no more similar than those of dizygotic twins or singletons. A third twin study, an unpublished work by Ingram & Goad (1988), compared the phonological and morpho-syntactic development of a set of monozygotic twin boys over a five-month period. Only the phonological data will be summarized here. Essentially, Ingram & Goad (1988) conclude that their subjects' phonologies were remarkably similar with the single exception of the application of a word-initial voicing rule. One twin voiced only initial /p/, while the co-twin voiced all initial stops. Despite the authors' criticisms, Edwards & Bernhardt's analysis—that one twin's phonology forms a subset of the co-twin's phonology-would accurately account for this data.

86

A very recent study by Locke & Mather (1989) analyzed the articulation abilities of monozygotic twins and compared the results with those of age- and sexmatched dizygotic twins. The authors concluded that monozygotic twins were significantly more likely than dizygotic twins to misproduce the same sounds on a standardized articulation test. However, based on a broad analysis of errors (which included only the categories substitution, omission and distortion), these authors found that monozygotic twins were no more likely than singletons or dizygotic twins to produce the same surface realization of the misproduced target. Locke & Mather (1989) admit that a more detailed phonetic analysis of these errors is required. It is possible that such an analysis will in fact reveal concordance between the phonological errors of twins and indicate that one twin's phonological abilities form a subset of the co-twin's phonology. No studies documenting the remediation of phonologically delayed twins, other than the present one, exist in either the twin or the speech-language pathology literature. As such, the current research allows not only the phonological inventories at the outset of the project to be analyzed, but also progress resulting from therapy may be compared. As previously mentioned, the twins' phonological abilities, both segmental and syllabic, were very similar at the beginning of the project. What is more interesting than the initial similarities, however, is the differences between the twins which surfaced once they began acquiring the targets of therapy. Combining the segmental and syllabic data based on the mini probe results presents a summary of the phonological acquisition results of the six-month therapy portion of the project. As is well illustrated by figures 17, 18, 19 and 23 , Subject 2 not only acquired the segmental goals of his twin's opposite condition without any direct intervention, he acquired all the goals of therapy before Subject 1. Subject 2 began generalizing treatment goals to novel stimuli some time after the second mini probe.

87

Except for the acquisition of /h/, Subject 1 did not show any acquisition of therapy targets until after mini probe #4, a full four months after his brother. The first possible explanation for why one twin should respond to therapy so well while the other twin showed a four month delay has to do with the targets of intervention. That is, it is possible that the syllabic goals of Subject 2 were simply easier to acquire and therefore generalized earlier than the segmental goals of Subject 1. This hypothesis seems especially plausible given that a requirement of the syllabic condition was that segments be established in some word position in the phonetic inventory. Thus, the segments themselves did not need to be learned; instead, only the use of these segments in unfamiliar syllable shapes needed to be acquired. At the end of the second cycle of therapy, it was found that, while generalization had not yet occurred, Subject 2 produced the syllabic goals regularly in therapy and only needed cueing during conversational activities. Subject 1, however, was having great difficulty producing the segmental targets. The target /h/ was often produced as a voiced laryngeal fricative with much tension visible in the throat and upper chest area, and /f/ was most often produced as a labio-dental stop with occasional friction preceding the stop. The hypothesis that the syllabic goals were actually easier than the segmental ones was confirmed when each twin was tested to see if he was stimulable for the production of the other twin's targets in the opposite condition. Subject 1 was easily able to produce /s/-clusters, while Subject 2 exhibited the same difficulties with the segmental goals that his brother demonstrated. These results are taken as evidence that Subject 2's syllabic goals were in fact easier targets to acquire. As previously mentioned, Bernhardt (1990) alternated targeting segmental and syllabic goals with her singleton subjects and found that syllabic goals were acquired faster than segmental ones by all of her subjects. This finding would

88

agree with the results presented here and implies that syllabic targets are in fact easier to acquire than segmental ones. The ease of acquisition hypothesis can only be accepted as a partial explanation of Subject 2's earlier acquisition of therapy targets. This same explanation cannot account for the fact that Subject 2 acquired Subject l's segmental goals without any direct intervention. Part of the explanation, therefore, must be attributed to the differences in the twins' personalities.

Twins and Psychology While most twin studies were primarily concerned with investigating the nature-nurture issue, Rene Zazzo (1960, 1974, 1977) was more interested in the differences exhibited by identical twins. He noted that in almost all monozygotic and dizygotic twin pairs there is a dominant or leading twin. Research by Canter (1973) indicated that identical twins reared in different environments were more similar on several personality indices than twins who shared the same environment. In other words, twins exhibit less similar personalities when they live together.

Observations

such as these led Zazzo (1977) to the conclusion that despite sharing identical genetic backgrounds and living in the same environment, identical twins do not share the same psychology. Zazzo (1977) attributes the personality differences of identical twins raised together to the twin situation itself. He describes the main problem of this situation as being the conflict between the attachment to the other twin and the selfaffirmation, between the resemblance that has been accepted and suffered during all childhood and the more or less sudden need for differentiation, personal autonomy, independence. (Zazzo, 1974: 347)

89

He concludes, therefore, that the twin situation leads each member of the pair to actively search out any possible social and personality differences within the genetic limitations set for them. Several personality differences were exhibited by the twin subjects of this study. The first difference refers to dominance which Zazzo (1977) claims may result from higher birthweight, better physical condition in infancy, or may simply result from being designated and subsequently treated as the older twin. Interestingly, the father of the twins in this study reported that Subject 1, who was the first born and the heavier twin at birth, was the dominant one until he was ill with a respiratory infection for an extended period of time at the age of two. Subject 2 became the dominant twin after this illness and was still obviously so at the time of this study. Zazzo (1977) reports that, in the twin pairs he observed, the dominant twin generally looked after dealing with "the outer world," while the co-twin concerned himself with the pair's relationship. He states that in some cases the same twin looked after both relationships. In Zazzo's (1977) terms, Subject 2 certainly looked after affairs with the external world and probably also those of the twin pair. It is this superior social awareness demonstrated by Subject 2 which is central in accounting for his greater success in therapy. Increased social contact with communicators other than his cotwin not only motivated Subject 2 to improve his phonology in order to avoid communication break-down, but also provided him with more frequently correct speech and language models. The difference in progress resulting from therapy can also be accounted for from the point of view of Subject 1. The language acquisition literature provides several such hypotheses. First, it is possible that Subject 1 is simply a very independent person. Such a personality would not seek out the companionship of others (beyond his co-twin) and would not be very concerned if his language were not understood. In this respect, an independent personality could lead to the slower acquisition of an

90

adult language system. Alternatively, the phonological acquisition literature describes some language learners as risk-takers (Macken & Ferguson, 1983). In these terms, it could be postulated that Subject 1 did not have a risk-taking personality and, therefore, avoided using phonological forms that were not well established in his inventory. A third possibility, described by Wilson, Brown & Matheny (1970) in their twin study, is that Subject 1 is less analytical in his approach to cognitive tasks. In this case, a longer period of therapy would be required so that a greater number of examples of the target form could be presented and attempted before the speech or language pattern was recognized and assimilated into Subject l's phonology. Which of the above three possibilities, or combination thereof, actually accounts for Subject l's slower progress cannot be determined by the results of this research.

Speech and Language Therapy with Twins Because two areas within nonlinear phonology were investigated in this study, and because of the interest in investigating the possibility of spontaneous acquisition of the co-twin's therapy targets, each twin was placed on different goals during the initial two intervention blocks of this project. This arrangement necessarily meant that the parents' support through home follow-up activities or incidental modelling and teaching during spontaneous conversation could not be enlisted because of the possibility of influencing the acquisition of untrained treatment goals. However, parental support is crucial in regular speech and language remediation that does not need to be controlled for research purposes, because it allows for "continuous" therapy to take place. In transferring the therapy method and design of this study to the remediation of twins in the regular clinical setting, then, it is recommended that both twins be placed on the same therapy goals so that not only the support of the parents can be 1

enlisted, but also so that the twins may support each other's progress by

91

demonstrating the accurate production of targets. Given the cited research findings, as well as the results of this study, it is quite possible that one twin will acquire the therapy goals more rapidly than the co-twin. Using a cyclic approach, as was done in this study, may allow for the same goals to be targeted despite an unequal acquisition rate. The design of the current research project required the subjects to be seen separately for therapy. If, as is recommended above, both twins were placed on the same goals, one needs to ask if it would be beneficial for them to be seen at the same time for therapy. This question was investigated during the fifth and sixth treatment sessions of Block 3 when the twins were seen together for therapy. When the twins were together, the dominance relationship was emphasized to the point where Subject 2 completed all the activities for both himself and his brother. During this time, Subject 1 appeared to withdraw from the therapy situation, allowing his brother do all the work and "deal with the outside world." This withdrawal even extended to activities that did not involve language activities and, therefore, rules out the hypothesis that Subject 1 felt inhibited because of the feeling that he may be linguistically inferior to his brother. Whether Subject 2 was seen for therapy together with his brother or on his own made no difference to him. For Subject 1, the less aggressive of the pair, attending therapy with his twin, however, did not lead to productive and active therapy. Given the above findings, it is recommended that twins be seen separately for therapy, so that the less dominant twin has equal opportunity to interact with the clinician and to benefit from language stimulation. It should be noted, however, that some anecdotal reports from parents and teachers indicates that twins constantly monitor the well-being of the co-twin. Separating twins, therefore, such as placing them in different classes may interfere with this monitoring and lead to anxiety. It is appropriate to place twins in the same class during the preschool and early elementary

92

school years if a strong relationship is demonstrated. Separating twins for short periods of time, however, as was done during the project, should not be an anxiety provoking situation, especially if the same activities are being completed because the same therapy goals are being targeted. In this case, the first twin to attend therapy can be reassured that the co-twin will complete the same activity. Likewise, the second child can be told that his/her sibling just completed the same tasks. In this way, each twin has a good idea of what the sibling is doing during the time that they are separated.

Autonomous Language No study of the language of twins is complete without at least a fleeting reference to the existence or nonexistence of autonomous language. Also termed twin language or cryptophasia, autonomous language refers to a "secret language" which some researchers claim exists between twins (Zazzo, 1960; Luria & Yudovich, 1968). Even if a secret language does not exist between twins, not all of the language they direct to each other may be intelligible to those outside the twin situation. Where phonology is concerned, unintelligibility can result from protracted phonological development, as was demonstrated by the subjects of this study. Because their phonologies are developing at approximately the same rate, each twin is able to understand the utterances of the co-twin even if nonstandard forms unintelligible to outsiders are used. In the case of the subjects of this study, the highly productive form [kaku] continued to be used for thank-you well after superior phonological skills had been acquired. Although this utterance may be uninterpretable for many, it presented no problems within the twin pair. Such "regressive phonological idioms" have also been exhibited by singletons on a normal developmental timetable (see, e.g. Ferguson & Farwell, 1975).

93

To date no researchers have reported the use of novel syntactic forms—that is, forms that are completely idiosyncratic in grammatical structure—by members of a twin pair. Idiosyncratic forms, however, have been observed in the secret language of nontwin siblings (Diehl & Kolodzey, 1981). With regard to twins, several studies have observed the use incomplete utterances, shorter utterances and a greater number of emotionally toned phrases when communicating (e.g. Savic & Jocic, 1973; Zazzo, 1977). Zazzo (1977) postulates that the linguistic deficit he observed in his twin subjects is compensated for by an increased development of other nonverbal communicative functions. Richon & Plee's (1976) finding that twins generally score higher on the nonverbal tasks of an intelligence test supports Zazzo's (1977) hypothesis. This author questioned an adult identical twin about her feelings regarding the existence of a secret language between herself and her co-twin. She reported that it was not a secret language that existed between them, but rather a deeper emotional bond than she experiences with others very close to her. This emotional understanding of each other allows each twin to gain the intent of a phrase even when it is incompletely or inexplicitly stated. To someone outside the twin pair, however, it may appear that a secret language exists. The connected utterances of the subjects of this study have not been analyzed to see if shorter or incomplete phrases were used. That a deep emotional bond existed between the subjects, however, cannot be disputed.

Summary and Conclusions This chapter re-interpreted the findings of this phonological study taking the atypical subjects into account. Twins are considered to be atypical in that they create a unique environment for each other by sharing all experiences and by mirroring each other's development.

94

The subjects of this study entered the project with essentially identical phonological inventories. This observation is in agreement with the twin research literature, which reports that monozygotic twins tend to be more similar than dizygotic twins or singleton children. This similarity in language acquisition is generally accepted as evidence that at least part of language development is under genetic control. Qualitative analyses of the language development of monozygotic twins have shown that, while both twins will follow the same path to acquisition, qualitative differences may be apparent at any one time. For example, similar branches of the phonological hierarchy may be emerging, but their specific acquisition time may differ between twins. This relative acquisition time was the most obvious difference between the twin subjects of this study. Subject 2 began generalizing multiple therapy goals during the third cycle of therapy. His brother, however, required a further three cycles of treatment before a similar acquisition spurt was observed. Two hypotheses were cited as possibly accounting for this observed difference in therapy progress. Initial investigation led to the hypothesis that Subject 2's syllabic goals were easier to acquire than Subject l's segmental goals. This hypothesis requires further investigation—possibly through comparing the results of intervention with another twin pair where the dominant twin is placed on segmental goals (but see Bernhardt, 1990). The second hypothesis accounting for the therapy results is based on the social awareness of the twins. Subject 2 was found to be the dominant twin of the pair and, as such, demonstrated superior social skills and awareness. This social awareness emphasized situations where communication breakdown occurred and motivated Subject 2 to acquire the phonological targets of therapy. Increased awareness also drew Subject 2's attention to correctly modelled target forms. It was suggested that Subject 1 was either not a risk-taking language learner or was not analytical in his approach to cognitive tasks. It seems unlikely, however, that

95

monozygotic twins would employ different cognitive strategies. Researchers have claimed that the personality differences (rather than cognitive ones) between monozygotic twins develop from the need to distinguish themselves as individuals separate from the twin situation. The difference in therapy progress observed in this study is more convincingly explained by the latter hypothesis. Based on the method and design of the current intervention project, recommendations were made to place twins on the same speech or language goals in therapy so that follow-up home activities can be completed with the caregivers. In addition, it was recommended that twins be seen individually for therapy so that the dominant twin does not complete all the activities and provide all the language while the co-twin withdraws from the tasks. Finally, an evaluation regarding the possible existence of autonomous language for this twin pair was given. Essentially, the unintelligibility of these twins resulted only from delayed phonological development. This delay may mean that persons outside of the twin couple will not understand what is said, while the co-twin has no difficulty comprehending the utterance because their phonological inventories are essentially the same. At least some of the inexplicit language which has been observed between twins is attributed to the fact that a deeper emotional bond exists between the pair allowing less detail to be verbalized before the intent of the utterance is conveyed. The current study focussed only on phonological acquisition and did not investigate this latter area. It can be concluded that twins are unique subjects for research projects which investigate not only issues concerning heredity in language development, but also the personality and social factors that influence language acquisition.

96

CHAPTER 7 CONCLUSIONS

The main purpose of this research was to investigate the possibility and advantages of applying a nonlinear phonological theory to the assessment and remediation of phonological disorders. Researchers have recently been developing nonlinear theories to account for the phonological rules and operations observed in different languages. They have found that, where previous generative theories have required rules involving complex transformations, nonlinear phonological principles allow rules to be more simply stated and logically related to similar rules in other languages.

Similarly, linguists interested in child language acquisition have

discovered that when acquisition data is examined from the point of view of nonlinear phonology, the data no longer appears to be deviant or idiosyncratic, but instead, follows the general principles and rules governing all languages. To date only the recently completed project by Bernhardt (1990) and this research have investigated the possibility of applying nonlinear principles when assessing and determining goals for phonological remediation in the speech-language pathology clinic. Following the research design and methodology of Bernhardt (1990), the current project applied nonlinear principles to a six-month phonological intervention study with a set of identical male twins. An assessment of the twins' phonological abilities at the outset of the project revealed that their strengths and deficits could easily be described as "established" and "unestablished" branches of nonlinear hierarchies at both the segmental and syllabic levels. Although these two

97

levels are associated in nonlinear theory, they were separated in this project to create a segmental and a syllabic condition for investigation. The segmental portion of the current research compared the acquisition of "higher" nodes of the feature hierarchy with "lower," or more deeply embedded nodes. The hypothesis that higher nodes would be established more quickly than lower nodes was investigated by targeting pairs of segments whose salient features fall at different levels in the hierarchy, within the same cycle of therapy. Intervention at the segmental level allowed the acquisition of two pairs of features to be compared. In both cases, the segment, whose salient feature was higher in the hierarchy, was acquired first. Bernhardt (1990) reports similar findings for her singleton subjects. These results, then, support assessing, and subsequently determining goals for intervention, based on nonlinear feature hierarchies. Conversely, it can be concluded that the results of therapy with phonologically delayed subjects provide independent evidence supporting feature geometries. Within the syllabic condition, it was not possible to compare the acquisition of higher level branching with the acquisition of more embedded branching as was done in the segmental condition. Therefore, remediation activities within this condition contrasted two different prosodic theories—the moraic theory and the onset-rime theory—to ascertain whether acquisition results could provide support favouring one or the other theory. The majority of intervention within this condition targeted the production of various consonant clusters. The differences between the two syllable theories affected the presentation and stimulation activities of intervention. Intervention following moraic principles allowed clusters to be presented as separate segments, while intervention based on onset-rime principles presented clusters as a unit. Depending on the clusters targeted, support for both hierarchies was provided by the results. Separating the consonants of a cluster to emphasize the presence of two constituents facilitated acquisition for the subjects of this study. The moraic theory of

98

syllable organization better accounts for this approach, because onset segments are appended individually to the syllable node rather than first being grouped under an onset node in opposition to the rime. The onset-rime theory, however, is better able to account for the intermediate acquisition step of epenthesis, which is commonly used by language learners. A comparison of the results of the two conditions of this study revealed that the syllabic targets were more quickly acquired than the segmental ones. Trial stimulation of each twin on the co-twin's targets supported the conclusion that syllabic targets are in fact easier to acquire than segmental ones. The results of Bernhardt (1990) agree with these findings.

Comparison of Nonlinear Phonology and Phonological Process Theory for Phonological Intervention Outlining the advantages of nonlinear phonology requires that this theory be compared to other phonological theory options. Currently, the most popular method of analyzing phonological data in the clinic uses a process analysis approach. Stampe (1969), who first organized phonological data according to processes, claimed that these processes represented mental operations and were unordered in their application to the (assumedly) adultlike underlying representation.

Researchers

currently using a phonological process framework hesitate to make claims regarding the status of the underlying representation. Not enough is known about the language learner's perceptual mechanism to conclude that the underlying representation is adultlike. This being the case, then, it is difficult to claim that phonological processes provide more than simply a description of the language learner's surface production. The first advantage of nonlinear phonological theories over process theories lies in the theoretical claims made. Nonlinear theory argues that the hierarchies which are basic to every level of phonological organization reflect the mental representation

99

of speakers. As such, nonlinear phonology provides an explanation of phonological rules rather than just a description of them. When applied to language acquisition, phonological processes necessarily imply that children are avoiding certain segments or syllable shapes ('fronting', for example, implies that the language learner is avoiding a "more back" target than her surface realization). In contrast, nonlinear theory claims that any discrepancies occurring between the child's surface forms and the adult targets result from the child not yet having established certain branches of the feature or syllable hierarchies. The second advantage of using nonlinear theories over phonological processes becomes apparent when determining targets for remediation in the speech-language pathology clinic.

Since phonological processes are unordered, according to Stampe

(1969), the theory itself provides no guidelines regarding which processes should be targeted first in therapy. Clinicians currently base decisions regarding intervention goals on suggestions such as following the developmental pattern of typical language learners in eliminating processes, or targeting those processes which are most detrimental to the intelligibility of the child. Remediation according to nonlinear principles involves establishing the higher, or less embedded, branches of the hierarchy before establishing the lower ones. The language learner necessarily moves from acquiring the most basic (and also universal) branching to the more complex language-specific

system.

The strongest argument in favor of using nonlinear theories in the clinical setting would result from remediation based on nonlinear phonological principles leading to the faster acquisition of therapy targets than if goals were chosen based on targeting phonological processes. Such a finding would imply that nonlinear hierarchies provide a closer match than phonological processes with the language learner's mental representation. The scope of this intervention study could not incorporate a comparison of these two theories. Further research is required that

100

compares the acquisition of phonological targets in intervention based on following the principles of nonlinear phonology with the acquisition results of remediation based on phonological process analyses.

Limitations of the Study The main limitation of the current project is that only two subjects were used. Since two conditions were tested in the intervention portion of the research, each condition could only be tested on a single subject. Claims based on the results of only one subject are not convincing evidence. The findings of this project, however, agree with the results of intervention with six singleton subjects (Bernhardt, 1990). Results from the eight subjects, taken together, demonstrate that nonlinear phonological theories may be applied to the assessment and remediation of phonologically delayed children. Further research applying nonlinear principles in the speech-language pathology clinic is required to further substantiate this claim and to add to the body of "nonlinear phonological remediation" data. A second limitation results from having used twins as subjects for this research. Chapter 6 outlined the complex interrelationship between the members of a twin pair. The psychological factors entering into this twin relationship are difficult to tease apart, meaning that research findings can only be tentatively generalized to the clinical population as a whole. The comparison of rates of acquisition with the personality differences of the twin subjects of this study led to the general claim that social awareness is prerequisite to phonological acquisition. This claim must be made with caution, because the two conditions of this study were unequal in difficulty. The findings may have been different had the opposite twin been assigned to each condition. A repetition of this study assigning the dominant twin to the segmental condition is required in order to substantiate any claims made regarding the role of personality factors in phonological acquisition.

101

Since nonlinear theory is currently motivating a great deal of research and is undergoing rapid development, changes in feature and prosodic hierarchies may lead to more refined clinical decisions regarding treatment goals. The main conclusion of this study, then, is that nonlinear phonological theory provides a workable and coherent framework for phonological assessment and intervention. Based on the positive results discussed here, it is claimed that intervention following nonlinear principles presents independent evidence supporting this theory. And finally, it is concluded that nonlinear theory provides a better theoretical framework than phonological process theory because of the explanatory value of nonlinear hierarchies and the logical process of choosing targets for treatment.

102

REFERENCES Archangeli, D. (1988). Aspects of Underspecification Theory. Phonology Yearbook 5: 183-207. Bernhardt, B. M . (1990). Application of nonlinear phonological theory to intervention with six phonologically disordered children. Unpublished doctoral dissertation: The University of British Columbia, Vancouver. Bruggemann, C . E . (1970). Twins: early grammatical development. Folia Phoniatrica 22, 197-215. Canter, S. (1973). Personality traits in twins. In Claridge, G . (ed.). Personality differences and biological variations. A study of twins. Oxford: Pergamon Press. Carrow-Woolfolk, E . (1985). Test for Auditory Comprehension of Language (Revised). Allen, T X : D L M . Chomsky, N., & Halle, M . (1968). The Sound Pattern of English. New York: Harper and Row. Christdas, P. (1988). The phonology and morphology of Tamil. Unpublished doctoral dissertation: Cornell University. Clements, G . N. & Keyser, S. J. (1983). CV-phonology; a generative theory of the syllable. Cambridge, Mass: MIT Press. Clements, G . N. (1976). Vowel harmony in nonlinear generative phonology: an autosegmental model. [Distributed by Indiana University Linguistics Club in 1980.] Clements, G . N. (1985). The geometry of phonological features. Phonology Yearbook 2, 305-328. Clements, G . N. (1989). A Unified Set of Features for Consonants & Vowels. Address to the Institut de Phonetique, Paris France, August, 1989. Davis, E . A. (1937). The development of linguistic skill in twins, singletons, and siblings, and only children from five to ten. University of Minnesota Institute of Child Welfare Monograph 14.

103 Day, E . J. (1932). The development of language in twins: a comparison of twins and single children. Child Development 3, 179-199. Diehl, R.L., & Kolodzey, K.F. (1981). Spaka: A private language. Language 57, 406424. Dinnsen, D . A . , Elbert, M . , & Weismer, G . (1980). Some typological properties of functional misarticulation systems. In: W.O. Dressler(ed-) Phonologies 1980, 83-88. Innsbruck: Innsbrucker Beitrage zur Sprachwissenschaft. Dunn, L . M . & Dunn, L . M . (1981). Peabody Picture Vocabulary Test. (Revised). Circle Pines, M N . : American Guidance Service. Edwards, M . L . & Bernhardt, B. (1973). Twin speech as the sharing of a phonological system.. Unpublished manuscript: Institute for Childhood Aphasia, Stanford University. Gandour, J. (1981). The nondeviant nature of deviant phonological systems. Journal of Communication Disorders 14, 11-29. Gardner, M . (1979). Expressive One-Word Picture Vocabulary Test. Novato, C A : Academic Therapy Publications. Goldsmith, J. (1976). Autosegmental phonology. Doctoral dissertation, Indiana University; Published 1979, New York: Garland Press. Grunwell, P. (1985). Phonological assessment of child speech. San Diego, C A : College-Hill Press. Halle, M . , & Vergnaud, J. R. (1980). Three dimensional phonology. Journal of Linguistic Research 1, 83-105. Hayes, B. (1982). Extrametricality and English stress. Linguistic Inquiry 13, 227276. Hayes, B. (1989). Compensatory Lengthening in Moraic Phonology. Linguistic Inquiry 20, 253-306. Hodson, B. & Paden, E . (1983). Targeting intelligible speech - a phonological approach to remediation. San Diego, C A : College-Hill Press. Hyman, L . (1985). A theory of phonolical weight. Paris, Dordrecht: Foris. Hyman, L . M . (1975). Phonology: Theory and Analysis. New York: Holt, Rinehart and Winston. Ingram, D . & Goad, H . (1988). Intratwin language acquisition. Unpublished manuscript: The University of British Columbia, Vancouver.

104 Ingram, D. (1976). Phonological disability in children. London: Edward Arnold Ltd. Jacobson, R. (1968). Child language, aphasia, and phonological universals. (A. R. Keiler, trans.). The Hague: Mouton. Kahn, D . (1979). Syllable-Based Generalizations in English Phonology. New York: Garland Press. Kaye, J., Lowenstamm, J., & Vergnaud, J.-R. (1987). Constituent structure and government in phonology. Unpublished manuscript: University of McGill, Montreal. Kiparsky P. (1979). Metrical Structure Assignment is Cyclic. Linguistic Inquiry 10, 421-441. Kiparsky, P. (1982). Lexical morphology and phonology. In Yang, S. (ed.), Linguistics in the morning calm. Seoul: Hanshin. Koch, H . L . (1977). Twins and Twin Relations. Chicago: The University of Chicago Press. Leben, W. (1971). Suprasegmental and segmental representation of tone. Studies in African Linguistics. Supplement 2, 183-200. Leben, W. (1973). Suprasegmental phonology. Indiana University Linguistics Club. Leinonen-Davies, E . (1988). Assessing the functional adequacy of children's phonological systems. Clinical linguistics and phonetics 2, 257-270. Leonard, L . B. (1985). Unusual and subtle phonological behavior in the speech of phonologically disordered children. Journal of Speech and Hearing Disorders 50,4-13. Leonard, L . , Newhoff, M . & Meselam, L . (1980). Individual differences in early child phonology. Applied Lingusitics 1,7-30. Levin, J. (1985). A metrical theory of syllabicity. Unpublished doctoral dissertation: Massachusetts Institute of Technology, Cambridge, M A . Liberman, M . (1975). The intonational structure of English. Indiana University Linguistics Club. Liberman, M . , & Prince, A. (1977). On stress and linguistic rhythm. Linguistic Inquiry 8, 249-336. Locke, J. L . , & Mather, P. L . (1989). Genetic factors in the ontogeny of spoken language: evidence from monozygotic and dizygotic twins. Journal of Child Language 16, 553-559.

105 Lorentz, J. P. (1976). An analysis of some deviant phonological rules of English. In: D. M . Morehead & A. E . Morehead (eds.), Normal and Deficient Child Language. Baltimore: University Park Press. Lytton, C , Conway, D . & Sauve, R. (1977). The impact of twinship on parent-child interaction. Journal of Personality and Social Psychology 35, 97-107. Macken, M . A . , & Ferguson, C . A. (1983). Cognitive aspects of phonological acquisition: model, evidence and issues. In Nelson, K. (ed.), Children's Language Vol. 4. Hillsdale, New Jersey: Erlbaum Associates Inc., pp. 256-282. Magnusson, E. (1984). The phonology of language disordered children: Production, perception, and awareness. Lund, Sweden: Liberforlang, Lund. Malmstrom, P. M . , & Silva, M . N. (1986). Twin talk: Manifestations of twin status in the speech of toddlers. Journal of Child Language 13, 293-304. Matheny, A . & Bruggemann, C. E . (1973). Children's speech: Hereditary components and sex differences. Folia Phoniatrica 25, 442-449. McCarthy, J. & Prince, A . (1986). Prosodic morphology. Unpublished manuscript: University of Massachusetts and Brandeis University. McCarthy, J. (1979). Formal problems in Semitic phonology and morphology. Unpublished doctoral dissertation: Massachusetts Institute of Technology, Cambridge, M A . McCarthy, J. J. (1988). Feature geometry and dependency: a review. Phonetica 43, 84-108. Munsinger, H . & Douglass, A. (1976). The syntactic abilities of identical twins, fraternal twins, and their siblings. Child Development 47, 40-50. Nettelbladt, V . (1983). Developmental studies of dysphonology in children. Lund, Sweden: Liberforlag Lund. Newcomer, P. & Hamill, P. (1982). Test of Language Development — Primary (Revised). Austin, T X : PRO-ED. Pike, K . , & Pike, E . V . (1947). Immediate constituents of Mazateco syllables. International Journal of American Linguistics 13, 78-91. Pollock, K. (1983). Individual preferences: Case study of a phonologically delayed child. Topics in Language Disorders 3, 10-23. Pulleyblank, D. (1986). Underspecification and low vowel harmony in Okpe. Studies in African Linguistics 17, 119-153.

106 Richon, G., & Plee, B. (1976). A propos de la dichotomie verbal/non verbal. Enfance 4-5, 495-509. Robbins, J. & Klee, T. (1987). Clinical assessment of oropharyngeal motor development in young children. Journal of Speech and Hearing Disorders, Volume 52,271-277. Sagey, E . (1986). The representation of features and relations in non-linear phonology. Unpublished doctoral dissertation: Massachusetts Institute of Technology, Cambridge, M A . Savic, S. & Jocic, M . (1975). Some features of dialogues between twins. International Journal of Psycholinguistics 4, 34-51. Selkirk, E . (1980a). The role of prosodic categories in English word stress. Linguistic Inquiry 11, 563-605. Smith, N. V . (1973). The Acquisition of Phonology. Cambridge: Cambridge University Press. Sonnenschein, S., & Whitehurst, G. J. (1980). The development of communication: when a bad model makes a good teacher. Journal of Experimental Child Psychology 3, 371-390. Spencer, A. (1984). A nonlinear analysis of phonological disability. Journal of Communication Disorders 17, 325-348. Spencer, A. J. (1986). Towards a theory of phonological development. Lingua 68, 3-38. Spencer, A . J. (1984). A nonlinear analysis of phonological disability. Journal of Communication Disorders 17, 325-348. Spencer, A . J. (1986). A theory of phonological development. Lingua 68, 3-38. Stampe, D. (1969). The acquisition of phonetic representation. In Papers from the Fifth Regional Meeting, Chicago Linguistic Society, pp.443-454. Stemberger, J. P. (1988). Between-word processes in child phonology. Journal of Child Language 15,39-61. Steriade, D . (1988). Reduplication and syllable transfer in Sanskrit and elsewhere. Phonology 5, 73-155. Waterman, P., & Shatz, M . (1982). The acquisition of personal pronouns and proper names by an identical twin pair. Journal of Speech and Hearing Research 25, 149-154.

107 Werner, E . & Kresheck, J. (1983). Structured Photographic Expressive Language Test — / / . Sandwich, ELL: Janelle Publications. Wilson, R. S. (1974). Twins: mental development in the preschool years. Developmental Psychology 10, 580-588. Wilson, R. S. (1975). Twins: patterns of cognitive development as measured on the Wechsler Preschool and Primary Scale of Intelligence. Developmental Psychology 11, 126-134. Wilson, R. S., Brown, A . M . , & Matheny, A . P. (1971). Emergence and persistence of behavioral differences in twins. Child Development 42, 1381-1398. Zazzo, R. (1960). Les Jumeaux: le couple et la personne. Presses Universitaires de Frances. Zazzo, R. (1977). Genesis and peculiarities of the personality of twins. In Nance, W. E. (ed.) (1978). Progress in clinical and biological research, Vol 24a. Twin research: Proceedings of the Second International Congress on Twin Studies. New York: Alan R. Liss, Inc., pp. 1-11.

108

APPENDIX 1 N O N L I N E A R M I N I M A L SPECIFICATIONS O F E N G L I S H C O N S O N A N T S

109

APPENDIX 1

Nonlinear minimal specifications of English consonants:

Abbreviations:

/m/

[cons] = [consonantal] [son] = [sonorant] [+cont] = [+continuant] [*cont] = complex continuant marking required for affricates [-ant] = [-anterior] [dist] = [distributed] L N = L A R Y N G E A L NODE

ROOT

ROOT

ROOT

[cons] [nasal]

[cons]

PLACE

PLACE

PLACE [voice]

/

/

labial

/n/

[cons]

labial

ROOT

labial

ROOT

ROOT

/d/

[cons] [nasal]

[cons]

[cons] LN [voice]

fkj

ROOT

ROOT

[cons] [nasal] PLACE dorsal

ROOT •[cons]

[cons]

PLACE

I

dorsal

PLACE [voice] dorsal

110

HI

ROOT

ROOT

M

[cons] / M+cont] PLACE

-[cons] [+cont]

LN PLACE [voice]

Isl

labial

labial

ROOT \Mcons] ^ [+cont]

ROOT

ROOT [cons] [+cont]

LN

[cons] [+cont]

LN

[voice]

III

ROOT \Mcons] ^ [+cont] PLACE

/

Izl

^7V[ LN / [voice]

coronal

1

[cons] [*cont]

/d3/

x

I

[dist]

ROOT [cons] [*cont]

LN

[cons] [+cont]

LN PLACE

PLACE [voice]

[voice]

[-ant]

coronal

ROOT

PLACE

I

i

j [+cont] PLACE | coronal

ROOT P\Mcons] j \[+cont] PLACE

[-ant]

ROOT

coronal

c o n s

I

[-ant]

Ml

IBI

ROOT

coronal

I

[-ant]

coronal

I [dist]

Ill

/w/

ROOT r^[son] PLACE

1)1

ROOT ^-[son]

/r/

ROOT J\"-[cons] j ^[son] PLACE

I labial

N

ROOT [cons] [son]

I

coronal

I

[dist]

APPENDIX 2 G R A N D PROBE WORD LIST

APPENDIX 2

Grand Probe word list (based on Bernhardt, 1990): Vowel-initial:

/nV

M

ivi

airplane all ear eat eating egg eyes ice icecubes icy itch itchy off

mask me milk mom mommy money mouth mouthy music musicbox

basket bee bib black blue boot boots bootie brown brush brushing ' buzz buzzing

page pages pie

SJll orange up /s/ Santa Claus scarf screwdriver see sewing machine sis sleep sleeping smooth snake snakey snow snowing soap spoon spring squirrel star starry

AV

/d/

nine no noise noisy

dad daddy doll dollhouse dolly dress dressing

glasses glove glovey go green gum gummy

M

/s/ (cont.) string sun sunglasses sunny sweater

Pig piggy plum pour pouring present pup purple A/

television toe toot tooth , toothbrush toothy truck tub tubby T.V. Ata/

cage cagey candle clock coffee comb combing cook cooking crayons key quarter

"G" judge jump jumping A;/

chair cherries chicken church

HI

NI

Ihl

M

feather fingers fish fishing five flower frog

van

hang hanger hi horse hug hugging

zipper zoo

M

N

/w/

/e/

rabbit radio read reading red roar row rub rubbing

la laugh laughing leaf leafy Lyle

wagon watch watching wave waving why

three throw throwing thumb

yellow

III shoes shirt

m that

APPENDIX 3 G R A N D PROBE D A T A FOR B O T H SUBJECTS

116

APPENDIX 3 Grand probe data for Subject 1: Target

$.™dJ*9teJ±

/vowel . i n i t i a l / j

lGrandProbe#2

JGrand

j

j

JGrand Probe#4 . j

airplane

i P '

all

j

i

!

jaco

ear

jer

j

j

jir

ears

|

eat

3 r

__

e

n

^ P?

Probe #3

9r

in

i P5 Er

in

:?TP

ein

lirs

jit

eating

;i?irj

?gg

|eg

|

|eg

jefl

?gg?

|

[eg?

|

|

eyes

?2U).ts

jais

jaiz

jaiz

ice

jeis

!

i

j

icecubes

j a i k j u b s i

icy

jajsi

|

itch

jus

j

jits)

itchy

iitsi

i

:its)i

off

ias

jas

jas

jas

Ian

i

n

j _j

j

ii?s}i

ZIZZZZZZZZZ^ZZZZZZZZZZZZZ^ on

ian j

orange j

i [(pra^ges! I i P l i n t s i p r a n t s w i o r e n t s e z j 5 ™ i L . j-?™t?. i |

ffi

|

I

bp

hs.

elephant

j

;

JAjahat

jaeamant

j

i

j

Initial/p/j page pages

j p e ^ s j p e r d z | j

j

jpe(fx)a

jpeidaaz

j

j p e i d s . J E ^ d s jpeids) jpeidza

117

pie

P»g.. p»ggy.

ipai JPM..

M9...

m\..

M9L

plum

iP.

pour

jpor

r^unng

jppnrj

present.

!P ?. .^.

Ipefiant

[P"P

JR*R

I

Am

e

jpor

n

M9...

.•P^m..

.|P ..

jpor...

jpor...

.mm

jp?ng j^presents^ jpej3?ts

jparpol

P.urp.1?.. pitcher

Am

jparpol ]P">?.... jptfsar r

Initial Ibi ibsskit

basket

jbsskit

jbjesnkEt

bee

Ibi

Ibi'

bumble bee

jbAmbAmbi

jbAmbAbi

bib

jbib

jbib

black

jbsek

;ba?k

boot

ibut

jbut

Ibut

boots

ibuts

ibuts

boo tie

ibudi

ibuts i(booties) Ibudis

brush

ibAS

ibi

ibaek

jbaek

ibAS

\b\J

ibAS

bnjshingi^S"]

i^sn)

$/±[}J}..

jbASirj

buzz

ibAS

buzzing blue

ibAS jpA(s)iig

ibu

ibAds

jbAds

jbj\dsnj

ibAzPirj

ibu

ibu

118

b r o w n I

IbEeon

I

ibaon

box

jtKits

jbats

j

jbaks

I n i t ^ /m/

j

j

j

j

mask

: ..?.?.H.

m

e

imaesk

m

i

m

i

i

milk

1™.!®:^.i ?©H.

mom

imam

mommy

imAmi

money

imAni

mouth

i .