The Possession-Modification Scale: A universal of ... - CiteSeerX [PDF]

The Possession-Modification Scale: A universal of nominal morphosyntax Irina Nikolaeva School of Oriental and African Studies and Andrew Spencer University of Essex March 6, 2010

Contents 1 Introduction

4

2 Semantic functions

4

2.1

Inalienable possession

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

2.2

Alienable possession . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

2.3

Attributive modification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

2.4

Modification-By-Noun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8

2.4.1

Noun-noun compounding . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9

2.4.2

Relational adjectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Overview of encoding strategies

11

3.1

Strategies without agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.2

Strategies with agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.3

Construction markers which register features of their host . . . . . . . . . . . . . . . . . 16

4 Patterns of polyfunctionality

18

4.1

A 6= B 6= C 6= D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.2

A = B 6= C 6= D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.3

A = B = C 6= D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4.4

A 6= B = C 6= D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4.5

A6=B6=C=D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.6

A 6= B = C = D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.7

A = B6= C = D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1

4.8

A = B = C = D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5 The Possession-Modification Scale

34

5.1

Monotonicity and its motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

5.2

Counter-examples to the scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.3

Juxtaposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6 Conclusions and prospect

41

2

Abstract In many languages possessive constructions and attributive modification are expressed by similar or identical encoding strategies. We compare four construction types: attributive modification proper, as in tall tree (A), modification-by-noun, as in cat food (B), alienable possession, as in Mary’s book (C) and inalienable possession, as in Mary’s hand/daughter. We present an overview of encoding strategies and then survey the way that different languages deploy those strategies to express the four constructions. For instance, in some languages all four types are expressed by different morphosyntax, while in others, constructions A, B might be expressed by one construction and C, D might together be expressed by a distinct construction. We propose a Possession-Modification Scale, A < B < C < D, with respect to which languages observe a monotonicity requirement: if a language expresses, say, constructions A and C in the same way then it will express construction B in that same way, too. This scale reflects strong tendencies rather than an absolute universal and we discuss certain classes of counter-example to monotonicity. Keywords: possession, attributive modification, alienable possession, inalienable possession, relational adjective, possessive adjective, noun-noun compound, ezafe

3

1 Introduction Many languages employ the same or similar morphosyntactic strategies for expressing both the possessive relation and what has traditionally been termed ‘attributive modification’. For example, possession is often expressed by a preposition of the type ‘of’, but this construction can also be used to form attributive modifiers (‘sea of green’ = ‘green sea’). On the other hand, possession is often expressed by means of a word which is formally an adjective. In languages such as Persian exactly the same construction (‘ezˆafe’) is used to linked possessors with their possessed nouns and attributive adjectives with their heads. Similarly, the associative marker in Bantu links the head noun with various types of dependents including possessors and property words, while in Mandarin Chinese both possessor NPs and attributive modifiers can be connected to the head by the same linking particle de. This paper explores these similarities and other patterns of polyfunctionality in the expression of adnominal dependents. It is therefore partly similar in its goals to Gil (2005), where languages are shown to vary in the degree to which they make grammatical distinctions between different semantic types of attribution. Gil’s paper concentrates on possessors, adjectives and relative clauses and shows that in some languages one and the same encoding strategy can be used for more than one semantic function. This is the ultimate conclusion of the present paper too, but unlike in Gil’s data, the patterns of polyfunctionality we have encountered are not random but seem to suggest an implicational relation. We seek to provide a motivation for it, which is essentially semantic in nature. Unlike Gil, we only address monoclausal constructions, i.e. relative clauses are not considered, unless some kind of clausal structure is the principle means of expressing the meaning we are interested in. The structure of the paper is as follows. In Section 2 we describe the basic semantic types of possession and modification. Section 3 provides a general overview of morphosyntactic strategies that are used to link the nominal head with its dependents. We then turn to the patterns of polyfunctionality in the expression of the meanings detected earlier and show in Section 4 that only some of the theoretically possible patterns seem are frequently attested. Section 5 suggests an explanation for this fact based on the simple idea that cross-linguistically recurrent identity in form reflects similarity in meaning. We will also discuss apparent counterexamples to our generalization and their significance for the understanding of canonical ways of encoding. Section 6 concludes the paper and suggests that the implicational scale we have proposed has important consequences for our understanding of word classes and the gradual transition from nouns to adjectives.

2 Semantic functions We present here the basic semantic characterization of adnominal dependents, excluding quantifiers and determiners. Following much previous research, alienable and inalienable possession are taken here to be the two basic semantic types of the possessive relation. We further suggest that in addition to canonical attributive modification, nouns can be modified by a noun-like entity. This type is referred to as ‘modification-by-noun’. This gives us four semantic types of adnominal dependents: inalienable possession, alienable possession, modification-by-noun, and attributive modification (‘modification-byadjective’).

2.1 Inalienable possession Constructions referred to as possessive are known to convey a wide variety of meanings, of which ownership or possession, in a strict sense, is only one type. Many authors, including Kempson (1977: 125), R. Hawkins (1981), Williams (1982: 283), Sperber and Wilson (1986: 188) and Herslund and Baron (2001: 13), stress that the list of possible interpretations of the relation between the possessor and

4

the possessee (possessum, possessed noun) can be extended indefinitely, subject to contextual factors.1 However, these widely diverse possible possessive meanings do not have the same status. In some instances there are relations that appear to be the only or at least the default interpretation for a specific possessive construction. As has been repeatedly noted in the literature (Chomsky 1970; Jackendoff 1977; Partee 1997; Partee and Borsch¨ev 2003 among others) the semantics (and perhaps the syntax) of possessives is largely determined by the nature of the possessed nominal. Following much previous work, Barker (1995) distinguishes between ‘lexical (or intrinsic) possession’ and ‘extrinsic possession’. For the former the semantic interpretation is determined by the meaning associated with a relational head noun. In this sense they are semantically dependent. The meaning of the head (possessed noun) provides a determinate specification of the nature of the relation between it and the possessor. Thus, the phrase John’s wife biases an interpretation of inalienable possession (‘the woman to whom John is married’). This is ensured by the fact that the nominal wife is a relational noun. In principle, this does not preclude the possibility that the expression could denote somebody else’s wife, whom John has selected as a winner in a competition for wives, but such an interpretation is constrained to occur in very special pragmatically stretched contexts and does not sound completely natural. Similarly, deverbal relational nominals and gerunds express lexical possession relations: for example, in the phrase my purchase the possessor is restricted to being a ‘purchaser’, with the semantic determiner of this relation being, of course, the intrinsic meaning of the verb. Lexical possessives contrast in Barker’s theory with extrinsic possessives involving non-relational nouns, e.g. John’s book. In cultures that have an institution of legal ownership this expression will often be paraphrasable as something like ‘book which John owns’. However, this is by no means the only possible interpretation. It can also refer to the book that John wrote, the book that he stole, the book he always talks about, the book he wants to buy or the book that he has just been given to review as a class assignment. The possessive relation here is not narrowly determined on the basis of inherent semantic properties of the possessed noun and, consequently, allows for various readings. Although there is often a bias for an ownership interpretation, consistent with a paraphrase using the verb own, other relations are also inferable. Whatever the precise factors and mechanisms might be, the interpretation of the intended relation between the possessor and the possessee is extrinsic to the lexical semantics of possessed noun. That John’s book, unlike John’s wife, is semantically ambiguous is due to the fact that the noun book is not intrinsically relational. Its semantic structure is such that there is no exclusive candidate for elaboration by the possessor nominal. In this paper we will rely heavily on these insights but use the terms ‘alienable’ and ‘inalienable’ possession instead. The traditional inalienable/alienable distinction corresponds to Barker’s lexical/extrinsic distinction. In inalienable possession the head noun is expected to be a relational noun such as a kin term, a meronym (part-of term, for instance, a body part), a topological noun or a noun denoting an inherent property, although the entities classified as inalienable are known to vary considerably from language to language. There does not seem to be any clear semantic basis for the alienable/inalienable opposition in a cross-linguistic sense, nor any universally applicable hierarchy of what counts as inalienable (Chappell and McGregor 1996: 89, Heine 1997). In each language the choice is determined lexically: it has to do with the lexical semantics of the possessed noun rather than style, the speaker’s construal of the situation or other similar factors (cf. Nichols 1992: 120). The twofold classification of possessives into lexical/intrinsic/inalienable and nonlexical/extrinsic/alienable raises the question which type is semantically ‘primary’. It is sometimes thought that the prototypical possessive relationships include part/whole, kinship and legal ownership/disposal (Langacker 1991: 169, Langacker 1995: 59, Koptjevskaja-Tamm 2001: 961). However, 1 We will mainly use the term ‘possessee’ as a synonym for other terms used in the literature such as ‘possessed noun’ or ‘possessum’. In the context of possessive constructions the term ‘head’ refers to the possessee and the term ‘dependent’ refers to the possessor.

5

there is no apparent way of grouping these three meanings together in semantic terms, while the alienable/inalienable distinction may be expressed in the argument structure as the distinction between relational vs. non-relational head nouns (for semantic representations see Section 5). Moreover, languages do have unambiguous encoding of the alienable/inalienable distinction (Chappell and McGregor 1996), while the distinction between part/whole, kinship and ownership, on the one hand, and other possessive meanings, on the other, never seems to be grammaticalized. We therefore prefer to treat alienables and inalienables differently and view the latter as prototypical possession.2

2.2 Alienable possession As mentioned above, in alienable possession the possessed noun is not relational. This affects the interpretation of the relation between the possessor and the possessee. If the possessee is a relational noun, the possessive relation is the relation denoted by the possessee. If the possessee is non-relational, the possessive relation largely depends on the context of use of the possessive construction. It does not make a great deal of sense semantically to think of these interpretations as possession in a strict sense, despite the fact that this is commonly the type of term used to describe these constructions. The only reason for referring to them as possession is because the formal means of expressing these meanings are often the same as those for expressing ownership. The question is how non-canonical possessive meanings arise and how a single construction comes to express such a profusion of semantic relations. Taylor (1989, 1995, 1996) suggested that the semantics of possessives is not amorphous but instead has an internal structure, which does not depend on a single definitional criterion. The possessive relationship is best regarded as a cluster of several independent properties (simplified after Taylor (1996: 340)): (1)

The possessive relationship a. The possessor is a specific human being; b. The possessed is an inanimate entity, usually a concrete physical object; c. The relation is exclusive, in the sense that for any possessed entity, there is usually only one possessor; d. The possessor has exclusive rights of access to the possessed; e. The possessed is typically an object of value, whether commercial or sentimental; f. The possessor’s rights of access to the possessed are invested in him through a special transaction; g. The possession relation is long term, measured in months and years, not in minutes or seconds; h. The possessor is typically located in the proximity of the possessed.

or inalienable possession is prototypical because it is characterized by essentially all the properties listed above. It involves an exclusive asymmetric long-term relation and physical proximity between two entities, for each possessee there is only one possessor who has the right to make use of the possessee, and the possessor is normally an individuated human being (cf. Heine 1997, 2001). But possessive constructions involving non-relational nouns are open to multiple interpretations which differ greatly with regard to their relative degree of prototypicality. The more relevant properties characterize the relation, the more it resembles the inalienable prototype. Like inalienable possession, legal ownership involves more or less all relevant properties. That is why it is often taken to be one of the prototypical possessive relationships and can be viewed as the unmarked default interpretation for extrinsic possessives in cultures with notion of personal possession.3 But since the prototype is a 2 While in most languages the alienable/inalienable opposition is binary, there may be different degrees of inalienability: some languages have several possessive classes. We will ignore these distinctions for present purposes. 3 Taylor (1989: 681) argues that when the possessor is in focus (This is JOHN ’s car), as well as in some other special contexts

6

Perm

complex cluster of properties, it permits deviations. Possessive relations deviating in one way or another from the prototype are restricted by discourse-based conditions and can be described with reference to various parameters, such as control and time (Heine 1997). For example, when John’s book is taken to mean ‘the book that John is reading at the moment (but which does not belong to him)’ properties c., d., g., and perhaps f. do not hold. The context here is characterized in terms of the speaker’s encyclopaedic knowledge of the usual relation(s) that obtain between the relevant entities, as well as the specific discourse situation that might mediate and modify this relation. In many languages alienable and inalienable possession do not formally differ. When the alienable/inalienable distinction is expressed, two major tendencies seem to be at work. In many languages alienable possession is morphologically marked and inalienable possession is not (Haiman 1985: 130; Croft 1990: 174–176; Nichols 1992: 117; Chappell and McGregor 1996: 45; Heine 1997: 172). For example, in Jarawara (Araw´a) alienable possession is expressed by the possessive marker kaa. In addition this language has about 150 inalienably possessed nouns, which are obligatorily preceded by a possessor without a possessive marker, cf.: (2)

Jarawara (Dixon 2000: 489, 490) a.

Manira kaa Manira POSS ‘Manira’s dog’

jomee dog

b.

Manira mani Manira arm ‘Manira’s arm’

The same pattern is observed in Chatino (Zapotecan) (Carleton and Waksler 2000) and a number of other languages of Australia and South America. On the other hand, in a number of languages alienable possession actually requires less morphological marking than inalienable possession. An example of this situation is provided by Dizi (Omotic) cited in Nichols (1992: 119). In this language inalienable possession requires a possessive genitive and alienable possession is expressed by mere juxtaposition accompanied by tonal alternations. (3)

Dizi (Nichols 1992: 119) a.

dadakn gel`ı boy.POSS head ‘boy’s head’

b.

dad`a k`ıa` n`u boy dog ‘boy’s dog’

As argued in Nichols (1992: 117), inalienable possession is commonly head-marked while alienable possession tends to be dependent-marked or has no morphological marking. In other words, inalienable possession is less dependent-marked.4 Below we will see that there are languages where alienable possessives pattern together with modificational constructions.

2.3 Attributive modification It is usually thought that the two main types of restrictive modifiers are non-intersective intensional modifiers (e.g. the alleged murderer) and intersective modifiers (the big house). Following a long typological tradition (e.g. Dixon 1991; Croft 1990) we take the basic type of modifier to be a word denoting some gradable property concept. But we differ from Gil (2005) in that we do not take colour in English, the ownership reading is the only available interpretation of the alienable possessive relation. Barker (1995) and Storto (2004) provide further evidence that (legal) ownership is an unmarked interpretation of extrinsic possessives. 4 This generalization may need further investigation: examples of inalienable possession cited in descriptive grammars usually involve pronominal possessors, and pronominal possessors are independently more likely to be marked on the head (via pronominal incorporation) than lexical possessors.

7

terms to represent a prototypical type of attribute. Colour terms tend to behave differently from other attributes. They are often grammaticalized from nouns and can sometimes retain nominal properties lacking in true adjectives. This is even found in English: for instance, in archaic or poetic registers we can say things like a sky of blue, a coat of green, a wine of deepest red and so on. These constructions are unavailable to non-colour terms: *a sky of bright, *a coat of warm, *a wine of sweet. For us prototypical modifiers are words denoting gradable properties of size, shape, age and the like.5 Modifiers of this type are usually represented as one-place predicates whose only argument is associated with the head noun. In the standard semantic analysis, the meaning of attributive modification by gradable property words such as the big house is set intersection. The denotation of the adjective identifies a certain subset in the denotation of the head noun by specifying which house is meant. Since modification narrows the concept associated with the head noun, it serves the purpose of classifying the respective entity. Distributionally, then, an adjective-noun group is identical to a common noun in isolation. For example, we can take the gradable property (intersective) adjective big and the common noun house to denote sets of individuals. Interpretation can proceed compositionally: the (house ∧ big) = λx [big(x) ∧ house(x)]. In other words, the big house is an entity that belongs both to the class of houses and the class ‘big’ of big things (modulo syncategorematicity) (cf. Siegel 1980; Higginbotham 1985; Larson and Segal 1995 and many other references). From this it follows that the basic type of nominal attributive modification will be intersective modification.

2.4 Modification-By-Noun It is also possible for languages to modify nouns with reference to other entities, even those that have purely nounlike denotations (such as concrete count nouns) and possibly belong to the lexical class of nouns. There does not seem to be a standard term for this semantic phenomenon. It was referred to as ‘non-anchoring relation’ in Koptjevskaja-Tamm (2000, 2004) and other work, or ‘specification’ in Heine (1997: 156-7), but the latter term is unfortunate because it is also used in defining phrase structure relations. We will refer to the relevant relation as ‘modification-by-noun’ or ‘nominal modification’. According to Koptjevskaja-Tamm, the dependent in non-anchoring relations serves to classify, describe or qualify the class of entities denoted by the head. Heine (1997: 157) states that the specifying dependent “refers to the same general entity as the specified” and characterizes the head noun by narrowing down the range of possible referents that may qualify as the specified. Similarly, we maintain that in modification-by-noun the denotation of the resulting phrase is a subset of the denotation of the head noun, much like in attributive modification addressed above (cf. Spencer 1999). We discuss this further in Section 5. The strongest typological claim we can make is that the option to modify a noun by another noun is open to all languages, even if they may have to use very different morphosyntactic devices. Below we summarize two frequent strategies for achieving modification-by-noun, which exist exclusively for this 5 It

is not entirely clear in some cases what the basis of Gil’s classification is. For instance, he describes Albanian as being ‘highly differentiated’ in that genitives (our possessives), adjectives and relative clauses all receive different encodings. Yet in Albanian the standard morphosyntax of both adjectives and ‘genitives’ involves a proclitic (or prefix) attached to the adjective or noun phrase and agreeing in gender, number, case and definiteness with the noun head (and this possessive strategy is also the productive way in which a noun may modify another noun generally.) Thus, Albanian represents for us a clear case in which all four of our primary modification functions are expressed in the same way, the exact opposite of Gil’s classification. Similarly, Chukchi is said to be highly differentiated, yet the possessive and modification-by-noun encodings are essentially a species of attributive adjective. Likewise, Japanese is said to collapse adjectives and genitives. However, the genitive construction is expressed by the postposition (‘case marker’) no, while the prototypical adjective (including basic colour adjectives) is to a large extent a kind of verb (though there is a large class of adjectives marked by a postpositional marker na as well as a much smaller number marked with no, a homophone of the genitive marker). Clearly, Gil’s description is classifying something rather different from what we are classifying. We leave it to future research to investigate the relationship between the two typologies.

8

purpose. In Section 4 we will look at languages where the construction used for nominal modification also conveys other meanings. 2.4.1

Noun-noun compounding

A typical instance of modification-by-noun is Germanic-style NN compounding. In a regular endocentric compound of the type London bus the first noun serves as the attributive modifier of the head noun. This is the canonical function of adjectives, and indeed in English we often find synonymous pairs in which a relational adjective alternates with an underived noun: Paris/Parisian lifestyle. Compounds do not have canonical properties of either possessors or modifiers. The first components of compounds take attributive modifiers like nouns, but do not take specifiers like adjectives and have a variety of other adjectival properties. In English it is especially difficult to determine whether the modifier London in London bus is categorially a noun or an adjective because there is little relevant morphology which distinguishes the two categories. But the only sense in which modifiers such as London have been claimed to be adjectives is in terms of their syntactic function and environment, not in terms of their morphological category. Spencer 2003 argues that in productive English syntactic NN compounding (as opposed to morphological compounding; see Huddleston and Pullum 2002: 448f for justification of this distinction) we have a relational adjective which retains the morphology of the input noun. Some authors have suggested analyses for compounding which are not compatible with the modification analysis. For instance, Ackema and Neeleman (2004: 80f) claim that the semantic relationship between the head N and the non-head N is semantically opaque and hence non-compositional. This can only be true of compounds which have acquired unpredictable lexicalized meanings. English NN compounds do tend to acquire fixed uses, especially if they have a high textual frequency, giving rise to the impression that the construction itself expresses some kind of determinate meaning over and above the meaning of its parts. However, compounding in English is entirely productive. It is difficult to see how completely productive compounding could be analysed as a form of modification if it were completely non-compositional. The meaning of new coinages can only be determined pragmatically. If a speaker of English coins a new compound, such as, say, tree paint, and this compound is interpretable in some specific context, then that interpretation can only be defined in terms of the (contextually given) understandings of the terms ‘tree’ and ‘paint’. Similarly, if a particular pen is used for signing cheques then it can be called chequebook pen. If a pen came as a free gift when somebody opened a new bank account s(h)e might also call it chequebook pen. There is no way that we can derive the special meaning of elephant gun from elephant and gun. In British English and related varieties there is an expression speed camera which all speakers interpret to mean ‘roadside camera whose function is to take photos of vehicles which exceed the speed limit’. Only the knowledge of the world (specifically of traffic regulations) will allow a speaker to derive this meaning. Downing (1977) shows on the basis of an experimental study that it is impossible to catalogue the compounding relationships exhaustively (see Levi 1978 for a misguided attempt to establish a set of fixed meanings). Any list will at most reflect the fact that certain types of relationships are “of greater classificatory values than others”. She further argues that, although the number of possible compounding relationships is in principle non-finite, it varies from context to context. For example, for the newly created non-lexicalized compound pea princess the following interpretations were suggested among others by the native speakers (Downing 1977: 820): ‘a fairy princess who rules the pea people’, ‘the fairytale princess who felt the pea under her mattresses’, ‘the princess of the Pea kingdom’, ‘the princess shaped like a pea or with the colour of a pea’, ‘a princess whose family is rich with pea farms’, and so on. There is nothing in the meaning of the compounding construction that can explain this. This discussion is important because it shows that the non-compositionality claim is founded on a misunderstanding. Ackema and Neeleman in effect mistake ‘semantically non-determinate’ for ‘non9

compositional’. They therefore think of all compounds as idiomatic. But the fact that a relationship has to be defined pragmatically rather than purely in terms of the fixed semantic representations of the constituents does not make a relationship non-compositional. That is to say that the interpretation of the productively formed compounds IS fully compositional. However, the real function of NN compounding is to permit a very general relationship to be expressed between two nouns. This relationship seems to be no less free than that between the possessor and possessed in alienable possessive constructions, but it is best thought of as a kind of modification. 2.4.2

Relational adjectives

Many languages do not permit NN compounding. An alternative strategy for realizing modificationby-noun is to use morphology to turn a noun into a word which has the grammatical properties of an adjective. Adjectives of this sort are usually termed ‘relational’6 and are noncanonical, in the sense that they do not denote a gradable property. However, they display canonical adjectival morphosyntax. In particular, they may show agreement with the head. The ‘pure’ form of relational adjective is one in which the derived adjectival form has exactly the same range of denotations as the original (undetermined) noun. In other words, a relational adjective is derived from the base noun by a process of transposition (Beard 1995) which alters the morphosyntax of the word but does not introduce any additional predicate to the semantic representation. Relational adjective formation is simply a way of bringing a noun into line with the grammar of the language so that it can serve as a syntactic modifier. In this sense, the construction ‘relational adjective + noun’ subserves effectively the same function as English NN compounding. English has relational adjectives but these are restricted in their occurrence, partly because the standard way of modifying a noun by a noun is by compounding. The relational adjective is essentially a borrowing from Romance languages (mainly Latin and Norman French) and Greek and often competes with other devices for realizing modification-by-noun, notably with NN compounding. In some cases only one or the other strategy can be applied. Thus, in linguistic terminology we talk about verb phrase but not #verbal phrase. On the other hand we speak of morphological operations not #morphology operations. Occasionally, we get free variation as in preposition/prepositional phrase. Relational adjectives in English seldom have a purely relational semantics, but often acquire additional nuances of meaning or restrictions on meaning. For instance, when it refers literally to ‘wood’ (and is not used, for instance, in a metaphorical sense, as in wooden acting), the adjective wooden almost always means ‘made out of wood’. This restriction is lacking in N N compounds with wood as first member. In other languages, however, especially those lacking the compounding strategy, a relational adjective will typically involve no additional semantics whatsoever. Relational adjective formation is simply a way of bringing a noun into line with the grammar of the language so that it can serve as a syntactic modifier. For example, in the Slavic and Romance language groups modificational compounding is not permitted and in those languages we often find adjectives that express some kind of indeterminate relation between two nominal denotations. In Russian relational adjectives are derived from nouns with the suffix -n-. Like canonical qualitative adjectives, they normally precede the head and agree with it in case, number and gender, although their syntactic behaviour does differ in part from the behaviour of qualitative adjectives: relational adjectives in Russian do not normally occur as predicates and lack so-called ‘short forms’. A typical example is kniˇznyj magazin ‘book shop’, where kniˇz- is the palatalized allomorph of the root knig- ‘book’ found before suffixes such as -n-, while -yj is an agreement affix. The meaning of relational adjectives in -n- is as general as in English NN compounds. For example, kniˇznyj usually means ‘having books, with books’ and zvezdnyj < zvezda ‘star’ typically means ‘starry, having stars, with stars’. However, stennoj derived 6 though

the reader is warned that the term ‘relational adjective’ is sometimes used in other senses, too.

10

from stena ‘wall’ does not mean ‘having a wall, with a wall’ but ‘located on a wall’ (relation of place), zamoˇcnyj derived from zamok ‘lock’ typically means ‘used for a lock’ (relation of purpose), moloˇcnyj derived from moloko ‘milk’ usually means ‘made of milk’ (relation of material), and noˇcnoj derived from noˇc ‘night’ means ‘in the night’ (time relation). But even with examples such as these we find alternative uses. This list of possible relations between two entities is not exhaustive, and each particular interpretation depends on the semantics of the based noun and general knowledge of the world. For instance, moloˇcnyj can occur in expressions such as moloˇcnyj kombinat/zavod ‘milk factory’, moloˇcnaja dieta ‘milk diet’ and moloˇcnyj saxar ‘lactose’ (literally ‘milk sugar’). While zvezdnyj usually has the proprietive meaning of ‘with/having stars’ as in zvezdnoe nebo ‘starry sky’, with other head nouns it can have other meanings, cf. zvezdnye vojny ‘star wars’, zvezdnyj god ‘sidereal year’, and zvezdnaja karta ‘star chart’ (see also other examples cited in Mezhevich 2002). Thus, the semantics of relational adjectives in -n- is essentially empty: they express some pragmatically defined relation between the head noun and its modifier.7 Overall, then, the range of functions and uses of relational adjectives corresponds very closely to the range of functions and uses of attributively modifying nouns in NN compounds. While in Russian the derivation of relational adjectives is lexically restricted, in some other languages essentially all nouns productively form a relational adjective. In Chukchi the standard way to modify a noun with another (inanimate) noun is to create a relational adjective using the adjectivizer -kin(e)/-ken(a) (Skorik 1961: 268-280). Such adjectives express various relations as seen in (4). (4)

emnuN lPeleN Nelw@lP@ weem

‘tundra’ ‘summer’ ‘herd’ ‘river’

emnuNkin g@nnik lPeleNkin ewirP@n Nelw@lP@kin Paacek weemkinet w@kw@t

‘tundra animal’ ‘summer clothing’ ‘youth from the herd’ ‘rocks in the river’

The relational adjective strategy is not incompatible with the compounding strategy. Chukchi also has NN compounding, but relational adjective formation is very productive in Chukchi and in a sense is the standard way to achieve modification-by-noun. To conclude, in relational adjectives the basic function of the noun category is at variance with its expected syntactic/distributional property: the modifying noun serves as an attribute, the canonical function of an adjective, not as the denotation of an argument. Indeed, such adjectives all too often retain some of their nominal morphosyntax, in particular, the ability of being modified by another adjective, see Nikolaeva and Spencer 2009a for details.

3 Overview of encoding strategies We have described briefly the major semantic types of adnominal dependents. This section presents an overview of most common encoding strategies used to express all types of possession and modification. By encoding strategy we understand the basic construction-internal morphological device (constructional marker) such as case/adpositional marking or agreement as well as the systematic lack of such, but ignore possible distinctions in the underlying constituent structure and word order. An overview of encoding strategies for possessives can be found in Koptjevskaja-Tamm (2001, 2002, 2003a); Chappell and McGregor (1996), and various other works. For attributive modification a survey of adjectival types is given in Wetzer (1996) and Bhat (1994). We expand the typology by treating possession and modification together and listing as many as is practicable of the logical possibilities for the morphological expression of the head-dependent relation within a NP. 7 Although the distinction between relational and qualitative adjectives is primarily semantic, the boundary between them is vague in part because many relational adjectives may acquire a metaphorical qualitative meaning, cf. serdeˇcnaja bolezn ‘heart disease’ and serdeˇcnyj cˇ elovek ‘warm hearted person’ (literally ‘heart person’). The former is interpreted as involving a relation between the notions ‘disease’ and ‘heart’, and the latter expresses a quality of the modified noun and does not contain a direct reference to heart.

11

We only deal with unambiguously NP-internal head-dependent relations, so we exclude external possessors, discontinuous modifiers and so-called ‘non-configurational’ languages where the notion of NP is problematic. Where the encoding only differs in inflectional class (for instance, the inflectionally distinct possessive adjectives and relational adjectives in Russian) we treat this as a single strategy. It is also important to note that we only address lexical possessors here. Possessive constructions involving pronominal possessors often employ different strategies from those involving lexical possessors, with pronominal possessors tending to induce more head marking than lexical possessors (Nichols 1986). For example, in Section 4.8 we show that lexical possessors in Northern Khanty do not normally trigger agreement on the head, but pronominal possessors in this language must be cross-referenced by agreement affixes and therefore involve a different strategy. Modification-by-noun as defined above precludes pronominal dependents by definition, therefore it only makes sense to compare it with lexical possessors. Moreover, in some syntactic frameworks pronominal possession is viewed as part of the functional structure of a nominal.

3.1 Strategies without agreement The relation between the head and the dependent can remain morphologically unexpressed or be indicated by means of an overt constructional marker, the position of which differs: it can be syntactically associated with the head or with the dependent, or it can be syntactically independent of either. This gives us the patterns in (5), where Nh stands for the nominal (possessed) head and Ndep for the dependent possessor or modifier. There is no assumption that the dependent Ndep is categorially a noun (for instance, it could be an adjective in form). For simplicity’s sake we take the order to be head-initial; for head-final constructions, just reverse the order of elements. The symbols X, Y denote realizations that express a construction-internal syntactic relation but do not specify their morphological status (i.e. whether the morpheme is an affix, clitic, phonologically unbound word, or corresponds to a systematic prosodic device). The symbol ‘=’ indicates some kind of morphological and/or syntactic binding and its absence in (5e) indicates that the element is syntactically associated with neither component of the construction (though in practice such elements tend to be prosodicially weak function words and therefore tend to be phonologically dependent with the left- or right-adjacent element, i.e. pronounced as en- or proclitics): (5) a. b. c. d. e.

head Nh Nh =X Nh Nh =X Nh

X

dependent Ndep Ndep Ndep =X Ndep =Y Ndep

The constructional markers X and Y can each take various shapes if they indicate the features of their host (see Section 3.3 below) or they can take different forms in different syntactic contexts. For example, the genitive in some Daghestanian languages has different exponents depending on the case of the head noun (absolutive vs. non-absolutive, see Koptjevskaja-Tamm 2003b). This latter type of covariation is not reflected in our schemas. The pattern represented in (5a) involves no constructional marking. We will use the neutral term ‘juxtaposition’ for it. A discussion of juxtaposition is provided below in Section 5.3. The head-marking pattern (5b) (Nh =X Ndep ) most generally calls to mind the familiar possessor agreement constructions of languages such as Hungarian. However, in the schemas in (5) we are discounting agreement as part of the encoding strategy. The head-marking pattern without agreement is instantiated therefore by the Construct State construction of Semitic languages, some Nilotic languages and others. In (6) we see examples from the Kenyan Nilotic language DhoLuo (Stafford 1967), in which 12

a typical construct state is formed by switching the voicing value of the final consonant of the root (cf. ot > od and ut > ud below8 ): (6)

base form

gloss

construct state

Nh kitab bidhi kidi sigana wer ot udi

gloss

Nh =X Ndep ‘book’ ‘spear’ ‘stone’ ‘story’ ‘song’ ‘house (SG)’ ‘house (PL)’

kitap kwano bith rec kit got sigand apwoyo wend Luo od winyo ut winyi

‘an arithmetic book’ ‘a fish spear’ ‘a stone from the hill’ ‘the story of the hare’ ‘a Luo song’ ‘a bird’s nest’ ‘birds’ nests’

In type (5c), dependent-marking possessives (Nh Ndep =X), are based on the genitive/possessive case or possessive adpositions, as in most Indo-European languages. As we illustrate later, genitives can be used for modificational purposes as well (see Koptjevskaja-Tamm 2001, 2003b,c, 2004 for extensive discussion). Many languages have derived adjectives which do not agree with the head. For example, the Hungarian privative adjectives are derived from nouns with the suffix -t(a)lan/-t(elen), e.g. ember-telen human-less. The adjectival suffix serves as a constructional marker in this case. The double marking in (5d), (Nh =X Ndep =Y), is most commonly found as a combination of strategies (5b) and (5c), as in certain types of Semitic Construct State constructions in which the possessed noun is in the construct form and the possessor is additionally marked with genitive case. According to Koptjevskaja-Tamm (2003b), Eastern Armenian has a somewhat similar kind of double marked possessive construction, in which genitive case is found on the possessor and some kind of obligatory definiteness marking appears on the possessee. The pattern represented in (5e) is instantiated by languages with a Persian-style ezafe. The ezafe marker -e in Persian is an invariable clitic attached as a phrasal affix to the right edge of the noun phrase.9 (7)

ketˆab-e an mard book-EZ that man ‘that man’s book’

The ezafe construction marks attributive modifiers, too. It is possible to have a sequence of such ezafemarked modifiers, showing fairly complex dependency structures, as in (8) (Ghomeshi 1997: 736): (8)

otˆaq-e kuchik-e zir-e shirvuni-e ali room-EZ small-EZ under-EZ root-EZ Ali ‘Ali’s small room under the roof’

The constituent structure implicit in example is shown in (9). Note in particular that the right most ezafe marker (on shirvuni) actually serves to link ali ‘Ali’ to otˆaq ‘room’. (9)

shirvuni]]-e ali [[otˆaq-e kuchik]-e [zir-e [[room-EZ small]-EZ [under-EZ roof]]-EZ ali

8 Stafford 9 For

does not provide tone markings so we are obliged to ignore them. extensive arguments in favour of this analysis see Samvelian (2007)

13

Another language which uses an ezafe-type construction, though only for attributive function, is Tagalog. In this language the principal possessive construction is expressed by marking the possessor with a pre-nominal marker ng (pronounced /naN/) for common nouns or /ni, nina/ for singular/plural proper nouns (the ‘common/proper noun’ distinction is found throughout the system of nominal markers in Tagalog). These markers are glossed ‘LKR’ (for ‘linker’): (10)

a.

lapis ng bata pencil LKR child ‘the/a child’s pencil’

b.

bahay nina Maria house LKR Maria ‘the house of Maria (and her family/group)’

This encoding is an instance of type (5c). However, attributive adjective modification and modificationby-noun use a different strategy. The modifying adjective or noun is connected to the head noun by means of a linker na/-ng. The ng form is pronounced /N/ and appears as an enclitic or phrasal affix on the rightmost constituent of the attributive phrase, provided the last word ends in a vowel, /h, P, N/, as in (11), (Schachter and Otanes 1972: 118). In all other cases the na allomorph is selected (12): (11)

a. /mabu:tih/ b. /mayu:mi/ c. /maya:man/

‘good’ ‘modest’ ‘rich’

mabuting tao mayuming tao mayaming tao

‘good person’ ‘modest person’ ‘rich person’

(12)

a. /masi:pag/ b. /maga:laN/ c. /mata:kaw/

‘diligent’ ‘courteous’ ‘greedy’

masipag na tao magalang na tao matakaw na tao

‘diligent person’ ‘courteous person’ ‘greedy person’

In contrast to the possessive construction, the order of head and dependent is not fixed here. This is equally true of attributive adjectives (13a) or modification by phrases, for instance, prepositional phrases (13b) (Schachter and Otanes 1972: 118): (13)

a. bagong libro vs. librong bago new.LKR book book.LKR new ‘new book’ b. nasa mesang libro vs. librong nasa mesa on table.LKR book book.LKR on table ‘the book on the table’

The difference in word order is related to information structure, with the new information tending to come after old information. Interestingly, this freedom of word, or rather the deployment of word order to express information structure, is also found with modification-by-noun. Schachter and Otanes (1972: 118) discuss at some length the distinction between (14a) and (14b), both of which would be translated into English as ‘toy stove’10 : (14)

a.

kalang stove.LKR

laruan toy

b. laruang toy.LKR

kalan stove

Given the fact that the na/-ng linker remains in the same linear position whatever the order of head and dependent we can conclude that it is not syntactically affiliated with either constituent. 10 These

compounds are ‘syntactic’ compounds, or productively formed compounds with compositional semantics. Tagalog also has lexical(ized) compounds in which the head noun is always to the right, and which differ morphosyntactically from the productive compounds. In these lexical compounds the -ng linker appears whenever the phonological conditions for it are met. However, there is no over linker otherwise, and Schachter and Otanes (1972: 107) regard this as indicating that the na linker simply has a zero allomorph in such contexts.

14

3.2 Strategies with agreement We understand agreement as a covariance in features between two different components of the construction, the head and the dependent. That is, using the conventions introduced in the previous subsection, if X is hosted by Nh it agrees with Ndep and vice versa. Agreement either functions as the construction marker itself, indicated here as agr, or is expressed by a separate realization in addition to X (X=agr). This yields the following basic types for single-marking patterns: (15) a. b. c. d. e.

head

dependent

Nh =agr Nh =X=agr Nh Nh Nh

Ndep Ndep Ndep =agr Ndep =X=agr Ndep

X=agr

In type (15a) the head agrees with the dependent. In the simplest case we have straightforward possessor agreement, as in Hungarian (and many other Uralic languages): (16)

Maria k¨onyv-e Maria book-3 SG . PX Ndep Nh ‘Maria’s book’

Type (15b) represents a case in which the head agrees with the dependent as in type (15a) and additionally assumes some sort of derived (or inflected) form special to the construction. This is the mirror image of the very well attested pattern shown in (15d). A head-initial pattern is found in Hoava, where the so-called ‘edible possession’ is rendered by a possessive classifier syntactically related to the head noun (but phonologically unbound). The classifier hosts possessive agreement with the possessor, see Section 4.1 for more details. In the next two types, (15c, d), the dependent agrees with the head, as indicated by the subscript on agr. This is the case of agreeing adjectives. In Russian krasiv-yj magazin ‘beautiful shop’, the non-derived gradable property adjective krasiv-yj ‘beautiful’ has no constructional marker except the agreement morpheme itself (Nh Ndep =agr). It is rare cross-linguistically for a noun to be given agreement markers that parallel those for adjectives without that noun first being turned into an adjective, but in Section 4.7 below we discuss Tundra Nenets examples which demonstrate exactly this. Type (15d), Nh Ndep =X=agr, is fairly well-represented cross-linguistically. For example, it is found in relational and possessive adjectives in Slavic languages. In Polish woda mineral-n-a ‘water mineral-ADJ - AGR’, ‘mineral water’, the suffix -n- indicates the relation between entities ‘mineral’ and ‘water’ and corresponds to X in (15d), while -a is a feminine singular attributive agreement suffix, cross-referencing the feminine gender noun woda ‘water’ and corresponding to agr. Our earlier Russian example of kniˇznyj magazin ‘book shop’ illustrates the same pattern with the opposite word order. The same strategy is sometimes used to express possessive relations: for instance, agreeing genitives in Romani, and possessive adjectives in Chukchi and some Slavic languages. In some languages the ‘X’ element is not obviously an affix, but rather is a particle, clitic or phrasal affix bearing the markers agreeing with the head, but still syntactically in construction with the dependent. Examples of such constructions are the Hindi-Urdu ‘genitive’ postposition kaa (Spencer 2005, McGregor 1995; Kachru 2006), the Bantu a-of-association (see Section 4.8), and the inflecting ‘articles’ of Albanian and Rumanian (see Koptjevskaja-Tamm 2003b: 661f). 15

In (15e) we have shown a hypothetical version of (5e) in which the syntactically unaffiliated ezafe marker agrees, either with the head or the dependent (or, presumably, with both). We have not encountered precisely this construction type, even though it appears to be a theoretical possibility. Perhaps this is simply because it is rather rare for a possessor marker to remain unaffiliated in this way. The only double marking patterns with agreement we are aware of are those in which the head (possessee) agrees with the possessor and the possessor occurs in some kind of possessive case (typically the genitive). The ‘definite’ izafet in Turkish and a number of other Turkic languages illustrates this pattern (see Section 4.5 below). Hungarian and a number of other languages have an optional variant of this type of construction in which the NP-internal possessor is marked in the dative (Hungarian lacks a genitive case).

3.3 Construction markers which register features of their host The patterns we have illustrated are, in a sense, the types of construction we would expect from combining the ingredients of invariable head or dependent marking (whether affix, adposition or whatever) with agreement. However, they do not exhaust the typological space of construction types. We sometimes find that a constructional marker registers the features of its own host. A clear example of this is provided by Hausa. In the original Hausa construction possession is indicated by means of a linking morph which is in syntactic construction with the possessor but agrees with the possessee head for gender na masculine, ta feminine, in the element order possessee - linker - possessor, (17), (18) (Kraft and Kirk-Greene 1973: 41-42, Newman 2000) (Aud`u is a woman’s name): (17)

gid¯a na Aud`u house.M . LKR . M Audu Nh Ndep ‘Audu’s house’

(18)

r`¯ıg¯a ta Aud`u gown.F. LKR . F Audu ‘Audu’s gown’

This pattern is identical constructionally to the Hindi-Urdu or Albanian ‘genitive’ and the Bantu ‘a-ofassociation’. However, the linker element can, under certain circumstances, become an enclitic, essentially suffixing to the word on the immediate left, and hence effectively marking the head possessee: (19)

Aud`u gida-n house.M .- LKR . M Audu ‘Audu’s house’

(20)

r`¯ıga-˜r Aud`u gown.F.- LKR . F Audu ‘Audu’s gown’

The r˜ allomorph in (20) represents a coronal tap/roll as opposed to /r/ which is a retroflex tap. It is only found after feminine singular nouns ending in -a; otherwise, the -n allomorph is found. This construction is now no longer an instance of the Albanian/Bantu dependent-marking agreeing type (15d), but rather represents head-marking non-agreeing pattern (5b), where the constructional marker indicates the gender 16

of the head itself. The pattern seen in (19, 20) is restricted. It is not found, for instance, when the head noun is otherwise modified, as in (21, 22): (21)

dˆok`ın nˆan na Bell`o horse this of.M Bello ‘this horse of Bello’s’

(22)

k`¯ek¯e biyu na Garb`a bicycles two of.M Garba ‘Garba’s two bicycles’

There are rather complex conditions on when the linker appears as an enclitic and when as a preposition. The linker construction can be used to signal attributive modification by an adjective, for instance, in which case the linker can only appear as the enclitic (attached to the prenominal adjectives as a repeating affix). Kurmanji Kurdish (Wurzel 1997; see also Kurdoev 1978: 70f) has a construction which is usually referrred to as ezafe, but with an importance difference from the standard Persian equivalent. The ezafe marker signals the gender/number of the possessed noun, though distinguishing gender in the singular only. Its forms are -a feminine singular, -ˆe masculine singular and -ˆen/-ˆed plural. Simple examples are shown in (23) (Wurzel 1997: 17): (23)

a.

welat-ˆe homeland.M - EZ . M . SG ‘my homeland’

mın my

b.

jın-a wife.F - EZ . F. SG ‘Hiso’s wife’

Hıso Hiso

In other respects the Kurmanji ezafe construction is much like its Persian congener. In particular, the ezafe is used to mark attributive modification as well as possession and from strings of ezafe constructions it is clear that the ezafe is a kind of edge inflection. The Kurmanji ezafe is effectively an instance of (5b) with the X redundantly signalling number/gender properties of its own host. In Aleut (Eskimo-Aleut) there are two construction-internal morphological markers in NPs (Bergsland 1997; Golovko 1997). The head of a complex NP bears a number marker (in the Absolutive singular -a, dual -kix or plural -(ng)is/-(ng)in). Golovko refers to it as ‘possessive’, as it is absent on nouns which do not take dependents and on which number is expressed differently (singular -ˆx, dual -(i)x/-ki and plural -(i)n/-(i)s). The dependent bears the so-called ‘relative’ marker, whose function is to indicate the dependent status of the phrase. The relative marker is also sensitive to the number feature of its host noun: singular and dual -(i)m, plural -s. The construction expresses both possession and (in the singular only) modification by noun e.g: (24)

a. sabaaka-m tutuusi-ki dog-REL . SG ear-POSS . DU ‘dog’s ears’ b. tayaˆgu-s hla-ngis man-REL . PL son-POSS . PL ‘men’s sons’ c. alaˆgu-m qa-a ocean.REL . SG fish-POSS . SG ‘ocean fish’ d. isuˆg-im ajaga-a seal-REL . SG female-POSS . SG ‘female seal’ 17

This is an instance of the double-marking pattern (5d), but both constructional markers vary in gender. A somewhat different case is represented by languages in which a pronominal adjective functions to link the possessor and possessum. An example from Spoken German is illustrated in (25): (25)

der Beate sein-e B¨ucher the.DAT Beate [her-N . NOM . PL book.N . NOM . PL] ‘Beate’s books’

In such constructions the possessive marker (typically a possessive pronoun/adjective, as here) agrees with the head noun in case, gender and number, just as in the Hindi, Bantu or Albanian constructions, but additionally it picks up certain features of the dependent noun. In German the possessor is additionally marked with dative case, so this is another instance of double marking combined with agreement similar to Turkish and Hungarian. For extensive discussion of the variants on this structural type in Creoles and in various Germanic languages (see Koptjevskaja-Tamm 2003b: 670f). We suspect that in this subsection we have only touched the surface of the complications that can arise from partial grammaticalization of possessive adpositions, possessive pronouns and such like, and it is possible that there are other construction types that we have not encountered or thought of. Typological study of these patterns is difficult because we are often dealing with incompletely morphologized cliticlike function words which are prone to be misdescribed and misanalysed, particularly if the grammarian has used misleading terminology for such constructions, such as ‘genitive case’. To summarize, we have outlined the principal types of encoding strategy found cross-linguistically, taking care to separate out those strategies that involve agreement from those that do not. This looks a little perverse against the backdrop of the head-/dependent marking distinction (Nichols 1986), in which there is an implied opposition between ‘head possessed noun agrees with dependent possessor’ and ‘dependent possessor noun marked with genitive case’. But we believe that it is more illuminating to factor out agreement strategies, since they can work in either direction: head agrees with dependent or dependent agrees with head (or both). As a result agreement does not really define a head/dependent marking dichotomy at all for possessive constructions.

4 Patterns of polyfunctionality In Section 2 we discussed four basic types of nominal constructions expressing possession and modification, all of which are ways of establishing some sort of relation between two denotations within an NP. The relevant construction types are as follows: (26)

A

canonical modification

C alienable possession

B

modification-by-noun

D canonical (inalienable) possession

In Section 3 we have presented the morphological strategies that can in principle encode these relations in the languages of the world. This section is meant to illustrate that at a very gross level of analysis, the same encoding strategies can be used for more than one function. It is important to notice that identity of surface strategies as defined above does not guarantee identity of syntax. For instance, Nikolaeva (2002) argues that the possession/modification-by-noun distinction in Northern Khanty is reflected only in constituent structure: although the possessor and the modifying noun do not differ in terms of morphological marking they occupy different structural positions, the latter being more closely associated with the head. Similarly, Jokinen (1991) and Christen (2001) show that in Finnish non-referential genitives must be adjacent to the head (27b), while for referential possessive 18

genitives this requirement does not hold. The referential genitives are syntactic specifiers and can be separated from the head by modificational elements (27a): (27)

a. kaupungin aktiiviset asukkaat town.GEN active.PL people.PL ‘the active people in/of the town’ b. aktiiviset kaupungin asukkaat active.PL town.GEN people.PL ‘(the) active town people’ (Christen 2001: 513)

However, we have abstracted away from word order and from syntactic structure generally in our typological survey. This is not because we do not regard word order as important (though there are no doubt languages in which word order never plays a significant role in encoding any of the constructions we discuss). Rather, the justification is practical. The descriptive materials at our disposal are likely to provide us with reasonably secure information about morphology but it’s much rarer to find reliable information about syntactic structure beyond remarks about typical word orders. At the same time, even for languages which have a long tradition of syntactic description, it is very difficult to know what we would be describing. For instance, a sufficiently careful survey of the literature in generative grammar on the English ’s construction would reveal analyses which treat it as a genitive case marker, as a marker assigning (abstract) genitive case, as a determiner, as a possessive functional head (with various different labels), and so on. Were we to try to compare the syntax of ’s-marked elements with, say, attributive adjectives we would be hampered by the fact that there is virtually no discussion of the syntax of English attributive adjectives in any popular generative framework.11 For these reasons we ignore the purely syntactic aspects of possession/modification until such time as it becomes possible to say something substantive about the issue. It is, of course, quite possible that none of our conclusions will be valid once syntax is properly taken into account, but for the present we see no practicable alternative. There is one very simple syntactic property, however, which we will wish to say something about, and that is the property of simple juxtaposition, which we discuss in more detail in Section 5.3. We do not discuss marginal strategies that only apply to a certain subset of lexical items. Thus, the notion of encoding strategy as understood here neutralizes lexical restrictions and other subtle distinctions of interpretation, as well as differences in syntactic constituency. The following patterns of polyfunctionality have been found so far (the signs ‘=’ and ‘6=’ mean ‘the same strategy’ and ‘different strategy’, respectively. The list of relevant languages is not, of course, exhaustive). (28)

A6 = B6 = C6 = D A = B6 = C6 = D A = B = C6 = D A6 = B = C6 = D A6 = B 6 = C = D A 6= B = C = D A = B 6= C = D A=B=C=D

(Lele, Tuvaluan, Hoava) (Maori, Samoani) ? (Maltese Arabic, Miya, Yamphu) (Kolyma Yukaghir, Turkish, Russian) (Finnish, Udihe, Swedish, Hausa) (Tundra Nenets, Malagasy, Taleshi, T¨umpisa Shoshone, Hungarian, English, Tagalog) (Persian, Tajik, Hindi, Albanian, Northern Khanty, Indonesian, Bantu, Chukchi)

These patterns are illustrated below. 11 Informal confirmation of this claim can be found by taking any recent textbook of generative grammar and looking up ‘adjective’ in the index.

19

4.1 A 6= B 6= C 6= D Languages of this type crucially employ encoding strategies which distinguish between all four semantic functions of adnominal dependents. Lele (NigerKongo, data from Frajzingier 2001) exhibits the opposition alienable/inalienable possession. The inalienable construction is head marked: the head noun follows the dependent and bears possessive agreement that indicates the gender of the possessor (29a, 29b). In alienable possession the word order is reversed: the dependent precedes the head. The dependent is followed by the genitival ‘linker’ kV agreeing with the head in gender (29c). (29)

a. k`urmb`alo c`ay chief head3 M ‘chief’s head’ b. k`urmb`alo kamday chief wife.PL 3 M ‘chief’s wives’ c. gr`a c`anig´e k`ey dog Canige GEN .3 M ‘Canige’s dog’

In contrast, nominal modification is achieved by simple juxtaposition of two nouns involving no agreement and no linking elements, as shown in (30): (30)

a. k`ar`a t´ug´u people village ‘village people’ b. w`el`e k`as`ug`u day market ‘market day (= Sunday)’ c. Godu tam´a monkey woman ‘monkey woman’

Property-denoting words used as attributive modifies take one of two forms. Either we see a postnominal attribute agreeing in number and gender with the head (31a, 31b), or we see a postnominal non-agreeing verbal form (a deverbal noun) (31c): (31)

a. tam´a t`unya woman(F) tall.F ‘tall woman’ b. kama j`alwa waters cold.PL ‘cold waters’ c. k`ulb´a b´or´e cow be.white.VN ‘white cow’

Like many other Austronesian languages, Tuvaluan has two possessive constructions, the so-called O-construction and the A-construction. Languages seem to differ as to the conditions of their use, but 20

according to Besnier (2000) in Tuvaluan the A/O distinction corresponds to the alienability opposition and is lexically determined. A-possession counts as alienable, while inalienably owned nouns require the O-construction. The latter include parts in part/whole relationships (e.g. body parts), words denoting inherent properties, emotions and sensations, inherent objects acquired through inheritance (clothing, canoes, homes, land), as well as things and persons that are intimately connected with the possessor (e.g. kinship terms). (32)

a. te ato o te fale the roof O the house ‘the roof of the house’ b. te tala a Evotia the story A Evotia ‘Evotia’s story’

Most nouns unambiguously belong to one or the other class, but sometimes the choice of a class depends on the argument structure of the noun or a context. Jackson and Jackson (1999: 27–28) mention that even words that normally belong to the O-class can participate in the A-construction if the relation between the ‘possessor’ and the ‘possessee’ is determined contextually rather than lexically. For instance, ‘my house’ will normally be rendered via O-possession, as in most cases this expression refers to the house which the speaker owns. However, if it means the house which the speaker does not own but is assigned to clean, the A-construction will be used. The point is that the semantic relation between the possessor and possessed noun in O-construction is established by the meaning of the latter, while in the A-construction it is not predetermined by the argument structure of its components. Modification-by-noun in Tuvaluan is achieved by what Jackson and Jackson’s grammar refers to as ‘nominal adjectives’. Nominal adjectives denote non-gradable properties, and they do not participate in comparative constructions or in expressing the degree of a quality. They do not agree with the head in number: (33)

a. te tifa fatu the plate stone ‘the porcelain plate’ b. tifa fatu plate stone ‘porcelain plates’

On the other hand, canonical modification is expressed by ‘verbal’ adjectives. These denote gradable property concepts and behave in some ways like verbs. When used as modifiers, they agree in number: (34)

a. se tagata valea a man ignorant ‘an ignorant man’ b. ne taagata vaallea some man.PL ignorant.PL ‘(some) ignorant men’

Another Austronesian language, Hoava, has four possessive classes depending on the meaning of the head noun (Davis 2003). Inalienable classes are as follows: (i) with part/whole, inherent properties and most family relations possessive agreement is required on the head noun; (ii) so-called ‘edible’ possession is expressed by an agreeing classifier; (iii) ‘exclusive’ possession denotes exclusive rights 21

not shared by other people and is expressed by a possessive pronoun preceding the possessed. All other nominals count as alienable. Alienable possessives involve the possessive preposition ta, te, tana, which may host agreement. These types are exemplified in (35a, b, c, d), respectively: (35)

a. sa

belena sa boko tail.3 SG ART pig ‘pig’s tail’ b. ana napo sa koburu CLASS .3 SG drink ART child ‘child’s drink’ c. nana siki Jakia 3 SG . POSS dog Jakia ‘Jakia’s dog’ d. sa hore te Iani ART canoe of Iani ‘Iani’s canoe’ ART

As can be seen in (35), all four possessive constructions contain some sort of construction-internal marker, although its nature differs. In contrast, modification-by-noun is expressed by ‘compounds’, i.e. juxtaposition where the dependent follows the head. As in English, compounds are fully productive. Although their meaning may be idiomatic, this is not necessarily the case. Some examples are nikana vaka ‘European man (literally: ship man)’, kabasa raro ‘kitchen (literally: house of pots)’, kabasa hinigala ‘garden house’ and kaha qato ‘tree bark (literally: tree skin)’. Words expressing gradable properties are verbal in nature. When used as modifiers, they agree in number with the head: (36)

ria tavete lavatidi ART works< NOM > be.big3PL ‘the major works’

Note that this kind of agreement is also possible on a small closed class of modifiers derived from nouns by reduplication, but the majority of nominal modifiers do not agree.

4.2 A = B 6= C 6= D Maori is another Austronesian language with O- and A-possession. According to Bauer (1997), these two categories express the nature of the relation between the possessor and the possessee. The Aconstruction is used when the possessor exercises some dominance or control over the possessee, and the O-construction is used otherwise. From this point of view the O-relation is unmarked. Most items only occur with one type of possession because of their intrinsic semantic properties. The A-class includes the items which the possessor acquires in his/her lifetime (e.g. wife, husband, children, uninherited objects) and personal property such as small portable objects and food. The O-class includes clothing, houses, means of transport, furniture, body parts and emotions, and, for some speakers, water. However, Bauer notices that this distinction is rather vague in the sense that some nouns may occur in both constructions depending on the situation and the construal of the relationship between two nouns by the speaker. In addition, there seem to be some variations amongst native speakers in the treatment of new items acquired by the Maori culture. (37)

a. te rongoa a Pou the medicine A Pou ‘Pou’s medicine (which he made)’ 22

b. te rongoa o Pou the medicine O Pou ‘Pou’s medicine (for him to take)’ Although the distinction between two types of possessives deviates from the standard content of the alienable/inalienable opposition and is described in terms of ‘dominance’ or ‘control’ rather than inalienablity per se, at its core the possessive classification is still lexically based. There is a default or open class and a specified class of nouns that systematically fall into the closed A-category. For these nouns the possessive relation can only have one interpretation. Presumably this property must be represented in their argument structure. Modification is by postnominal attribute. There is no agreement or any other construction-internal marker. The two main types of modifiers are illustrated below: in (38) we show modification-by-noun and (39) demonstrates a property word equivalent in meaning to a European adjective. (38)

(39)

a. te huruhuru manu the feather bird ‘the bird feather’ b. te tamariki ariki the children chief ‘the children of chiefly descent’ c. te whakap¯u ahi the siren fire ‘the fire siren’ w¯ahi pai place good ‘good place’

Property words are actually stative verbs (‘state intransitives’, in Bauer’s terminology) and in nonattributive use exhibit verbal properties. Some stative intransitives undergo reduplication to realize a property that Bauer refers to as ‘distributiveness’, i.e. the attribution of the property to every member of the group. Distributive/plural reduplication does not seem to be obligatory, since some speakers accept non-reduplicated forms modifying plural nouns. In any case, it is only observed in a small subgroup of property words; most of them do not show number agreement, just like in Hoava mentioned in the previous subsection. The situation in Samoan as described in Mosel and Hovdhaugen (1992) is basically the same.

4.3 A = B = C 6= D This seems to be a rare pattern and we do not have clear examples. Presumably part of the reason for this is that when alienable and inalienable possession are distinct, inalienable possession is often expressed by simple juxtaposition. This means that the language would have to use, say, an adjectival strategy with agreement or a case/adpositional construction for all the other cases. Nonetheless, it is difficult to see why such a situation has not come about more often.

4.4 A 6= B = C 6= D In this pattern modification-by-noun is formally indistinguishable from alienable possession, but inalienable possession and canonical modification are expressed by other means. In Maltese Arabic (Semitic) 23

there are two possessive constructions (Koptjevskaja-Tamm 1997). Inalienable possession requires the Construct State: the possessed noun is followed by the definite form of the possessor and the head noun does not host its own determiner. Alienable possession is expressed by the ‘analytical genitive’, where the possessor is preceded by the genitive preposition and the head noun can host the definite article. These two options are illustrated in (40): (40)

a.

bin issult¯an son DEF.king ‘the king’s son’

issi˙gg˙ u ta’ DEF.chair of ‘Peter’s chair’

b.

Pietru Peter

The Construct State is mainly used with head nouns referring to kinship relations and body parts, i.e. it instantiates the two most typical inalienable meanings. The meaning of the genitival construction is very broad. It is not limited to alienable possession exemplified in (40b), where some kind of ownership relation obtains between the possessor and the possessee and the possessor is a referential expression, it is also employed to encode various qualitative meanings studied in Koptjevskaja-Tamm’s (1997) paper, for example the relation of material, purpose, quality, time and so on. In such cases the dependent noun is non-referential and, unlike the Construct State possessor, it does not have to host the definite article: (41)

a.

gèajnejn ta’ eye.DU of ‘hawk eyes’

c.

vja˙gg˙ ta’ sagètejn journey of hour.DU ‘a two hours’ journey’

serq hawk

b.

kittieb ta’ talent writer of talent ‘a writer of big talent’

kbir big

Adjectives are right-adjacent to the head noun and agree with it in definiteness: (42)

issi˙gg˙ u z˙ z˙ gèir DEF.chair DEF.little ‘the little chair’

Definiteness agreement distinguishes the adjectival construction from the Construct State, as in the Construct State the head cannot host the definite article. Miya (Chadic) opposes the so-called ‘direct’ and ‘indirect’ genitives (Schuh 1998: 246f). Direct genitives indicate inalienable possession and involve tonal alternations but no agreement-sensitive linking element. (43)

a.

b´aa V`aziya father Vaziya ‘Vaziya’s father’

b.

a´ t´ın laah@ nose jackal ‘jackal’s nose’

Indirect or ‘linked’ genitives have the following structure: ‘head noun - agreeing linker - dependent noun’. The linker seems to be syntactically attached to the possessor. This construction is ambiguous between alienable possession and modificational interpretations, as illustrated in (44) and (45), respectively. (44)

a.

mb`@rgu na ram LKR . M ‘Vaziya’s ram

Vaziya Vaziya

t`@makw´ıy niy Vaziya sheep.PL LKR . PL Vaziya ‘Vaziya’s sheep’

b.

24

(45)

a.

dl´ang@r naak´am animal LKR . M .house ‘domestic animal’

b.

ts`@rak`@ standing ‘midday’

taam´uku LKR . F.sun

In these examples variations in the form of the masculine singular linker na/n are determined phonologically by the quality of the following segment. Attributive adjectives either follow or precede the head. Postnominal adjectives show agreement in gender and number, while prenominal adjectives are invariable. (46)

a.

s´aaboo nd`uwul new pot ‘the/a new pot’

b.

nd`uwul saaboona pot new.M ‘the/a new pot’

The prenominal construction is rather limited in usage and is judged by the author of the description as a marginal strategy. In Yamphu (Tibeto-Burman), as described in Rutgers (1998), the genitive in mi(n) is used for inalienable possession, although it is not entirely clear what counts as inalienable in this language. The genitive does not agree. (47)

a.

Chatra;ba;smi egeyo Chatrabasmi.GEN this.side ‘this side of Chatra’

b.

nambajimi jimma father-in-law.NON . SG . GEN land ‘father-in-laws’ land’ (pp. 62-63)

Other types of adnominal dependent host an attributivizing morpheme referred to as attributive ‘nominaliser’ by the author of the description (Rutgers 1998: 86f), but we will simply gloss it as ATTR. The attributivizer agrees with the head in number: m(a) in the singular and h(a) in the plural. It is used on attributive adjectives, although in this case it is optional. This exemplifies agreeing strategy (15c). (48)

a.

utthri;ma white.SG . ATTR ‘white pig’

akma pig

b.

utthri;ha white.PL . ATTR ‘white flowers’

uN flowers

Attributivizers are also used on nouns if they combine with postpositions and oblique cases. These elements express various relationships between the head and dependent noun, for example, the similative and the sociative. (49)

hæN a. pasadokma child.SIM . SG . ATTR you ‘you who are like a child’ yamiji b. akoknuNha load.SOC . PL . ATTR person.NON . SG ‘people with a load, load people’

In addition, the plural attributivizer is present on the so-called ‘possessive’ case in -ææ (the singular attributivizer is omitted in this instance). The possessive case expresses “a certain belonging together” (Rutgers 1998: 70), which can be understood as some kind of possessive relation but “the nature of the relationship need not be so strict”. This corresponds to our understanding of semantically undetermined alienable possession.

25

(50)

a.

Sittambæ khimbe Sittambæ.POSS house.LOC ‘the house of Sittamb’

b.

ma;ksæ gotthabe bear.POSS shed.LOC ‘the shed of the bear’

We would argue that examples (49a) and (50), i.e. nominal modification and alienable possession, exemplify the same strategy, namely, type (15d) in our classification. In this strategy the dependent agrees with the head and hosts an additional construction-internal marker. In Yamphu it corresponds to a case marker that specifies the exact nature of the semantic relationship between the head and the dependent. In our examples it is similative, sociative or possessive. The point is that this strategy differs both from non-agreeing inalienable genitive and canonical modification, where construction-internal marker other than agreement is absent.

4.5 A6=B6=C=D In this pattern three distinct strategies are used to render the four relevant semantic types, due to the fact that the alienablity opposition is absent. Kolyma Yukaghir (Uralo-Yukaghir) has a small closed class of basic adjectives which include about half a dozen items such as ‘round’, ‘long’ and ‘new’, but they are mostly used in frozen collocations (Maslova 2003; Nikolaeva 2005). The overwhelming majority of property words belong to the class of verbs. They have the full finite paradigm and most non-finite forms, and occur in all or most syntactic environments where non-qualitative verbs can occur. Canonical modification is expressed by attributive verbal forms, i.e. participles. Thus, the modifying constructions formally look like relative clauses, although their clausal status is rather questionable. Examples (51) show a regular intransitive verb jaqte ‘to sing’ and a ‘qualitative’ verb lige ‘to be old’ in the modifying function: (51)

a. jaqte-je terike sing-PART woman ‘a singing woman, the woman who is singing’ b. lige-je terike old-PART woman ‘an old woman’

Modification-by-noun is achieved by what was referred to as the ‘attributive form of nouns’ in Maslova (2003). This might be a misnomer given that modification is not its only function: it is also used on objects of postpositions and sometimes on objects in non-finite clauses. However it is true that in modern Kolyma Yukaghir it is mainly used for nominal modification. The dependent is non-referential and the relationship between the head and the dependent is one of the typical ‘non-anchoring’ relationships such as material, species, kind and the like (Maslova 2003: 116–118): (52)

a.

sˇa:-n qanmujaje tree-ATTR spoon ‘wooden spoon’

b.

jaqa-n pajpe Yakut-ATTR woman ‘Yakut woman’

c.

touke-n mi:d’i: dog-ATTR sledge ‘dog sledge’

d.

mure-d igeje shoe-ATTR rope ‘shoe lace’

Alienable and inalienable possession are not formally distinguished. Both relations are encoded by juxtaposition of the possessor and possessed:

26

(53)

a.

met terike aNd’e I woman eye ‘my wife’s eye(s)’

b.

ord’¯ol lebie sˇoromo middle land man ‘man of/from the middle earth’

Both alienably and inalienably possessed nouns can host a possessive marker (gi in the Nominative or de in oblique cases), cf.: (54)

a. q¯aq¯a num¨o-gi bear house-3 ‘the den of the bear’

b. taN sˇo¨ jl’bul iri-de-ge that mouse belly-3-LOC ‘in the belly of that mouse’

This marking becomes obligatory if the possessor is not overtly present in the same NP, but is optional when it is overt and seems to depend on the discourse prominence of the possessor. In Turkish (Turkic) attributive adjectives precede the head and show no agreement, e.g. y¨uksek a˘gac¸ ‘tall tree’ (G¨oksel and Kerslake 2005). Nominal dependents participate in two types of izafet constructions: so-called definite and indefinite izafet. Definite izafet is a double-marking construction, while indefinite izafet is head-marking. Definite izafet expresses both alienable and inalienable possession, while indefinite izafet involves non-referential dependents and renders various kinds of semantic relations between two nominals, cf. examples (55) and (56) from Spencer (1991: 316) and G¨oksel and Kerslake (2005: 103f): (55)

a. Fatma-nın ev-i Fatma-GEN house-3 ‘Fatma’s house’

b. Ali-nın o˘gl-u Ali-GEN son-3 ‘Ali’s son’

(56)

a. c¸ay barda˘g-ı tea glass-3 ‘tea glass’

b. otob¨us bilet-ler-i bus ticket-PL-3 ‘bus tickets’

c.

g¨oz hastalık-lar-ı hastane-si eye disease-PL-3 SG hospital-3 SG ‘hospital for eye diseases’

There are also minor strategies of nominal modification: the relationship of material and gender, as well as some nationalities and names may be expressed by juxtaposition, just like for adjectives (Boeder and Schroeder 2000), e.g. kadın doktor ‘female doctor’ and kız arkadas¸ ‘girlfriend’. Another language that belongs to this type is Russian (Indo-European). In Russian possession is expressed by a non-agreeing genitive. In Section 2.3 we cited derived relational adjectives, which express nominal modification and agree with the head just like qualitative adjectives. The difference between them lies in the fact that relational adjectives represent strategy (15d), i.e. Nh Ndep =X-agr, whereas non-relational qualitative adjectives exemplify (15c), i.e. Nh Ndep -agr.

4.6 A 6= B = C = D In these languages the standard possessive construction is ambiguous: it is used both for possession and nominal modification. However, it contrasts sharply with adjectival modification. One example is Finnish (Uralic) where the class of adjectives is fairly clearly distinguished from the class of nouns, in particular because adjectives exhibit attributive concord in case and number. Examples (57) below are from Sulkala and Karjalainen (1992) and Christen (2001). 27

(57)

a.

pieni-ss¨a talo-i-ssa small.PL - INESS house-PL - INESS ‘in the small houses’

b.

valkea-lla auto-lla white-ADESS car-ADESS ‘on the white car’

Nominal dependents do not show agreement but host the genitive -n. The referential genitive expresses alienable and inalienable possession (58), but non-referential genitives are used for modificational purposes. (58)

a.

(59)

tyt¨o-n koira girl-GEN dog ‘girl’s dog’

b.

tyt¨o-n k¨asi girl-GEN hand ‘girl’s hand’

naisen k¨asiala woman.GEN handwriting ‘woman’s handwriting’

As mentioned above, referential and non-referential genitives have different positions within the NP, but we treat them as representing the same encoding strategy. Some modificational relationships in Finnish such as the relation of material are expressed by juxtaposition rather than the genitive construction, e.g. kultasormus (= kulta ‘gold’, sormus ‘ring’) ‘gold ring’ (Christen 2001). An example of a headmarking language which exhibits the same pattern is Udihe/Udeghe (Tungusic), (Nikolaeva and Tolskaya 2001) The possessive construction in Udihe is head-final; the possessor stands in the nominative and must be crossreferenced by person/number agreement on the head. The 3rd person singular possessive affix is -ni. The construction conveys the usual range of possessive meanings, both inalienable (kin and social relation, part/whole relation and inherent property) and alienable (legal ownership, disposal, location and so on). (60)

a.

giuse ule:-ni roe flesh-3 SG ‘roe’s flesh’

b.

mama tege-ni grandmother gown-3 SG ‘grandmother’s gown’

In addition, the possessive construction serves to express nominal modification, which, unlike in true possessives, usually involves non-referential dependents. This makes it ambiguous between two readings, cf.: (61)

keige sitani in’ei dilini xoto skolani niNka sexini

modification

possession

‘kitten’ ‘dog head’ ‘city school’ ‘Chinese fabric’

‘cat’s young’ ‘dog’s head’ ‘the school in/of the city’ ‘fabric of a/the Chinese person’

Although adjectives overlap with nouns in certain respects, they can be unambiguously identified by a number of morphological and syntactic criteria. Adjectivehood tests relevant for Tungusic are explored in Nikolaeva (2008). In short, adjectives and nouns have similar inflectional properties, but differ in their syntactic behaviour: unlike nouns, adjectives can function as secondary predicates. They can also assume the head position within an NP. Udihe does not have attributive agreement NP-internally, but when adjectives are located discontinuously to the head, they agree with it in case and sometimes number. This indicates that Udihe has a category of adjective. Adjectives precede the head without inducing possessive marking on it. Qualitative adjectives have meanings associated with canonical adjectives

28

in other languages such as form, size, colour, age, human propensity and so on, e.g. sagdi zugdi ‘big house’, imexi mo: ‘new firewood’ and ge: m¨ana ‘bad flour’. It is worth mentioning that Udihe also has relational adjectives fairly productively derived from nouns with the suffix ma/me/mo. They show the same syntactic properties as non-derived adjectives and most typically express material, e.g. aisi-me ‘golden’ < aisi ‘gold’. In some instances, however, they have other meanings such as ‘resembling X, looking like X’ (miki-me kuliga ‘a snake resembling an adder’) and ‘being X’ (g’ai-ma anda ‘the crow-friend’) where X is the base noun, as well as temporal relations (teuze-me zugdi ‘winter house’, i.e. the house used in winter) and some kind of loose association between two concepts which is difficult to characterize in precise terms (bo:boi-me olondo ‘miraculous ginseng’). This makes it difficult to classify Udihe unambiguously, as it also shows properties of the A = B 6= C = D type addressed in the following subsection. However, similar splits occur in other languages: nominal modification often employs several strategies, as noted above for Turkish, Miya and Finnish. Swedish (Germanic) is basically like Finnish since genitives render various modificational meanings (Koptjevskaja-Tamm 2003c), but it also has Germanic-type compounding. An intriguing instance of this pattern is Hausa. Modification of a noun by a noun in any capacity is accomplished by means of the linker, the -n/ r˜ head-marking suffix/enclitic exemplified in Section 3.3 (though in possession constructions this alternates with a dependent-marking preposition na/ ta). The -n/ r˜ linker redundantly realizes the number and gender of its own host (rather like the Kurmanji ezafe). Now exactly the same marker, with the same allomorphy, is used to signal attributive modification. Attributes are normally prenominal in Hausa and yet the linker still marks the leftmost element(s) in the construction, namely, the adjective(s). As a result, it is the dependent, not the head, that is marked by the suffixed linker in attributive constructions, though the linker still agrees with the head, just as in the case of possession and modification-by-noun. In some respects, therefore, this construction represents the A=B=C=D pattern, but this is obscured by the fact that (i) the linker is effectively a second position enclitic/suffix and (ii) it is the attributive adjective, not the head noun, that has to appear in leftmost position.

4.7 A = B6= C = D This type involves two major strategies, one for possession and another for modification, and is remarkably widespread. For example, in Taleshi (Iranian) alienable and inalienable possession is expressed by the number-dependent ‘oblique’ marker on the dependent noun.12 (62)

a.

Huseyn-i Huseyn-OBL . SG ‘Huseyn’s house’

ka house

c.

palang-i pust panther-OBL . SG skin ‘the skin of the panther’

b.

merdak-un men-OBL . PL ‘men’s house’

ka house

Other semantic types employ the uninflectable linking clitic -a. In (63) we show examples of nominal modification expressing various semantic relations, whereas (64) shows attributive modification: (63)

12 The

a.

palang-a pust panther-LKR skin ‘panther skin’

b.

s@ng-a ka stone-LKR house ‘stone house’

Taleshi data are obtained from Gerardo De Caro.

29

c.

ner-a palang male-LKR panther ‘male panther’

(64)

a.

xub-a aˆ dam good-LKR man ‘a/the good man’

b.

qaˇsang-a k@la beautiful-LKR girl ‘a/the beautiful girl’

In other languages modifying nouns are not turned into adjectives by an overt derivational process, but show canonical adjectival properties such as attributive agreement. For instance, in Tundra Nenets (Uralic) the possessive construction with lexical possessors is dependent-marked: the possessor takes the genitive expressed by -h in the singular and -q in the plural on non-possessed nouns, or a variety of other affixes on possessed nouns. The possessive construction renders the usual alienable and inalienable meanings. (65)

a.

ny´ıbya-h mal° needle-GEN end ‘the end of the needle’

b.

puxacya-h pad°ko woman-GEN bag ‘the woman’s bag’

In contrast, qualitative and nominal modification is expressed by prenominal juxtaposition. The semantic difference between these two constructions has been studied in detail by Tereˇscˇ enko (1967, 1973: 219ff) and Tereˇscˇ enko (1967, 1973: 219ff.) and Nikolaeva (2002), where it is argued that the dependent genitive expresses the relationship between two referents, while the dependent nominative is non-referential and expresses a property. Consider the following contrast: (66)

a. ti-h ya reindeer-GEN soup ‘reindeers’ soup, soup for the reindeer’ b. t´ı ya reindeer soup ‘reindeer soup (soup made of reindeer meat)’

Other examples of nominal modification are sax°r xidya ‘sugar cup, cup used for sugar’ and myercya pyi ‘windy night’. The dependent nominative shows optional number agreement with the head. In this it does not differ from qualitative adjectives, cf.: (67)

a.

nyarawa-q loNkey°-q copper-PL button-PL ‘copper buttons’

b.

serako-q loNkey°-q white-PL button-PL ‘white buttons’

A roughly similar pattern obtains in a Numic language, T¨umpisa (Panamint) Shoshone, described by Dayley (1989), where genitives do not agree, whereas adjectives and modificational nouns show case agreement with the head. Malagasy (Austronesian) has no attributive agreement.13 The postnominal genitive is indicated by the definite article, which normally triggers phonological changes in the head noun and expresses alienable and inalienable possession. (68)

a.

ny

fakan’ ny root. EP DEF ‘the root of the tree’ DEF

13 The

hazo tree

b.

ny

i

DEF

alikan’ dog. EP ‘Jeanne’s dog’

DEF

authors are indebted to Charles Randriamasimanana for the Malagasy data.

30

Jeanne Jeanne

In nominal modification the dependent is non-referential and does not take an article. This construction expresses various modificational relations between two nouns such as material, gender, nationality, measure, some kind of association, quality, and the like. (69)

a.

ny

trano vato house stone ‘the stone house’

b.

alika lahy dog male ‘a male dog’

teny anglisy language English ‘English language’

d.

akanjona mpamosavy clothes witch ‘witch clothes’

DEF

c.

iray one

Unlike nouns, adjectives never occur in the genitive. In modificational function they follow the head noun and do not agree. (70)

a.

trano house good

tsara good house

b.

ny

lamba DEF clothes the white

fotsy white clothes

Modifying adjectives precede the genitive and are not compatible with the relative pronoun izay ‘which’. In this they contrast with modifying verbs (relative clauses), which must follow the genitive, as seen in (71): (71)

Jeanne a. ny rona (*izay) petak’ i def nose which flat DEF.GEN Jeanne ‘Jeanne’s flat nose’ ndRakoto (izay) novidiko b. ilay satroka fotsy that hat white.GEN Koto which bought.1 SG ‘the white hat of Koto that I bought’

The null copula relative clause with an adjective as a predicate also follows the genitive, cf.: (72)

hitako ny oron’ i Jeanne izay petaka I.saw DEF nose.GEN DEF Jeanne which flat ‘the nose of Jeanne, which is flat’

This indicates that adjectives form a separate class, distinct both from nouns and verbs. Other languages that belong to this type are Hungarian, English and Tagalog.

4.8 A = B = C = D Finally, the four relevant semantic types may show identical encoding, although the actual strategies vary. We mentioned in Section 3.1 that the ezafe construction in Persian marks possession and attributive modification by adjectives (and sometimes by other types of phrase, such as prepositional phrases). We can conveniently refer to constructions of this sort the ‘ezafe family’ of constructions. For example, in Tajik ezafe is used for all our relevant types of adnominals. Examples (73 - 75) from Perry (2005) show attributive modification, modification-by-noun and alienable/inalienable possession. (73)

a.

duxtar-i zebo girl-EZ beautiful ‘the beautiful girl’

b. habo-i na˘ga weather-EZ nice ‘nice weather’

31

(74)

a.

dandon-i tooth-EZ ‘gold tooth’

tillo gold

(75)

a.

duxtar-i mard daughter-EZ man ‘the man’s daughter’

b. safar-i Buxoro trip-EZ Bukhara ‘the Bukhara trip’

c.

kurta-i xob shirt-EZ night ‘nightshirt’

b. xalta-i piramard-ro sack-EZ old.man-OBJ ‘the old man’s sack’

The pattern represented by Tajik is also characteristic of many other Iranian languages related to Persian. The construction-internal relation is signalled by an invariant marker which either is not attached to either constituent or is encliticized to the right edge of the leftmost member of the construction (which typically means the head). The neutral strategy is exemplified by Northern Khanty/Ostyak (Uralic) as described in Nikolaeva (1999). In this language attributive adjectives are morphologically unmarked and do not agree with the head. (76)

jam xo:t-@t-na good house-PL - LOC ‘in the good houses’

The juxtaposition of two nouns renders inalienable (77a) or alienable (77b) possession. But the same construction may have a non-possessive modificational meaning (77c): (77)

a.

e:wi se:m girl eye ‘girl’s eye(s)’

b.

e:wi la:ra´s girl box ‘girl’s box’

c.

n´ a´n la:ra´s bread box ‘box for bread, bread box’

Nominal modifiers serve for qualification of the head noun by reference to its material, purpose/function, quality, origin and the like, e.g. u:r xo:t ‘forest tent, tent in a forest’, niN n´ a:wre:m ‘daughter (literally: female child)’ and naN po:r@x ‘larch tree stump’. Like attributive adjectives, modifying nouns do not agree with the head and do not take determiners, but unlike adjectives they can be modified by an adjective and are recursive, for example kala:N sax sax ‘coat made of reindeer skin (literally: reindeer skin coat)’. We can therefore expect that at least in some instances the juxtapositional construction would be ambiguous between the possessive and non-possessive (modificational) readings. This prediction turns out to be correct: kala:N se:m can mean both ‘reindeer’s eyes, eyes of the/a reindeer’ and ‘reindeer eyes’. The semantics of the construction strongly depends, of course, on the meaning of its components. Possessors tend to be animate and often human, while non-possessive modifiers tend to be inanimate and are often mass or abstract nouns. Inanimate concrete nouns are likely to serve in both functions, and this is exactly the area in which ambiguity is the most plausible. The data cited above raises the question of word classes, i. e. whether Khanty has a distinct class of adjectives. Indeed, in null-headed phrases adjectives take nominal morphology and are not distinguishable from nouns as far as their inflectional properties is concerned. However, the syntactic distribution of adjectives differs from that of nouns in several respects: unlike nouns, adjectives function as resultative secondary predicates and can modify a verb. Most importantly, the 3rd person possessor nouns optionally trigger possessive agreement on the head, but this is totally impossible for adjectives or modifying nouns, for that matter, cf.:14 14 Agreement

is conditioned by the discourse prominence (topicality) of the possessor, pretty much as in Kolyma Yukaghir.

32

(78)

a.

e:wi la:ra´s-@l girl box-3SG ‘girl’s box’

b. n´ a´n la:ra´s-@l c. jam la:ra´s-@l bread box-3SG good box-3SG 6= ‘box for bread, bread box’ 6= ‘good box’

The expressions in (78b) and (78c) are in principle grammatical but only in the meaning ‘his/her box for bread’ and ‘his/her good box’, respectively. Another example of a language with neutral marking in NPs is standard Indonesian, where the situation is exactly like in Khanty except that word order is its mirror image: the phrase is head-initial (Sneddon 1996). Many languages of the Bantu group show essentially the same pattern in the so-called a-ofrelationship or a-of-association. A type of dependent marking is demonstrated by Tswana, in which the a-of-relationship construction has become morphologized, with the -a element becoming a prefix to the possessor. According to Cole (1979: 159ff), possessives are formed by prefixing possessive affixes to the dependent noun. The affixes indicate agreement with the head in class and sometimes also number. The class of the head may also be shown by a prefix; in (79a) the head noun ‘dog’ is class 5 and in (79b) the head noun ‘wife’ is class 1: (79)

a.

mo-sadi wa-motˇsomi CL5-dog CL5-hunter ‘hunter’s dog’

b.

n-tˇsa ya-motˇsomi CL1-wife CL1-hunter ‘hunter’s wife’

The same strategy serves in so-called ‘descriptive possessives’, which “denote some quality, function, feature of other characteristic of the antecedent” (Cole 1979: 167). Examples below show that descriptive possessives correspond to what we refer to as modification-by-noun and express various semantic relations between the head and the modifying noun. (80)

a.

n-tlo ya-ditene CL5-house CL5-brick ‘brick house’

b.

n-kgˆo ya-bojalwa CL5-pot CL5-beer ‘a pot of beer’

c.

n-tlo ya-thapˆelˆo CL5-house CL5-prayer ‘a house for prayer, church’

d.

di-tlhako tsa-senna PL-shoe CL 4. PL-man ‘men’s shoes’

Property words also agree in class and number, but employ a different set of concord affixes. For example, in class 5 the singular prefix is ya for possessors and eˆ N for property words, cf. ‘hunter’s dog’ in (79b) above and (81): (81)

n-tˇsa eˆ m-pe CL5-dog CL5-ugly ‘ugly dog’

These distinction can be treated in terms of declension classes, which are partly semantically based: under this analysis, we are dealing with semantic grouping of nominals which requires a presence of a particular morpheme. Still, we can reasonably claim that we are dealing with the same encoding strategy. Roughly the same pattern is observed in Zulu (Poulos and Msimang 1998). The a-of-relationship construction, then, is a device for turning any type of phrase into a phrase with the same agreement morphosyntax as an adjective. In this sense we are dealing with a single, across-the-board, modification encoding strategy.

33

5 The Possession-Modification Scale The material from the previous section demonstrates that languages show a very strong tendency towards a certain kind of uniformity with respect to encoding strategy. In this section we posit a more finelygrained hierarchy and a universal principle according to which languages must respect it. We will then look at some exceptions.

5.1 Monotonicity and its motivation We argue that polyfunctional construction types reflect the Possession-Modification Scale as shown in (82): (82)

Possession-Modification Scale A < B < C < D

The strongest claim we can advance is that encoding strategies respect a monotonicity requirement along the Possession-Modification Scale, according to which a given strategy will cover only continuous segments of the scale. For instance, we might find that a language employs two strategies, one for A and B constructions and the other for C, D constructions, schematically A=B6=C=D. We have seen examples of this situation above. However, what we predict will not occur is a language in which there is a non-monotonic relationship between encoding strategies and the Possession-Modification Scale. For instance, we predict that no language will exhibit encoding patterns of the form, say, A=C6=B=D, A=C=D6=B, A6=B=D6=C and so on. More concretely, we do not expect to encounter a language which groups together qualitative attributes/alienable possession or modification-by-noun/inalienable possession to the exclusion of other types. Recall that we state this claim over the most liberal interpretation of ‘encoding strategy’ possible, which permits differences in linear order of constituents. What might motivate such a scale? Essentially, we claim that this is a scale which is heavily grounded in semantics. It can be thought of as a two-dimension representation of crosslinguistic regularities in semantic structure of adnominal constructions. The basic idea is similar to that of the semantic maps approach (e.g. Haspelmath 2003, Croft 2003: 133f, among others). This approach provides an empirically testable tool for studying meaning variations across languages and show convergence with grammaticalization theory, as well as with the research using (implicational) hierarchies, as found in functional typology and Optimality Theory. Existing semantic map analyses deal with polyfunctional morphemes that are realized as adpositions or bound affixes and serve as exponences of categories such as case, tense or modality. Categories of this kind are claimed to form a complex network of submeanings and languages vary in how they categorize them. In this study we are dealing with the meaning of syntactic constructions rather than individual morphemes. The form of the constructions is defined in terms of what we have referred to as encoding strategy. The claim is that encoding strategies employed to express the relevant constructional meaning exhibit monotonicity along the Possession-Modification Scale because the neighbouring points on the scale are semantically/conceptually close. First, it is very common cross-linguistically to find alienable and inalienable possession grouped together, so that the possessive construction is ambiguous between these two interpretations. Since the possessive construction is capable of expressing numerous kinds of relationships between two entities, in some accounts the possessive relation is even considered semantically indeterminate. For example, Kay and Zimmer (1976) argue that the possessive (in their terms, Genitive) NP is simply a metalinguistic instruction to the hearer that there is some kind of relation between the possessor and the possessee. The hearer automatically supplies the appropriate interpretation in a manner that makes sense, given the context in which they appear.

34

However, the meaning of the possessive construction itself is not totally empty. Crucially, the relationship between the possessor and the possessee is largely asymmetric: possessors function as pragmatic anchors (J. A. Hawkins 1978; R. Hawkins 1981; Fraurud 1990; Koptjevskaja-Tamm 2000, 2004) or reference points (Langacker 1993, 1995) for identifying the possessee, but not the other way round. On Langacker’s account, all possessives are based on one common cognitive feature, “the ability to invoke the conception of one entity for purpose of establishing mental contact with another, i.e. to single it out for individual conscious awareness” (Langacker 1993: 5). This ability motivates the asymmetrical relationship: the possessor serves as a conceptual reference point for locating the possessed. The relationship between them cannot normally be reversed: we have the cat’s fleas but not *the fleas’ cat. In inalienable possession the conceptual distance between two entities is especially close, because “inalienability denotes an indissoluble connection between two entities, a permanent and inherent association between the possessor and the possessed” (Chappell and McGregor 1996: 4). In other words, inalienable possession naturally leads to the reference point interpretation and provides the clearest instance of it. But alienable possession is also based on the reference point function, as reflected in Taylor’s list of prototypical possessive properties. Thus, the reference point function constitutes the constructional meaning common to all possessives and it is compatible with a diverse range of relationships that hold between the possessor and the possessee.15 Since the possessor is a conceptual reference point for identifying the possessee, it tends to be referential (e.g. Haspelmath 1999). Assuming that the carrier of referentiality is DP, as is generally assumed in the syntactic literature, this suggests that possessors, both alienable and inalienable, are canonically fully-fledged DPs. This semantics is corroborated by syntax, where possessives are typically handled in terms of DPinternal predication. In syntactic models possessors are often associates with a specifier position within a maximal projection of a functional category D (e.g. Abney 1987; Alexiadou and Wilder 1998; Bittner and Hale 1996; Ritter 1991 as well as references therein). As has been argued by Anderson (1983), Szabolcsi (1983, 1994) and others, the category D assigns a theta-role to the possessor. This role is arbitrary and attributed to a functional component of the construction. It merely serves to satisfy the theta-criterion without having a specified content, so that any semantically unrestricted relation can obtain between the possessor and possessed. What remains invariant is that, whatever the exact interpretation of this relation, the possessor is structurally and therefore functionally more prominent that the possessed. In inalienable possession the possessor DP is an argument in a two-place relation determined by the meaning of the head noun. Inalienable possession is therefore understood grammatically as a two-place predicate that denotes a relation over a pair of entities. In alienable possession the relationship between the possessor DP and the possessed noun is semantically unspecified and can be designated simply as a two-place relation, ‘ℜ’ (Kathol 2002). For example, the rough semantic representation for John’s book would be something like ℜx[(John, x) ∧ book (x)]. In the default case ℜ is interpreted as ownership (whatever that means in different cultures), but many languages allow non-ownership readings based on a contextually established association between two entities. Next, we have seen that some languages treat alienable possession and modification-by-noun as one and the same construction. The rationale behind this pattern is that in both cases we are dealing with contextually/pragmatically determined relationship between two entities. In alienable possessives this is the relation between the possessor DP and the head noun. In a similar manner, modification-bynoun involves two nouns (or more accurately, two nominal denotations) and some kind of semantically indeterminate relation between them. The difference between modification-by-noun and alienable possession is that in modification-bynoun the modifying noun does not have the status of co-argument with the head noun. Instead its meaning is part of the complex denotation of the modifier as a one-place predicate. In Taylor’s (1996: 17) words, “in opting to use a possessive expression, the speaker is instructing the hearer on how best 15 There

are other problems with “semantic incompleteness” accounts, see, for instance, Taylor 1996 and Nikiforidou 1991.

35

to identify the referent that he, the speaker, intends”. In contrast, when a noun modifies another noun, “the dependent-head combination refers to a subclass of a broader class and often functions as a classificatory label for it, suggesting that the dependent and the head together correspond to one concept” (Koptjevskaja-Tamm 2004: 156). The entity denoted by the dependent is not individuated but rather denotes a class of objects. In this sense it is a non-referential counterpart to the (typically referential) alienable possessor, cf. dog’s collar (alienable possession denoting some dog’s collar) vs. dog collar (modification-by-noun denoting a type of collar). This means that modification-by-noun involves modification by non-referential dependents, that is, by Ns or perhaps NPs, rather than by DPs, as in the case of possession.16 Finally, many languages group together modification-by-noun and canonical modification. The similarity between them lies in the fact that both types involve set intersection, where the denotation of the dependent identifies a certain subset in the denotation of the head noun. Just like regular adjectival modification, modification-by-noun serves to classify, describe or qualify the entity rather than to identify it by reference to another entity. For example, in the expression cat food we have a compound whose denotation is a subset of the denotation of the head noun. The subset is conventionally defined as the result of some relation between the denotation of the modifying N and the head N, hence cat food means ‘food which bears some relation to the notion of ‘cat’ ’. This relation is pragmatically (contextually) determined. The notion ‘food’ can be given some kind of relation to any other concept, thereby narrowing down its reference. The expression cat food selects that subset of food denotations which is defined by reference to the denotation of ‘cat’. Thus, modification-by-noun is a kind of modification, but not a canonical kind defined in Section 2.2. In contrast to standard adjectival modification with semantically simple property concepts, in modification-by-noun the semantic structure of the modifier is complex because it involves another noun-like entity (cf. Beck 2002: 88). In canonical modification there is no relation between two noun-like entities, because, of course, there is only one such entity. The difference between the four semantic types can be schematically represented as in (83), where ℜ stands for our unspecified semantic relation and P and A denote semantically specified predicates. (83)

a. b. c. d.

inalienable possession alienable possession modification-by-noun canonical modification

P (DP, N(DP)) ℜ(DP, N) ℜ(N, N) A (N)

The representation for adjectives in (83) takes an adjective to be a second order predicate taking noun denotations as its argument, though this is mainly for expository convenience. In modification by noun the semantically unspecified relation ℜ relates two noun denotations, while in alienable possession it relates a modifying phrase, DP, and a modified noun. Finally, in our representation of inalienable possession we follow essentially Barker’s (1997) notion of ‘lexical possession’17 . Inalienably possessed nouns are relational nouns, such as kin terms and nouns denoting objects with parts. They are represented as having an obligatory argument, so that ‘mother’ really means ‘(someone’s) mother’ and ‘leg’ means ‘(someone’s) leg’. We treat inalienable possession, then, as a special relation holding between a modifying DP and a relational noun, such that the modifying DP is identified with the noun’s argument. As can be seen from these representations, one of the crucial parameters is referentiality of the dependent. In alienable and inalienable possession the dependent is canonically referential and corresponds to a DP. In modificational constructions the dependent is canonically non-referential; it is either a lexical N which does not project a DP or an adjective. Hence the split between possession and modification (A = B 6= C = D) is a particularly natural split, grammaticalized in many languages. 16 It seems to be possible to modify a noun by complex constructions involving referential DPs, such as adpositional phrases or some kind of relative clause. These are non-canonical cases, which we do not address in this paper. 17 The representation in (83a) is meant to be an approximation to Barker’s representation λx λy [child(x, y)] (Barker 1995: 52).

36

5.2 Counter-examples to the scale Although we believe that our Possession-Modification Scale represents real tendencies in grammatical and semantic organization, we are well aware that such patterns are the outcome of complex grammaticalization processes and therefore allow a certain degree of variation. We have so far only encountered very limited instances of violations of Possession-Modification Scale, but the historical processes being what they are one might expect a variety of counter-examples to arise. There are three basic logically possible ways in which monotonicity can be violated, shown schematically in (84), where a star, an apostrophe and the lack of diacritics all indicate different encodings. (84)

Possible counter-examples to monotonicity A A′ A∗

B∗ B∗ B∗

C C C

D∗ D∗ D∗

b. A∗ A∗

B′ B

C C

D∗ D∗

A∗ A∗

B′ B

C∗ C∗

D D∗

a.

c.

First, we might expect to encounter languages where modification-by-noun receives identical expression with inalienable possession, while alienable possession differs (B = D), Type (a). There are three possibilities for the encoding of the remaining two semantic functions: (i) A = C, (ii) A 6= C, and (iii) A 6= C but A = B. This type seems to exist and is in fact quite frequent, especially its A∗ B∗ C D∗ subtype, where attributive modification, modification-by-noun and inalienable possession are expressed by juxtaposition, while alienable possession is marked by some kind of genitive. A number of West African languages exhibit this type of construction (cf. Creissels 2000: 249). For instance, in Ewe inalienable possession is expressed by simple juxtaposition (with the element order dependent-head). Attributive modification is also by juxtaposition (though with the opposite element order of head-dependent, apparently. Recall we are ignoring word order, otherwise Ewe may not even be a counterexample to the scale). (85)

a.

fia dada chief mother ‘the chief’s mother’

b.

xO nyu´ı house good ‘a good house’

Noun-noun compounding is dependent-head juxtaposition (though sometimes with tone sandhi): (86)

a.

eVe t´O Ewe man ‘an Ewe man’

b.

agble t´O farm owner ‘a farm owner’

However, alienable possession is expressed by the adposition Fe´ (presumably a postposition, though this is not clear from Westermann’s account (Westermann 1930: 49): (87)

fia Fe´ xO chief POSS house ‘the chief’s house’

Another case in point would be Kabba (Nilo-Saharan), as described in Moser (2004). Consider examples (88): 37

(88)

a.

d`ew ngall person tall ‘a tall person’

b.

buw´a hole ‘water hole’

m`aann water

c.

k`ul`a l´e bb`@t@ work GEN woman ‘woman’s work’

d.

m`@k`@j`@ g`Ol-`E knee leg-3 SG ‘the knee of his leg’

A similar pattern is found in the Mande languages. In Jeli (West Mande), as described in Tr¨obs (1998), there is a similar split between alienable and inalienable possession: only the former requires an overt constructional marker. Modification-by-noun is also expressed by some kind of compounding. However, attributive modification requires a special attributive form of a qualitative verb, derived with the suffix -ra/-rE and referred to as the resultative participle by Tr¨obs (1998). Unlike other dependents, the participles follow the head. (89)

a.

wulu gbOgO-rE dog become.black-PART ‘black dog’

b.

gba wu tree head ‘tree head’

b.

Soma ra Soma POSS ‘Soma’s car’

d.

Mama kEli Mama husband.DEF ‘Mama’s husband’

monbilo car.DEF

Thus, in the case of Jeli we are dealing with the A′ B∗ C D∗ subtype. The A B∗ C D∗ pattern involves a double violation of the scale. Our only example to date is Lango (Western Nilotic). Noonan (1992) describes two types of ‘associative construction’. Inalienable possession is expressed by the juxtaposition of the head and the dependent if the dependent is a lexical NP. The inalienably possessed nouns denote ‘parts’ in ‘part-of’ relations, including body parts, pictures, character features and other notions relating to the individual’s self, blood relatives, locational notions and the like. (90)

a.

w´ı rw`ot head king ‘the king’s head’

b.

c´al l´oc`@ picture man ‘the man’s picture’

Both alienable possession and attributive modification employ the so-called associative construction, which consists of the head noun followed by the attributive particle a` and the dependent NP or AP. (The final consonant is geminated before a` ). (91)

a.

gwˆokk a` dog ATTR ‘the man’s dog’

l´oc`@ man

b.

gwˆokk a` dog ATTR ‘the good dog’

b`Er good

Unlike possessors, attributive adjectives show number agreement with the head, so on our definitions they follow a different strategy. However, nominal modifiers do not take the attributive a` or at least this is not mentioned in the grammar. Instead nominal modification is achieved by what Noonan calls a fully productive compounding strategy, which expresses “any contextually reasonable association between the compound elements” (Noonan 1992: 115). (92)

a.

gw´ok "r´OmˆO dog sheep ‘dog sheep’

b.

d´Og dˆel mouth skin ‘lips’ 38

As Noonan (1992: 115) explicitly notes compounded nouns “are simply juxtaposed in a syntactic configuration like that of inalienable associative constructions”. Second, we can expect cases where attributive modification is expressed similarly to inalienable possession (A = D), but alienable possession and modification-by-noun have a different expression, Type (b). We can identify two sub-variants of this violation, one in which the B and C functions are encoded distinctly and another in which B and C receive the same encoding. This latter seems to us to be a particularly bad violation of monotonicity, because the B = C encoding seems to imply that the language works with a split between inalienable possession and the rest, which is subverted by the A = D encoding. This is not a likely scenario. It is of course easy to find languages in which attributive adjectives and inalienable possessors are both expressed by simply juxtaposition. The question is whether there are such languages in which both B or C are expressed by other means. For instance, we might expect to find a language in which both modification-by-noun and alienable possession were expressed by a special preposition (subtype A∗ B C D∗ ) or in which modification-by-noun was expressed by a preposition and alienable possession by possessor agreement (A∗ B′ C D∗ ). However, we are still looking for such a language. Finally, attributive modification can be encoded similarly to alienable possession (A = C), Type (c). Again, there are two possibilities here: (i) B 6= D, and (ii) B 6= D but C = D. For Type (c) what is needed is a language which marks (alienable) possession as attributive modification (say, with a possessive adjective construction such as that found in Chukchi) but which encodes modification-by-noun in some distinct manner (for instance, by compounding). In fact, although NN compounding is indeed a minor encoding strategy in Chukchi, most of the examples seem to have become quickly lexicalized and so the strategy cannot be regarded as productive in the same way as relational adjective formation is productive. We therefore just need to find some kind of ‘Chukchi-prime’, which is exactly like Chukchi except that the relational adjective morphology has been lost and compounding has become completely productive, as in Germanic. We have not observed other types of counterexamples so far. Types (b) and (c) turn out to be surprisingly infrequent and we have not been able to identify any reliable examples of either sort, even when we factor out word order and other possible syntactic determinants. Although we do not a priori exclude them, they are expected to be rare. This may be a failing of our construction-driven typology or it may simply reflect the workings of the scale.

5.3 Juxtaposition When certain data appear to violate a scale motivated by a well-founded directionality hypothesis, one can look for a non-semantic explanation. Crucially, juxtaposition is involved in all the types of exceptions we have encountered so far. This suggests that there may be something special about it, as opposed to other types of constructional encodings. In our introductory remarks to Section 4 we mentioned juxtaposition as one of the strategies used to encode NP-internal head-dependent relations. However it is important to emphasize that we may be dealing with very different syntactic relationships here. Unfortunately, juxtaposition may be difficult to distinguish from other grammatical phenomena which involve more than just linear ordering and adjacency. For instance, frequently juxtaposed elements may receive special prosody (such as English ‘compound stress’) or trigger special phrase phonology. In addition, we may find we have to distinguish between ‘syntactic adjacency’ and some other, tighter, degree of adjacency. In languages which clearly distinguish a syntactic word level and a phrase level, syntactic adjacency is in general defined over phrases. For instance, in English, French and various other languages a noun is modified by a phrase whose lexical head is an adjective. Yet when prenominal an attributive adjective is able to project only very limited types of phrase, compared to its postnominal or predicative usage. This prompts Sadler and Arnold (1994) to speak about ‘small constructions’, mid-way between word- and phrase-level, and 39

Kageyama (2001) argues for a similar notion (his ‘Word+’) on the basis on evidence from Japanese. In the literature on noun incorporation we frequently see reference to N-V constructions in which the incorporated noun does not really form a compound as such with the verb, but which does not project a phrase either (see, for instance, Miner’s 1986 distinction between ‘loose compounding’ and genuine incorporation, echoed by various subsequent authors). Abstracting away from these complications, we can understand simple juxtaposition as the absence of constructional marking and stipulated adjacency between the head and the dependent, usually in a fixed order. Understood in this way, juxtaposition is relatively easy to discern. We mentioned above that juxtaposition often encodes inalienable possession. In typology there has been much discussion of languages where inalienable constructions are signalled by simple juxtaposition, while alienable possession is signalled by some kind of overt morphosyntactic marking. A popular type of explanation rests on the notion of iconicity. It has been argued in various places that inalienable possession involves conceptually closer relationship between the possessor and the possessed than alienable possession, and this is often iconically reflected in ‘linguistic distance’ between them. This reasoning goes back to Haiman (1983) and is favoured by Chappell and McGregor (1989). They suggest two scales that iconically reflect the conceptual proximity between the head and the dependent. The first one, the ‘constituent status’ scale, has to do with the degree of formal separateness between the two and indirectly reflects the degree of referentiality of the dependent: it goes from a combination of two phrases to juxtaposition and then to lexical compounding to a single lexeme. The second scale has to do with morphological marking: it goes from complex formal markers relating the head and the adnominal to the absence of marking and is said to iconically reflect the conceptual proximity between the two concepts. On both scales inalienable possession is between alienable possession and ‘classification’ (modification-by-noun, in our terminology), but closer to the latter, which would explain why the two are often expressed by similar means. Basically the same reasoning is adopted in Croft (1990: 175-176), Koptjevskaja-Tamm (1997), and Lazard (2005) and is reflected in some syntactic accounts. Inalienable possessor are analyzed as forming some sort of complex predicate with the possessed noun and therefore structurally closer to the latter than alienable possessor (Vergnaud and Zubizarreta 1992; Alexiadou 2003). The iconicity explanation has recently been challenged by Haspelmath (2008), who provides a number of counter-arguments to it. In particular, iconicity predicts that the constructional marker in alienable constructions should occur between the possessor and the possessum, but this is not always the case. The possessive marker may occur at the periphery of the phrase, as in Dogon, Puluwat, O’odham, Koyukon, and Achagua, among other languages. Haspelmath suggests that the markedness pattern in inalienable possession is based on economy rather than iconicity. Coding asymmetries in possessive constructions are due to their differential predictability, something which can be measured by relative frequency. Frequent predictable patterns that can be easily inferred are known to need less formal markedness. Relational nouns normally, or at least very frequently, occur as possessums in possessive NPs, while this is much less frequent for non-relational nouns. This implies that for the former the possessive relation can be inferred and so its overt marking is relatively redundant, while for the latter the possessive relation is not expected and therefore has to be formally marked. While we basically agree with the essence of the frequency explanation when it comes to the inalienablility split in possessives, our question is more general in nature and follows from the fact that we look at a wider range of constructions, including not only possession but modification. Juxtaposition is commonly used to express all relevant semantic functions. There are languages where all possessives are morphologically unmarked, e.g. Celtic (Koptjevskaja-Tamm 2003b: 649), and Kolyma Yukaghir (see example (53), Section 4.5) and languages where juxtaposition serves to express alienable possession alone (see example (2), Section 2.2). Non-agreeing non-derived modifiers represented in many languages are also juxtaposed to the head by our definition of juxtaposition. Juxtaposition embraces numerous compound-like structures in English and elsewhere. We have argued that compounding of

40

some sort is a common strategy for expressing modification-by-noun, and this is true not only of Germanic. For instance, it exists in Japanese and Mandarin Chinese (Liu 2003: 62f), although the degree of morphosyntactic tightness between the elements of such compounds is not always clear and the existing descriptions do not always resolve this issue. Perhaps a relevant consideration here is the fact that juxtaposition (or specific word orders generally) are, in an important sense, construction-neutral. It is just one aspect of the more general phenomenon of expression by syntactic structure. An interesting question arises here for the canonical approach to typology (in the sense of Corbett 2007). Is juxtaposition a canonical encoding strategy of an NP-internal head-dependent relation or a non-canonical one? If so, is it canonical for possession or modification? The answer to this question is not easy to determine, but at least in some respects juxtaposition seems to be canonical because it does not rely on anything other than what all languages have to have, namely, linearized sets of words. Simple linear adjacency can be used to express more or less any relationship and without some language-specific convention it is not even possible to identify which element is the head and which the dependent. Moreover, it is completely independent of grammatical category/grammatical relations (pace Kayne 1994). In this sense it is the default encoding strategy for any grammatical relation.18 The important point is that juxtaposition is inherently neutral with respect to the possession/modification distinction, unlike morphosyntactic constructions such as possessor agreement or adjective agreement. It can appear unpredictably (from the point of view of our implicational scale) for expressing any type of relevant semantic or grammatical function, ranging from inalienable possession to both types of modification.

6 Conclusions and prospect In this paper we have addressed four types of adnominal construction with different, though related, meanings: attributive modification (type A), modification-by-noun (type B), alienable possession (type C) and inalienable possession (type D). In some languages each type has as its default or principal mode of expression a distinct morphosyntactic construction, but in most cases we find that some or all of the four types are canonically expressed in that language using the same morphosyntactic means. We have argued that the four construction types define an implicational scale which models the polyfunctionality of the morphosyntactic encodings. The scale goes from property-denoting adjectival encoding to objectdenoting nominal encoding. The formal overlap in these encodings was shown to respect a monotonicity requirement: wherever we have overlapping encoding, each morphosyntactic strategy covers some continuous segment of the scale. In other words, if a language opts to treat adjectival modifiers using a nominal strategy then it will tend to use this strategy for the other constructions, and if it tends to use an adjectival strategy for the possessor construction it will tend to use this strategy throughout. We expect exceptions to creep in but they are surprisingly rare. We suggested a semantic explanation for this generalization, based on the idea that adjacent points on the scale exhibit semantic affinity. For instance, if a language elects to express alienable possession systematically by means of a possessive adjective strategy it will be under great pressure to express modification-by-noun using a similar adjectival strategy (i.e. using a relational adjective). The semantic pressure to conform to canonical content-to-form mappings is sufficiently strong to prevent languages 18 Juxtaposition

is also likely to be historically primary. We can easily imagine a grammaticalization scenario in which a split develops between alienable and inalienable possession, with alienable possession (C) being marked by an innovating genitive case marker and inalienable possession (D) being minimally marked by juxtaposition. We can equally imagine that the new genitive would then take over the role of coding modification-by-noun (C). This is what Haspelmath calls “inhibition of expansion”. A novel construction can make an existing meaning more transparent by including an additional morpheme and expand to new contexts, but it will not spread to the contexts in which the relevant meaning occurs most often (inalienable possession). All this time, however, we might find that adjectives (A) modify by pure juxtaposition. This is probably what happened in Miya, but we leave this speculation to future research.

41

from grammaticalizing constructions against this gradient. This is a somewhat unexpected finding unless we assume that languages take the semantic notion of ‘modification’ as a starting point and define a partitioning of the four construction types by appeal to something like the representations we have proposed in (83). Compared to many typological studies our database and our analysis may seem rather a meagre affair, presenting, as it does, only a one-dimensional scale rather than a luxuriant multidimensional map. There is a good reason for this. The kind of semantic affinity which gives rise to our implicational scale is different from the usually studied types of semantic effect on constructions. A fair amount of research has been devoted to the study of polysemy effects, in which a given morphosyntactic construction with a core meaning (such as, say, ‘reflexive verb’) undergoes semantic drift and acquires a host of polysemous or homonymous usages. In our case, the effects of the semantic relatedness is the precise opposite: a single semantic thread runs through our four construction types, namely, the poorly understood notion of ‘modifier of a noun’. This semantic core serves to restrict the types of encoding strategies that can be found across the four types. This type of semantically-driven restriction across morphosyntactic expression types is a phenomenon which has not been widely discussed in the typological literature and it is a line of inquiry which we believe may prove very fruitful if applied to a greater variety of construction types. In addition to potential significance for typology generally, we believe that ensembles of constructions such as adjectival modifier - nominal possessor can throw important light on the question of what constitutes a lexical category. By focusing on aspects of attributive modification and possession which regularly pattern together we highlight those properties that have to be regarded as systematically related. We have argued that a detailed study of the various ways in which noun modification is expressed cross-linguistically is essentially for understanding not only the category of ‘adjective’, but also the category of ‘noun’ itself. Our study has raised a question which is not often asked, but our answers raise additional questions, of central concern to mainstream typology and linguistics generally. A deeper understanding of the Possession-Modification scale and what kinds of lexical category can instantiate it, particularly one that includes more detailed information about syntactic behaviour, would illuminate these murky questions of word class. This would require a study which was able to resolve difficult issues of what constitutes a modifier of a noun as opposed to a specifier of a noun, raising complex questions about the nature of nominal syntax, but the principles behind such a study should be clear. Further research should also factor in the contribution of other means of noun modification such as relative clauses, participial modifiers, adposition phrase modifiers and so on. One particularly important aspect of syntactic organization, touched upon in Section 5.3 is the varying roles that simple juxtaposition can play, and more generally, the way that adjacency of syntactic terminal elements (syntactic words), interacts with the prosodic and morphophonological level of organization. Again, these questions merit serious discussion of their own. At the same time, some of the puzzles we have identified can only be properly resolved against the background of a suitably elaborated model of lexical representation and lexical relatedness, such as that argued for in Spencer (2010). For instance, modification-by-noun and certain types of possessive construction involve encoding strategies that inherit aspects of noun morphosyntax as well as acquiring aspects of adjectival morphosyntax, such as agreement. Sometimes this can create words which seem to belong simultaneously to the category of noun and of adjective, so-called mixed categories. We have discussed some of the implications of such category mixing in earlier work Nikolaeva and Spencer (2009a) where we argue for a factorization of components of a lexical entry. There is always the danger when discussing mixed categories of trying to shoehorn constructions into instantiations of one category or the other. It should be clear from our discussion that the correct question to ask in such cases is not (necessarily) ‘which category does this word belong to?’ but rather ‘which properties does this word inherit from category X and which properties does it inherit from category Y (and in what contexts, with what restrictions and so on)?’ This is a much more nuanced approach to the problem of lexical

42

categories and one which generally requires rather sophisticated datasets. We hope that by asking such questions we can stimulate researchers on specific languages or language groups to investigate more thoroughly these generally uncharted areas. Finally, for future consideration, we believe that our study provides an interesting test case for the benefits of the canonical approach to typology proposed recently by Corbett (2006, 2007). When we describe a category or construction type in the canonical approach, we establish a small number of uncontroversial defining properties for that category or construction, thus defining a logical ideal from which real exemplars deviate to varying degrees. In this respect, we can treat inalienable possession as a canonical type of possessive relationship (one driven by the semantics of the possessed noun) and we can treat modification by gradable property-denoting predicate as canonical attributive modification (driven by the semantics of the adjective). On this perspective, modification-by-noun and alienable possession both turn out to be non-canonical varieties of attributive modification and of inalienable possession. In Nikolaeva and Spencer (2009b) we offer a very preliminary analysis of the adjectivepossessor distinction along such lines, arguing, inter alia, that such an approach allows us to factor out the various components of ‘mixed category’ encodings often found with the intermediate constructions, modification-by-noun and alienable possession. We believe that such a canonical approach to these matters would help resolve much of the controversy that surrounds these constructions while at the same time providing us with useful analytical tools for investigating implicational scales or other universal properties of the kind proposed here.

43

Abbreviations ADESS adessive, AGR agreement, ATTR attributivizer, ART article, CL noun class, DEF definite, DU dual, EP epenthesis, EZ ezafe, GEN genitive, INDEF indefinite, INES inessive, INF infinitive, LOC locative, LKR linker, M masculine, NOM nominative, OBJ object, OBL oblique, PART participle, PASS passive, PL plural, POSS possessive, REL relative, SG singular, SIM similative, SOC sociative, VN verbal noun.

References Abney, Stephen P. 1987. The English Noun Phrase in Its Sentential Aspect. Ph.D. thesis, MIT. Ackema, Peter and Ad Neeleman. 2004. Beyond Morphology: Interface Conditions on Word Formation. Oxford: Oxford University Press. Alexiadou, Artemis. 2003. Some notes on the structure of alienable and inalienable possessors. In M. Coene and Y. d’Hulst, eds., From NP to DP. Vol. 2. The Expression of Possession in the Noun Phrase, pages 167–188. Amsterdam: John Benjamins. Alexiadou, Artemis and Chris Wilder, eds. 1998. Possessors, Predicates and Movement in the Determiner Phrase. Linguistik Aktuell/Linguistics Today 22. Amsterdam: John Benjamins. Anderson, Mona. 1983. Prenominal genitive NPs. The Linguistic Review 3:1–24. Barker, Chris. 1995. Possessive Descriptions. Stanford: CSLI Publications. Bauer, Winifred. 1997. The Reed Reference Grammar of Maori (with William Parker, Te Kareongawai Evans, and Te Aroha Noti Teepa). Auckland: Reed Books. Beard, Robert. 1995. Lexeme-Morpheme Base Morphology. Stony Brook, NY: SUNY Press. Beck, David. 2002. The Typology of Parts of Speech Systems: the Markedness of Adjectives. London: Routledge. Bergsland, Knut. 1997. Aleut grammar. Fairbanks, Alaska: Alaska Native Center. Besnier, Niko. 2000. Tuvaluan: a Polynesian Language of the Central Pacific.. London: Routledge. Bhat, D. N. S. 1994. The Adjective Category: Criteria for Differentiation and Identification. Amsterdam: John Benjamins. Bittner, Maria and Ken Hale. 1996. The structural determination of case and agreement. Linguistic Inquiry 27(1):1–68. Boeder, Winfried and Christoph Schroeder. 2000. Relational coding in Georgian and Turkish noun phrases: syntax, derivational morphology, and “linking” by means of participles. Turkic Languages 4:153—204. Carleton, Troi and Rachelle Waksler. 2000. Pronominal markers in Zenzontepec Chatino. International Journal of American Linguistics 66(3):381—395. Chappell, Hilary and William McGregor. 1989. Alienability, inalienability and nominal classification. In Proceedings of the Fifteenth Annual Meeting of the Berkeley Linguistics Society, vol. 15, pages 24–36. Berkeley Linguistics Society. Chappell, Hilary and William McGregor. 1996. Prolegomena to a theory of inalienability. In H. Chappell and W. McGregor, eds., The Grammar of Inalienability: A Typological Perspective on Body Part Terms and the Part-Whole Relation, pages 3–30. Berlin: Mouton de Gruyter. 44

Chomsky, Noam. 1970. Remarks on nominalization. In R. Jacobs and P. Rosenbaum, eds., Readings in English Transformational Grammar, pages 184—221. Waltham, MA: Blaisdell. Christen, Simon. 2001. Genitive positions in Baltic and Finnic languages. In O. Dahl and M. Koptjevskaja-Tamm, eds., Circum-Baltic Languages: Their Typology and Contacts, pages 499– 520. Amsterdam: John Benjamins. Cole, D. T. 1979. An Introduction to Tswana Grammar. Johannesburg: Longman. Corbett, Greville G. 2006. Agreement. Cambridge: Cambridge University Press. Corbett, Greville G. 2007. Canonical typology, suppletion, and possible words. Language 83(1):8–42. Creissels, Denis. 2000. Typology. In B. Heine and D. Nurse, eds., African Linguistics. An Introduction, chap. 9, pages 231–258. Cambridge: Cambridge University Press. Croft, William. 1990. Typology and Universals. Cambridge: Cambridge University Press. Croft, William. 2003. Typology and Universals. Cambridge: Cambridge University Press, 2nd edn. Davis, Karen. 2003. A Grammar of the Hoava Language, Western Solomons.. Canberra: Pacific Linguistics, Research School of Pacific and Asian Studies, Australian National University. Dayley, Jon P. 1989. T¨umpisa (Panamint) Shoshone Grammar. Berkeley: University of California Press. Dixon, Robert M. W. 1991. A New Approach to English Grammar, on Semantic Principles. Oxford: The Clarendon Press. Dixon, Robert M. W. 2000. Categories of the Noun Phrase in Jarawara. Journal of Linguistics 36:487– 510. Downing, Pamela. 1977. On the creation and use of English nominal compounds. Language 55:810– 842. Frajzingier, Zygmunt. 2001. A Grammar of Lele. Stanford, CA: CSLI Publications. Fraurud, Karl. 1990. Definiteness and the processing of noun phrases in natural discourse. Journal of Semantics 7:395–433. Ghomeshi, Jila. 1997. Non-projecting nouns and the ezafe: construction in Persian. Natural Language and Linguistic Theory 15:729—788. Gil, David. 2005. Genitives, adjectives, and relative clauses. Article 60. In M. Haspelmath, M. S. Dryer, D. Gil, and B. Comrie, eds., World Atlas of Linguistic Structures, pages 246–249. Oxford: Oxford University Press. G¨oksel, Aslı and Celia Kerslake. 2005. Turkish. A Comprehensive Grammar. London: Routledge. Golovko, E. V. 1997. Aleutskij jazyk [aleut]. In A. P. Volodin, N. B. Vaxtin, and A. A. Kibrik, eds., Jazyki Mira. Paleoaziatskie jazyki [Languages of the World. Paleosiberian Languages], pages 101– 116. Izdatel’stvo “Indrik”. Haiman, John. 1983. Iconic and economic motivation. Language 59:781–819. Haiman, John. 1985. Natural Syntax: Iconicity and Erosion. Cambridge: Cambridge University Press. Haspelmath, Martin. 1999. Explaining article-possessor complementarity: economic motivation in noun phrase syntax. Language 75(2):227–243.

45

Haspelmath, Martin. 2003. The geometry of grammatical meaning: semantic maps and crosslinguistic comparison. In M. Tomasello, ed., The New Psychology of Language, vol. 2, pages 211–242. Mahwah, NJ: Lawrence Erlbaum Associates. Haspelmath, Martin. 2008. Alienable vs. inalienable possessive http://www.eva.mpg.de/lingua/conference/08 springschool/pdf/course materials/ Haspelmath Possessives.pdf.

constructions.

Hawkins, John A. 1978. Definiteness and Indefiniteness: A Study in Reference and Grammaticality Prediction. London: Croom Helm. Hawkins, Roger. 1981. Towards an account of the possessive constructions: NPs and the N of NP. Journal of Linguistics 17:247—269. Heine, Bernd. 1997. Possession. Cognitive sources, Forces, and Grammaticalization. Cambridge: Cambridge University Press. Heine, Bernd. 2001. Ways of explaining possession. In I. Baron, M. Herslund, and F. Sorensen, eds., Dimensions of Possession, Typological studies in language 47, pages 311–328. Amsterdam: John Benjamins. Herslund, Michael and Ir`ene Baron. 2001. Introduction: Dimensions of possession. In I. Baron, M. Herslund, and F. Sorensen, eds., Dimensions of Possession, pages 1–26. Amsterdam: John Benjamins. Higginbotham, James. 1985. On semantics. Linguistic Inquiry 16(4):547–593. Huddleston, Rodney and Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Jackendoff, Ray S. 1977. X¯ Syntax: A Study of Phrase Structure. Cambridge, MA: The MIT Press. Jackson, Geoff and Jenny Jackson. 1999. An Introduction to Tuvaluan. Suva: Oceania Printers. Jokinen, Kristiina. 1991. On the two genitives in Finnish. EUROTYPE, Theme 7: Noun Phrase Structure. Working Paper 14, Strassbourg. Kachru, Yamuna. 2006. Hindi. Amsterdam: John Benjamins. Kageyama, Taroo. 2001. Word plus: The intersection of words and phrases. In J. van der Weijer and T. Nishihara, eds., Issues in Japanese Phonology and Morphology, pages 245–276. Berlin: Mouton de Gruyter. Kathol, Andreas. 2002. Nominal head-marking constructions: two case studies from Luise˜no. In F. V. Eynde, L. Hellan, and D. Beermann, eds., The Proceedings of the 8th International Conference on Head-Driven Phrase Structure Grammar, pages 189–201. Stanford: CSLI Publications. Kay, Paul and Karl Zimmer. 1976. On the semantics of compounds and genitives in english. In Proceedings of the Sixth California Linguistics Asociation Conference, pages 29–35. San Diego State University, San Diego, CA. Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA: The MIT Press. Kempson, Ruth M. 1977. Semantic Theory. Cambridge: Cambridge University Press. Koptjevskaja-Tamm, Maria. 1997. Possessive NPs in Maltese: Alienability, iconicity and grammaticalization. The Maltese NP Meets Typology (Special issue of Rivista di Linguistica, edited by A. Borg and Frans Plank) 8(1):245–274.

46

Koptjevskaja-Tamm, Maria. 2000. Romani genitives in cross-linguistic perspective. In V. Elˇs´ık and Y. Matras, eds., Grammatical Relations in Romani: The Noun Phrase, pages 123–149. John Benjamins. Koptjevskaja-Tamm, Maria. 2001. Adnominal possession. In M. Haspelmath, E. K¨onig, W. Oesterreicher, and W. Raible, eds., Language Typology and Language Universals, Vol. 2, vol. 2, pages 960–970. Walter de Gruyter. Koptjevskaja-Tamm, Maria. 2002. Adnominal possession in the European languages: form and function. Sprachtypologie und Universalienforschung 55:141–171. Koptjevskaja-Tamm, Maria. 2003a. Action nominal constructions in the languages of Europe. In Frans Plank, ed., Noun Phrase Structure in the Languages of Europe, pages 723–759. Berlin: Mouton de Gruyter. Koptjevskaja-Tamm, Maria. 2003b. Possessive noun phrases in the languages of Europe. In Frans Plank, ed., Noun Phrase Structure in the Languages of Europe, pages 621–722. Berlin: Mouton de Gruyter. Koptjevskaja-Tamm, Maria. 2003c. A woman of sin, a man of duty, and a hell of a mess: Non-determiner genitives in Swedish. In Frans Plank, ed., Noun Phrase Structure in the Languages of Europe, pages 515–558. Berlin: Mouton de Gruyter. Koptjevskaja-Tamm, Maria. 2004. “Maria’s ring of gold”: adnominal possession and non-anchoring relations in the European languages. In J.-Y. Kim, Y. Lander, and B. H. Partee, eds., Possessives and Beyond: Semantics and Syntax, pages 155–181. Amherst, MA: GLSA Publications. Kraft, Charles H. and A. M. H. Kirk-Greene. 1973. Teach Yourself Hausa. Sevenoaks: Hodder and Stoughton. Kurdoev, Kanat Kalashevich. 1978. Grammatika kurdskogo jazyka. Na materialie dialektov kurmandˇzi i sorani [A Grammar of Kurdish, Based on the Kurmanji and Sorani Dialects]. Moscow: Nauka. Langacker, Ronald W. 1991. Foundations of Cognitive Grammar, Vol. 2: Descriptive Applications. Stanford: Stanford University Press. Langacker, Ronald W. 1993. Reference-point constructions. Cognitive Linguistics 4:1–38. Langacker, Ronald W. 1995. Possession and possessive constructions. In J. R. Taylor and R. E. MacLayry, eds., Language and the Cognitive Construal of the World, Trends in Linguistics, Studies and Monographs 82, pages 51–79. Berlin: Mouton de Gruyter. Larson, Richard and Gabriel Segal. 1995. Knowledge of Meaning. An Introduction to Semantic Theory. Cambridge, MA: MIT Press. Lazard, Gilbert. 2005. What are we typologists doing? In Z. Frajzyngier, A. Hodges, and D. S. Rood, eds., Linguistic Diversity and Language Theories, pages 1–23. Amsterdam: John Benjamins. Levi, Judith N. 1978. The Syntax and Semantics of Complex Nominals. New York: Academic Press. Liu, Hsin-Yun. 2003. A Profile of the Mandarin Noun Phrase: Possessive Phrases and Classifier Phrases in Spoken Discourse. Munich: LINCOM Europa. Maslova, Elena S. 2003. A Grammar of Kolyma Yukaghir. Berlin: Mouton de Gruyter. McGregor, R. S. 1995. Outline of Hindi Grammar. Oxford/Delhi: Oxford University Press, 3rd edn. Mezhevich, Ilana. 2002. English compounds and Russian relational adjectives. In Proceedings of the North Western Linguistics Conference 2002, pages 95–114. 47

Miner, Kenneth. 1986. Noun stripping and loose incorporation in Zuni. International Journal of American Linguistics 2:242–254. Mosel, Ulrike and Even Hovdhaugen. 1992. Samoan Reference Grammar. Oslo: Scandinavian University Press. Moser, Rosemarie. 2004. Kabba: a Nilo-Saharan Language of the Central African Republic. Munich: LINCOM Europa. Newman, Paul. 2000. The Hausa Language: An Encyclopedic Reference Grammar. New Haven: Yale University Press. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62:56–119. Nichols, Johanna. 1992. Linguistic Diversity in Space and Time. Chicago: University of Chicago Press. Nikiforidou, Kiki. 1991. The meaning of the genitive: A case study in semantic structure and semantic change. Cognitive Linguistics 2:149–205. Nikolaeva, Irina A. 1999. Ostyak. Munich: LINCOM Europa. Nikolaeva, Irina A. 2002. Possession vs. nominal attribution in Uralic. In E. Helimski and A. Widmer, eds., W˘us´a w˘us´a – Sei gegr¨usst! Beitr¨age zur Finnougristik zu Ehren von Gert Sauer dargebracht zu seinem siebzigsten Geburtstag., Ver¨offentlichungen der Societas Uralo-Altaica 57, pages 239–250. Wiesbaden: Harrassowitz Verlag. Nikolaeva, Irina A. 2005. Review article on E. Maslova “A Grammar of Kolyma Yukaghir’. Linguistic Typology 9:299–325. Nikolaeva, Irina A. 2008. Between nouns and adjectives: A constructional view. Lingua 118:969–996. Nikolaeva, Irina A. and Andrew Spencer. 2009a. Adjectives as nouns and nouns as adjectives. Unpublished ms, SOAS/University of Essex. Nikolaeva, Irina A. and Andrew Spencer. 2009b. Canonical Typology and the Possession-Modification Scale. Unpublished ms, SOAS/University of Essex. Nikolaeva, Irina A. and Maria S. Tolskaya. 2001. A Grammar of Udihe. Berlin: Mouton de Gruyter. Noonan, Michael. 1992. A Grammar of Lango. Berlin: Mouton de Gruyter. Partee, Barbara H. 1997. Genitives, a case study. In J. van Benthem and A. ter Meulen, eds., Handbook of logic and language, pages 417–473. Dordrecht: Elsevier Publishers. Partee, Barbara H. and Vladimir Borsch¨ev. 2003. Genitives, relational nouns, and argument-modifier ambiguity. In C. M. E. Lang and C. Fabricius-Hansen, eds., Modifying Adjuncts, pages 67–112. Berlin: Mouton de Gruyter. Perry, John R. 2005. A Tajik Persian Reference Grammar. Leiden: E. J. Brill. Poulos, G. and C. T. Msimang. 1998. A Linguistic Analysis of Zulu. Cape Town: Via Afrika Limited. Ritter, Elizabeth. 1991. Two functional categories in noun phrases: Evidence from Modern Hebrew. In S. Rothstein, ed., Perspectives on Phrase Structure, Syntax and Semantics 25, pages 37–62. Academic Press. Rutgers, Roland. 1998. Yamphu: Grammar, Texts and Lexicon. Leiden: Research School CNWS (School of Asian, African and Amerindian studies). 48

Sadler, Louisa and Doug Arnold. 1994. Prenominal adjectives and the phrasal/lexical distinction. Journal of Linguistics 30:187–226. Samvelian, Pollet. 2007. 43(3):605–645.

A (phrasal) affix analysis of the Persian ezafe.

Journal of Linguistics

Schachter, Paul and Fe T. Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA: University of California Press. Schuh, Russell G. 1998. A Grammar of Miya. Berkeley, CA: University of California Press. Siegel, Muffy E. A. 1980. Capturing the Adjective. New York: Garland Publications. Skorik, Pjotr Ja. 1961. Grammatika cˇ ukotskogo jazyka, tom 1 [A Grammar of Chukchi, Volume 1]. Leningrad: Nauka. Sneddon, James Neil. 1996. Indonesian: a Comprehensive Grammar. London: Routledge. Spencer, Andrew. 1991. Morphological Theory: An Introduction to Word Structure in Generative Grammar. Oxford: Blackwell Publishers. Spencer, Andrew. 1999. Transpositions and argument structure. In G. Booij and J. van Marle, eds., Yearbook of Morphology 1998, pages 73–102. Dordrecht: Kluwer Academic Publishers. Spencer, Andrew. 2003. Does English have productive compounding? In G. Booij, J. DeCesaris, A. Ralli, and S. Scalise, eds., Topics in Morphology. Selected papers from the Third Mediterranean Morphology Meeting (Barcelona, September 20—22, 2001, pages 329–341. Institut Universitari de Ling¨u´ıstica Applicada, Universtitat Pompeu Fabra, Barcelona. Spencer, Andrew. 2005. Case in Hindi. In M. Butt and T. H. King, eds., The Proceedings of the LFG ’05 Conference, pages 429–446. University of Bergen, Norway. Spencer, Andrew. 2010. Lexical relatedness. Ms. in preparation. Sperber, Dan and Deirdre Wilson. 1986. Relevance: Communication and Cognition. Oxford: Blackwell Publishers. Stafford, R. 1967. The Luo Language. Nairobi: Longmans. Storto, Gianluca. 2004. Possessives in context. In J.-Y. Kim, Y. Lander, and B. H. Partee, eds., Possessives and Beyond: Semantics and Syntax, pages 59–86. Amherst, MA: GLSA Publications. Sulkala, Helena and Merja Karjalainen. 1992. Finnish. London: Routledge. Szabolcsi, Anna. 1983. The possessor that ran away from home. The Linguistic Review 3:89–102. Szabolcsi, Anna. 1994. The noun phrase. In F. Kiefer and K. E. Kiss, eds., The Syntactic Structure of Hungarian, pages 179–274. San Diego, CA: Academic Press. Taylor, John R. 1989. Possessive genitives in English. Linguistics 27:663–686. Taylor, John R. 1995. Linguistic Categorization: Prototypes in Linguistic Theory. Oxford: Oxford University Press. Taylor, John R. 1996. Possessives in English. An Exploration in Cognitive Grammar. Oxford: Oxford University Press. Tereˇscˇ enko, Natal’ja M. 1967. K razvitiju atributivnyx otnoˇsenij v samodijskiz jazykax [On the development of attributive relations in the Samoyedic languages]. Voprosy finno-ugorskogo jazykoznanija 4:234–242. 49

Tereˇscˇ enko, Natal’ja M. 1973. Sintaksis samodijskix jazykov [The Syntax of the Samoyedic Languages]. Leningrad: Nauka. Tr¨obs, Holger. 1998. Funktionale Sprachbeschreibung des Jeli (West-Mande). Cologne: K¨oppe. Vergnaud, Jean-Roger and Maria Luisa Zubizarreta. 1992. The definite determiner and the inalienable construction in French. Linguistic Inquiry 23:595–652. Westermann, Diedrich. 1930. A Study of the Ewe Language. London: Oxford University Press. Wetzer, Harrie. 1996. The Typology of Adjectival Predication. No. 17 in Empirical approaches to language typology. Berlin: Mouton de Gruyter. Williams, Edwin. 1982. The NP Cycle. Linguistic Inquiry 13(2):277–295. Wurzel, Petra. 1997. Rojbas. Einf¨uhrung in die kurdische Sprache. Wiesbaden: Dr. Ludwig Reichert Verlag.

50