1. Introduction
Australian languages, like many of the world's languages (Haspelmath Reference Haspelmath and Haspelmath2004), frequently express nominal coordination with an asyndetic construction in which the coordinated nominals are simply juxtaposed. Nominal juxtaposition is also used in many of these Australian languages to express a wide range of other, non-coordinated construction types (see e.g. Blake Reference Blake1987), including pronoun–noun appositions, part–whole constructions, generic–specific constructions and inclusory constructions, to be exemplified below. Juxtaposed nominal structures in Australian languages have received very little attention in the theoretical literature, yet they raise many interesting issues for syntactic analysis. In particular, the fact that such a range of coordinated and non-coordinated construction types are found with the same surface syntax has implications for many standard syntactic analyses which would assume that coordinations have a distinct syntactic structure from other nominal–nominal combinations, such as approaches postulating a CoordP for coordinate structures (Progovac Reference Progovac, Lindseth and Franks1997, Johannessen Reference Johannessen1998) for example. In this paper we provide an analysis of the full range of nominal juxtapositions in Australian languages within Lexical-Functional Grammar (LFG), an analysis which exploits the flexible architecture of LFG to capture the syntactic similarities of the various juxtaposed nominal constructions while allowing them to be suitably distinguished in the mapping to the semantics. On the view that we explore, the surface similarity between coordination and a whole range of other construction types which are expressed by means of juxtaposition motivates a similar treatment for each of these construction types in the syntax. Our analysis builds on the standard LFG approach to coordination (Kaplan & Maxwell Reference Kaplan and Maxwell1988, Dalrymple & Kaplan Reference Dalrymple and Kaplan2000, Dalrymple Reference Dalrymple2001), in which coordinated NPs are treated as set-valued grammatical functions at f-structure. We argue that non-coordinated juxtapositions, in which multiple nominals in juxtaposition share a single grammatical function in the clause, should likewise be treated as sets at f-structure, thus providing a unified analysis of a range of nominal juxtapositions common to many Australian languages. The differences between the various construction types are captured by distinguishing between them in the mapping to the semantics, which we discuss in some detail in section 5. Our modular approach, which differentiates syntactically between different types of nominal juxtaposition only in so far as such differentiation is overtly justified, provides a natural and unified account of nominal juxtaposition in Australian languages.
The paper is structured as follows. In section 2 we introduce a range of construction types which are widely and productively expressed by means of nominal juxtaposition in some Australian languages. Section 3 outlines the LFG approach to the syntax of coordination, which will form the basis of our syntactic account of nominal juxtapositions in Australian languages. Section 4 shows how this approach can straightforwardly account for asyndetic coordination structures and extends this syntactic analysis to account for the syntax of other juxtaposed nominal structures in Australian languages as well, providing both a natural account of these constructions and a unified account of juxtaposition in Australian languages. In section 5 we outline the semantics which differentiates these different construction types. Section 6 shows how a possibility inherent to our account of nominal juxtaposed structures provides for a straightforward account of inclusory constructions, and section 7 concludes.
2. Juxtaposed nominal constructions
The most common way to encode nominal coordination in Australian languages is through juxtaposition, whereby the coordinated nominals are simply listed with no particular marker of their coordinated status (1)–(6). The fact that such phrases have a conjunctive interpretation is shown by the form of elements which agree with the construction as a whole; for example, the coreferential pronouns in (3) and (4) and the verbal agreement marker in (5) and (6) pick up the resolved number feature.Footnote 2
(1) Niya kurrka-tha barruntha-ya wuran-ki nguku-y 3sg(nom) take-act yesterday-loc food-mloc water-mloc ‘Yesterday he took (with him) food and water.’
(Evans Reference Evans1995: 250 Kayardild)
(2) Gaj-ba ngurru manganyma yangaji eat-fut 1pl.inc(np) tucker(acc) meat(acc) ‘Let's eat the bread and meat.’
(Nordlinger Reference Nordlinger1998a: 257 Wambaya)
(3) Paanth thono pam pul mimp katp-r woman one(nom) man(nom) 3du(nom) cloth(acc) grasp-np ‘A woman and a man are holding up a piece of cloth.’
(Gaby Reference Gaby2006: 318 Kuuk Thaayorre)
(4) Dathin-a maku-wa bithiin-da bi-l-da warra-j that-nom woman-nom man-nom 3-pl-nom go-act ‘Those men and women are going.’
(Evans Reference Evans1995: 249 Kayardild)
(5) Ngayirni babi-rni ngiji-nginyi-nu kujkarrana yaminju-nu, nyu-rruku nyinawarra1 sg.erg older.brother-erg see-1du.ex-did two(m) shooting.star-did 2sg-went this.way ‘My brother and I saw two shooting stars when you'd gone.’
(Pensalfini Reference Pensalfini2003: 178 Jingulu)
(6) Pala-nga ngatu jarri-nya-pinti-ngi, mima-nikinyi-yi puluku, kujarra kangkuru-jirri waraja yalapara that-loc stationary inch-nm-ass-loc wait.for-impf-3pl.sub 3du.dat two kangaroo-du one goanna ‘And there, on the finishing line, the two kangaroos and one goanna waited for those two.’
(Sharp Reference Sharp2004: 315 Nyangumarta)
Consistent with the ‘free word order’ properties typical of many of these languages (Hale Reference Hale1983, Simpson Reference Simpson1991, Nordlinger Reference Nordlinger1998b, Austin Reference Austin and Simpson2001), such coordinations can also be discontinuous in some languages, as shown in (7) and (8) (compare these with the contiguous example from Kuuk Thaayorre in (3)).Footnote 3
(7) Ngul ngay kirk kempthe kal-m thul=yuk then 1sg(erg) spear(acc) apart carry-p.ipfv woomera(acc)=stuff ‘I used to carry spears and woomeras separately.’
(Gaby Reference Gaby2006: 320 Kuuk Thaayorre)
(8) Nganip-n ngancn yik-nhat nganam-un father-dat 1pl.ex(nom) say-p.pfv mother-dat ‘We said to Dad and Mum.’
(Gaby Reference Gaby2006: 320 Kuuk Thaayorre)
In addition to coordination, Australian languages are characterised by extensive use of nominal juxtaposition, exhibiting a substantial amount of flexibility as to how such nominal sequences are to be interpreted semantically (Blake Reference Blake1987, Reference Blake and Simpson2001; Dixon Reference Dixon2002). For example, many Australian languages make use of both generic–specific constructions and part–whole constructions in which two nominals are placed in syntactic juxtaposition and jointly determine the referent of the syntactic argument. Below we provide examples from a selection of languages – Kalkatungu, Kayardild and Yidiny – illustrating the use of juxtaposition to encode generic–specific and part–whole constructions, both across languages and within a single language. (9)–(14) illustrate generic–specific constructions.
(9) tjaa maa wartatji this vegetable.food orange ‘the/this orange’ (Blake Reference Blake and Simpson2001: 418 Kalkatungu)
(10) Ngayika ati-ntji ari-li thuwarr-ku I meat-dat eat-apass snake-dat ‘I'm eating snake.’
(Blake Reference Blake and Simpson2001: 419 Kalkatungu)
(11) Dathin-a dangka-a niya wumburung-kuru raa-ja wanku-ya kulkiji-y that-nom man-nom 3sg(nom) spear-prop spear-act elasmobranch-mloc shark-mloc ‘That man speared a shark with a spear.’Footnote 4
(Evans Reference Evans1995: 244 Kayardild)
(12) Dathin-a jardi-wuthin-da badi-ja jul-i wuran-ki that-nom mob-plenty-nom carry-act bone-mloc food-mloc ‘All those (ants) are carrying a bone.’
(Evans Reference Evans1995: 244 Kayardild)
(13) Gana mayi jimirr julaːlin try vegetable(abs) yam(abs) dig:going.imp ‘Go and try to dig some yams up!’Footnote 5
(Dixon Reference Dixon1977: 247 Yidiny)
(14) bama muulaːrri wulnggaːny banaː person(abs) initiated.man(abs) cover:past water:loc ‘The initiated men were drowned by the (rising) water.’
(Dixon Reference Dixon1977: 247 Yidiny)
In all of these examples we find a ‘generic’ nominal (e.g. ‘vegetable food’, ‘meat’, ‘person’) combined with a ‘specific’ nominal (e.g. ‘orange’, ‘snake’, ‘initiated man’) to jointly determine the reference of the NP as a whole. The appropriate English equivalent of these generic–specific constructions is really a single referential NP, as per the translation provided. These constructions, therefore, are not equivalent to non-restrictive appositional constructions in English such as ‘I am eating meat, namely snake’, or ‘I saw the elasmobranch, namely the shark’, in which one single nominal specifies or defines the reference and the other provides elaboration. Nor are they equivalent to such English constructions in discourse or pragmatic terms. Rather, the joint use of both nominals is, in these languages, a pragmatically neutral way to encode reference. See Wilkins (Reference Wilkins and Senft2000) for further discussion of this point in relation to the Australian data.
Part–whole juxtapositions operate similarly, as shown by the following examples:
(15) ngida wamburra tree trunk ‘tree trunk’
(Evans Reference Evans1995: 248 Kayardild)
(16) dangkaa ngabaya person spirit ‘person's spirit’
(Evans Reference Evans1995: 248 Kayardild)
(17) jugi gubu gana ngayu wanggi wawaːlna tree(abs) leaf(abs) try I(nom) up look:purp ‘I must try to look up at the leaves on the trees.’
(Dixon Reference Dixon1977: 248 Yidiny)
In the above examples of the part–whole construction we find juxtaposed nominals each bearing the same case marking, realising (part of) the same grammatical function, and jointly picking out the relevant referent for the NP as a whole.
In addition to generic–specific and part–whole constructions, nominal juxtapostion is very common in expressing a number of other construction types in which multiple nominals and pronouns are strung together in a single clause, jointly determining a single referent or having overlapping reference. Examples include the following:Footnote 6
(18) Garidi-ni bungmanyi-ni gin-amany yanybi husband-erg old.man-erg 3sg.m.a-p.twd get‘ (Her) old man husband came and got (her).’
(Nordlinger Reference Nordlinger1998a: 133 Wambaya)
(19) Pam-al ith nhul may carrots yakakerr man-erg that 3sg.ergveg carrots cut:rdp:p.pfv ‘The man cut up the carrots.’
(Gaby Reference Gaby2006: 290 Kuuk Thaayorre)
(20) Nhani mankada nhintha pani 3sg.f(nom) girl(abs) shame none ‘The girl is shameless.’
(Austin Reference Austin1981: 102 Diyari)
(21) yingu bama bunya bindam galiːny this(abs) person(abs) woman(abs) ‘name’(abs) go:past ‘This woman Bindam went.’
(Dixon Reference Dixon1977: 252 Yidiny)
(22) dathin-a dangka-a niya wirdi-j that-nom fellow-nom 3sg(nom) remain-act ‘That fellow (Kajurku) was waiting.’
(Evans Reference Evans1995: 278 Kayardild)
In (18)–(22) we again find strings of nominals and/or pronouns jointly expressing reference to a single entity (or single group of entities). All of these constructions, including the generic–specific and part–whole constructions exemplified above, are often described as ‘appositional’ in the Australianist literature (Heath Reference Heath1978, Reference Heath1984; Blake Reference Blake1979, Reference Blake1983, Reference Blake1987, Reference Blake and Simpson2001; Evans Reference Evans1995), where apposition is used in its broadest sense, as in the following traditional definition of this term (Crystal Reference Crystal1997: 24):
apposition: A traditional term retained in some models of grammatical description for a sequence of units which are constituents at the same grammatical level, and which have an identity or similarity of reference.
Examples such as (18) and (19) illustrate two straightforward variants of Australian-style ‘appositional’ constructions – a nominal-nominal appositional construction in (18), in which ‘old man’ is juxtaposed with ‘husband’ in subject function,Footnote 7 and a nominal–pronominal appositional construction in (19), in which the nominal elements ‘man’ and ‘that’ are juxtaposed with the coreferential third person pronoun nhul.Footnote 8 The use of such nominal appositions or juxtapositions is considerably less marked than appositional constructions in English (as in ‘Her husband, the old man, came and got her’, cf. (18)), and are usually best translated with simple non-apposed structures (e.g. ‘The man cut up the carrots’ in (19) and not ‘He, the man, cut up vegetables, being carrots’).Footnote 9 More generally, it is important to realise that the Australianist use of the term ‘apposition’ should not be taken to imply that these constructions are to be viewed as equivalent to the sorts of non-restrictive constructions often referred to as appositions in English.
Given the wide range of uses to which the term ‘apposition’ has been put in the literature, a terminological aside is important here to clarify our use of the term in this paper (and in the Australianist literature more generally). Note that we generally seek to avoid this particular terminological minefield in the present paper by using the more neutral term ‘nominal juxtaposition’ to describe the syntax of these constructions.
In the general literature on apposition a distinction is commonly drawn between restrictive and non-restrictive apposition (e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985), also known as close vs. loose apposition (e.g. Lekakou & Szendröi Reference Lekakou, Szendröi, Breheny and Velegrakis2007). Consider the pair of examples below. In (23a) the two terms of the apposition, Burns and the poet, serve jointly to determine the reference of the NP. The expression Burns itself does not pick out a unique entity; the restrictive phrase in a close (or restrictive) apposition restricts the denotation so that a unique referent is picked out. In (23b), on the other hand, Burns picks out a unique individual and the expression the poet makes a comment or provides further information about this individual. In English (and other languages) a loose apposition is set off intonationally, while in a close apposition the two elements consitute a single intonational unit.
(23)
(a) Burns the poet (close)
(b) Burns, the poet (loose)
There is an extensive body of literature which explores this distinction in a range of languages and different formal frameworks. Much of the syntactic discussion concerning non-restrictive, or loose, appositions has centered on non-restrictive relative clauses, and on whether the relative clause is integrated syntactically, and if so, how or at what level. The ‘radical orphanage’ approach (Fabb Reference Fabb1990, Espinal Reference Espinal1991, Peterson Reference Peterson, Butt and King2004) takes the non-restrictive modifier to be not integrated into the syntactic structure of the matrix at all, whereas integrated approaches generally take non-restrictive relative clauses to be adjoined to the nominal (Jackendoff Reference Jackendoff1977, Kempson Reference Kempson2003, Arnold Reference Arnold and Müller2004, Reference Arnold2007). Amongst the properties of loose or non-restrictive appositions which both integrated and non-integrated accounts have sought to accommodate are a range of interpretational facts: for example, loose appositions appear to escape the scope of sentence negation and that of propositional verbs. The (integrated) approach of Potts (Reference Potts2003, Reference Potts2005) captures this by putting the semantic content of (loose) appositional material in a separate dimension from which it is not accessible to the normal propositional content.
Although the prevalent use of the term apposition in the literature on European languages is to refer to non-restrictive nominal, clausal or other phrases (loose appositions), there has also been work on the nature of close or restrictive apposition and how it relates to modifier–head constructions both syntactically and semantically (Acuña-Fariña Reference Acuña-Fariña and Carlos1999; Keizer Reference Keizer2005, Reference Keizer2007; Lekakou & Szendröi Reference Lekakou, Szendröi, Breheny and Velegrakis2007).Footnote 10
This is an area which has been very little explored in Australianist work. It is clear from a look through a number of grammatical descriptions that there are examples which strongly suggest a close (restrictive) interpretation, but that equally, there are examples in which one nominal is set off from the sentence or phrase intonationally, suggestive of a loose or non-restrictive apposition. Unfortunately, however, the Australianist literature provides almost no discussion of how these types of appositions may differ syntactically.Footnote 11 In this paper, we restrict ourselves to examples of close appositions, such as the examples provided above, which we take to introduce semantic content which is integrated into the truth-conditional semantics. Our use of the term ‘apposition’ (or indeed nominal juxtaposition) should also be interpreted as referring to close appositions only.
Returning now to the Australian language data, a further type of juxtaposed construction common in Australian languages is the inclusory construction (Singer Reference Singer2001, Reference Singer2005) (also known in the literature as the ‘plural pronoun construction’, Schwartz Reference Schwartz, Barlow and Ferguson1988a), in which a plural pronoun referring to the superset is combined with a subset nominal. In many languages the inclusory construction involves simple juxtaposition of the two elements, as in the following example from Kayardild:
(24) Nga-rr-a kajakaja warra-ja thaa-th 1-du-nom daddy(nom) go-act return-act ‘Daddy and I will go.’ (lit. ‘We two, including Daddy, will go.’)
(Evans Reference Evans1995: 249 Kayardild)
Although inclusory constructions have much in common with the other juxtaposed constructions above, they differ in that the agreement features of the construction as a whole are those of only one of the constituent parts – namely the superset pronoun. We return to a discussion of inclusory constructions and their relationship to other juxtaposed constructions in section 6.
Thus, across Australian languages we find juxtaposed nominal constructions used with a range of functions beyond coordination, including generic–specific constructions, part–whole constructions, inclusory constructions and various types of nominal–nominal ‘appositions’. While the specific construction types and their properties can vary across different languages, the general phenomenon whereby sequences of nominals can have multiple interpretations including coordinated and non-coordinated meanings is pervasive among Australian languages and is frequently referred to in the literature as a common characteristic of Australian languages in general (e.g. Blake Reference Blake1987). Indeed, some researchers have gone so far as to claim for particular Australian languages that they have no noun phrases at all, with all nominals existing in apposition (in the broad sense of the term). This is the analysis given by Blake (Reference Blake1983) for Kalkatungu, for example, and by Heath (Reference Heath1978: 52) for Ngandi, in which ‘noun phrases which have more than one constituent are typically formed by apposition’. Even languages with clear NP structures for head–modifier relations also have a range of constructions in which nominals are seemingly juxtaposed without any evidence of syntactic asymmetry. For example, Evans (Reference Evans1995: 247), in discussing the Kayardild constructions exemplified in (11) and (15) above, states that ‘there are no syntactic reasons for considering one nominal to be the head, and it is better to treat them as apposed nominals’.
Significantly, all of these juxtaposed construction types share the syntactic properties of the asyndetic coordinated constructions in that neither nominal can be clearly identified as the head of the phrase. Evans (Reference Evans1995), for example, treats both nominals in the generic–specific and part–whole constructions as filling the ‘head’ slot in the NP structure (p. 235). He argues explicitly that the constructions are double-headed, on the grounds that either noun can function alone as the head of an NP so that there is no syntactic dependency, and because the nouns may appear in either order (pp. 244–248). Similar arguments have been made for analogous constructions in other Australian languages such as Kuuk Thaayorre (Gaby Reference Gaby2006: 283), Kalkatungu (Blake Reference Blake1983) and Mparntwe Arrernte (Wilkins Reference Wilkins and Senft2000).Footnote 12
Indeed, the only clear formal difference between coordinated constructions and other juxtapositions is that the former show resolved agreement on the verb (in languages with verbal agreement) whereas the latter do not.Footnote 13 This difference arises from the fact that the two nominals in a coordinated construction establish distinct referents, whereas the two nominals in other types of juxtapositions combine to identify a single referent. Consider, for example, the Wambaya example (25) below, repeated from (18). The fact that the auxiliary (gin-amany) shows singular number agreement forces the interpretation of the juxtaposed nominals as describing the same referent. If the auxiliary had dual number agreement here (gurl-amany), the NP would be interpreted as a coordination (i.e. ‘(her) husband and the old man’).
(25) Garidi-ni bungmanyi-ni gin-amany yanybi husband-erg old.man-erg 3sg.m.a-p.twd get‘ (Her) old man husband came and got (her).’
(Nordlinger Reference Nordlinger1998a: 133 Wambaya)
Note that verbal agreement in languages like English similarly disambiguates between true coordination and so-called boolean coordination in examples such as My wife and the mother of my children is/are singing. We return to a discussion of boolean coordination in section 7. Of course in both the Wambaya and the English cases, verbal agreement cannot disambiguate if both elements are themselves plural.
Apart from this semantically motivated difference (concerning the agreement features), all of these types of juxtaposed constructions – coordinations, generic–specific constructions, part–whole constructions, inclusory constructions and other types of nominal–nominal appositions – have the same basic syntactic structure in that they consist of a sequence of two nominals fulfilling the same grammatical function, neither of which is syntactically dependent upon the other. In fact, they all satisfy a broad definition of coordination, such as the following from Haspelmath (Reference Haspelmath and Shopen2007: 1):
The term coordination refers to syntactic constructions in which two or more units of the same type are combined into a larger unit and still have the same semantic relations with other surrounding elements.
This suggests that the appropriate analysis for all of these juxtaposed nominal constructions is as a multiply-headed structure in the syntax, in which each element is independently fulfilling the same grammatical function in parallel, similar to the analysis usually assumed for coordination (see Blake Reference Blake1983: 171–172 for a proposal along similar lines). In the following sections we will sketch out such an analysis in LFG and show how it provides an explanatory account of the Australian language data. We begin with a discussion of the standard treatment of coordination in LFG.
3. LFG analysis of coordination
LFG is a lexicalist constraint-based syntactic framework which posits two co-present levels of syntactic representation: c-structure, which encodes phrase structural relations (of precedence and dominance) between elements, and f-structure, which represents a level of internal syntactic structure based on grammatical functions, modelling predicate argument relations. C-structures are represented by phrase structure trees of a familiar sort while f-structures are represented by attribute value matrices. C-structures and f-structures (corresponding to a given string) are interrelated by means of a mapping function φ, which places elements of c-structure (in the domain of the function) in correspondence with f-structures. The mapping is expressed by means of equations (or f-descriptions) associated with lexical items and phrase structure nodes (phrase structure rules in LFG are annotated with such equations). The relation between c-structure and f-structure is many-to-one and into (f-structures can arise which are not related to specific c-structure nodes, and more than one c-structure node can correspond to a single f-structure). F-structures are subject to well-formedness constraints, notably those of completeness and coherence, which ensure that all and only the arguments of a predicate are present (as values of subcategorisable grammatical functions) in the f-structure projected by a predicate. For comprehensive introductions to the formalism see Bresnan (Reference Bresnan2001), Dalrymple (Reference Dalrymple2001) and Falk (Reference Falk2001).
This approach provides a highly modular and flexible framework for syntactic description: in particular, it has proved sufficiently flexible to be an appropriate tool for modelling the syntax of widely divergent languages ranging from the highly configurational to the radically non-configurational. A significant body of work has developed LFG analyses of phenomena in Australian Aboriginal languages (Simpson & Bresnan Reference Simpson and Bresnan1983, Simpson Reference Simpson1991, Austin & Bresnan Reference Austin and Bresnan1996, Nordlinger Reference Nordlinger1998b, Wilson Reference Wilson1999).
A simple sentence such as (26) would have the f-structure (27). F-structures are simple attribute-value matrixes, where values can be atomic or complex. A particular feature of LFG is the use of sets as values of attributes (that is, the value of the attribute is a set of f-structures), as in the case of the grammatical function adjunct in (27), alongside atomic and complex-valued features (tense and subj respectively in (27)).
(26) Kim wept yesterday.
(27)
When a property is asserted to hold of a set, its behaviour with respect to the set will depend on the nature of the particular property. The majority of properties are distributive: such a feature is an attribute of every member of the set. The grammatical functions are distributive; thus Kim is the subj of both the f-structure of shout and the f-structure of cry in the (set of) f-structures for Kim shouted and cried. As we will see below, agreement features and the conjform feature are non-distributive: when a feature is non-distributive, it and its value are a property of the set as a whole.
A standard analysis of NP coordination in LFG (Dalrymple & Kaplan Reference Dalrymple and Kaplan2000, Dalrymple Reference Dalrymple2001) assumes a c-structure coordination schema along the lines of (28) in which each coordinand is defined as belonging to a set-valued f-structure as specified by the annotation ↓∊↑.Footnote 14
(28)
The representation of a coordinate structure involves a hybrid object, that is, a set which additionally may itself have properties (the non-distributive properties referred to above) alongside the elements or members of the set. Consider the representation of (29), given in (30):
(29) John and I met.
(30)
In (30), each conjunct contributes an f-structure (labelled j and i in (30)) to the set of f-structures which is the value of the subj attribute. Additionally, the subj (labelled ji) has some features which express properties of the set as a whole – in this example, the feature conjform with the value and and the index feature which expresses the person/number/gender (png) agreement features of the coordinate structure as a whole (as noted above, agreement features are taken to be non-distributive).
A common pattern of agreement with coordinate NPs involves syntactic resolution, whereby the agreement features of all the conjuncts are taken into account in ‘calculating’ the agreement features of the coordinate structure as a whole. Dalrymple & Kaplan (Reference Dalrymple and Kaplan2000) develop an approach to syntactic feature resolution using set-valued agreement features and the simple operation of set union. Thus, for example, if pers features are represented as shown in (31), set union gives the standard resolution pattern for this feature.Footnote 15 Features such as conjform and index are non-distributive (i.e. resolving) while most features, including grammatical functions such as subj and obj, as already noted, are distributive.Footnote 16
(31)
Similarly, a two-gender (m, f) system with resolution to the masculine works as follows (with masc corresponding to the set {m} and fem to the empty set):
(32)
As shown in (33), the coordination schema for NP coordination in a language with syntactic feature resolution involves simple f-descriptions which ensure that the pers and gend features of each NP conjunct are a subset of the pers and gend features of the whole set (in the following, ind abbreviates index).Footnote 17
(33)
Resolution of the num feature, on the other hand, is not purely syntactic, as shown by the mimimally contrasting examples in (34), to which we return briefly in section 7.
- (34)
(a) The president and chief executive are attending the meeting in Beirut.
(b) The president and chief executive is attending the meeting in Beirut.
In LFG it is possible to associate a collection of f-descriptions (or equations) with a name in a template definition. The rule schema can then call the template (a template call is denoted in the formalism by the symbol @). This simple and convenient abbreviatory device has a number of useful properties. For example, because templates can call other templates, they can be organised to express linguistic generalisations succinctly (see Dalrymple et al. (Reference Dalrymple, Kaplan, King, Butt and King2004) for further discussion). The pers and gend equations for NP coordination can be expressed in a template:
(35)
We can therefore represent (33) more simply as in (36), in which each NP has been associated with the coordination template in (35):
(36)
4. Modelling nominal juxtaposition
In this section we show how the general LFG approach to coordination described above can be built upon to account for the full range of juxtaposed nominal constructions in Australian languages, discussed in section 2. We first show how the basic syntactic analysis of coordination can be extended to account for asyndetic coordination (section 4.1) and other non-coordinated juxtaposed constructions (section 4.2). We argue that these constructions can all be treated as essentially the same at f-structure, with the differences between them captured in the mapping to the semantics. A preliminary semantic treatment of some of the construction types is provided in section 5.Footnote 18
4.1. Juxtaposition as asyndetic coordination
As we saw in section 2, NP coordination in Australian languages is frequently encoded by juxtaposition, without any explicit marker of coordination. It seems reasonable to assume that such coordinate structures receive precisely the same syntactic treatment as (29) above, so that they differ in the presence (or absence) of a coordinator. This suggests the c-structure schema in (37) for asyndetic nominal coordination. In (37) X is a categorial metavariable ranging over N and NP;Footnote 19 otherwise it differs from the standard schema in (36) above in lacking a conj feature.Footnote 20 We defer full discussion of the agreement constraints (for the non-distributive index features) to section 5, but assume for the moment that the agreement template np-cnjt is associated with each daughter node, just as in (36).
(37)
The f-structure corresponding to the coordinated subject in (6) (repeated here as (38)) is then that in (39), with resolved index features but no conj feature in the outer f-structure.
(38) Pala-nga ngatu jarri-nya-pinti-ngi, mima-nikinyi-yi puluku, kujarra kangkuru-jirri waraja yalapara that-loc stationary inch-nm-ass-loc wait.for-impf-3pl.sub 3du.dat two kangaroo-du one goanna ‘And there, on the finishing line, the two kangaroos and one goanna waited for those two.’
(Sharp Reference Sharp2004: 315 Nyangumarta)
(39)
Recall from section 2 that coordination structures can be discontinuous, as in the following examples. That these are truly coordinations (and not afterthoughts, for example) is readily established by such things as discourse context (Blake Reference Blake and Simpson2001) and the fact that the verb shows resolved agreement, as in (40).
(40) Kintja-(ng)ku=yana intji-mi-ngi-yu ntiya-(ng)ku tjipa-yi kurlayingu-thu female-erg=and pelt-fut-me-they.du stone-erg this-erg male-erg ‘The girl and boy will both pelt me with stones.’
(Blake Reference Blake and Simpson2001: 423 Kalkatungu)
(41) Ngul ngay kirk kempthe kal-m thul=yuk then 1sg(erg) spear(acc) apart carry-p.ipfv woomera(acc)=stuff ‘I used to carry spears and woomeras separately.’
(Gaby Reference Gaby2006: 320 Kuuk Thaayorre)
Our analysis combines straightforwardly with the standard approach to discontinuity and non-configurationality in LFG to account for these cases. In LFG c-structure nodes are generally optional – by the Principle of Economy, only those nodes which are motivated by overt lexical material (or by some semantic requirement) are present (Bresnan Reference Bresnan2001).
In an example like (41) the two parts of the coordinate structure appear separated by the verb and its modifier: each is represented at c-structure as an instance of the coordinate structure schema (37) with just one daughter (42).
(42)
In non-configurational structures, the grammatical functions to be assigned to an element of c-structure are not identified in configurational terms but by morphological means. In the example at hand, it is the case marking on the nominals which indicates the subj (marked erg) and the (discontinuous) obj (marked acc). Since in principle a wide range of different grammatical function assignments might be compatible with a string of categories we assume NPs under S are freely annotated with the equation (↑gf)=↓, where gf is a meta-variable over the relevant set of grammatical functions. It is the case markers which identify which grammatical function a nominal element maps to (Simpson Reference Simpson1991, Austin & Bresnan Reference Austin and Bresnan1996, Nordlinger Reference Nordlinger1998b).
The two acc-marked NPs under S will both independently co-specify the obj f-structure (as shown by the labels a and b in (42)), and hence they both contribute elements to the (coordinate) set.
(43)
If we choose to associate some other functional annotation with the acc-marked NPs (treating one or both of them as contributing the whole obj function, for example), no complete and coherent f-structure would be produced. Our analysis of asyndetic coordination thus extends to account straightforwardly for the possibility of discontinuous coordinations in these languages.
4.2. Juxtaposition beyond coordination: a common syntax
In section 2 we saw that many Australian languages use nominal juxtaposition to encode a range of non-coordinated construction types, including generic–specific, part–whole, and nominal–pronoun appositions. These constructions show the same syntactic properties as asyndetic coordination: the juxtaposed nominals do not stand in a syntactic dependency relation with one another and they fulfill the same grammatical function. Our proposal is that the full range of nominal juxtapositions discussed in section 2 all have the same surface syntax, and thus should be given the same syntactic analysis (apart from the non-distributive agreement features) – namely, as hybrid structures at f-structure. We are therefore arguing that the LFG analysis of coordination be extended to these other juxtaposed constructions also, with the differences between the construction types captured in the mapping to the semantics. Not only does this approach provide a unified account of nominal juxtaposition in Australian languages – both coordinated and non-coordinated – but representation of non-coordinated juxtaposed nominals as members of a set directly captures the intuition expressed by many researchers that these nominals are co-heads (e.g. Blake Reference Blake1983, Evans Reference Evans1995, Wilkins Reference Wilkins and Senft2000, Gaby Reference Gaby2006).Footnote 21
On our analysis, the f-structure corresponding to the nominal juxtaposition in (18), repeated in (44), is as in (45). Apart from the value of the non-distributive (index) features of the set (which we discuss below), the f-structure in (45) is structurally identical to that associated with a coordination such as (39) repeated as (46).
(44) Garidi-ni bungmanyi-ni gin-amany yanybi husband-erg old.man-erg 3sg.m.a-p.twd get‘ (Her) old man husband came and got (her).’
(Nordlinger Reference Nordlinger1998a: 133 Wambaya)
(45) Apposition:
(46) Coordination:
This analysis directly reflects the fact that there is no visible syntactic distinction within the nominal strings themselves between nominal coordination and other types of nominal juxtaposition.Footnote 22 In fact, as discussed in section 2 above, the nominal phrase in (44) is itself ambiguous between a coordinate and an appositional interpretation, disambiguated only by the verbal morphology. The auxiliary form gin-amany ‘3sg.m.a-p.twd’ determines that the subj is 3sg. If this example meant ‘the old man and her husband (they) …’ then the finite auxiliary would be encoded with 3du. Crucially, the formal differences lie only in the agreement features of the set; there is no visible syntactic distinction within the nominal structure itself. Thus, as far as the syntax is concerned, our analysis needs to be able to account for the fact that the same nominal f-structure may sometimes involve feature resolution (i.e. in a coordination structure), and sometimes not (i.e. in an appositional structure).
Thus, we propose that both appositional and coordinate constructions are licensed by the basic phrase structure schema in (47), with differences arising in terms of the overall agreement features of the structures and in the semantics, as we shall soon see. Appositional and coordinate structures differ syntactically, on this view, purely in terms of the overall agreement features of the structure as a whole. Structurally, they are identical.
(47)
Recall from section 3 above that feature resolution in coordination structures is licensed by the np-cnjt template, repeated here from (35), which is associated with each NP in the coordinated phrase:
(48)
In non-coordinated juxtapositions such as (44), on the other hand, the juxtaposed constituents are co-referential and there is no feature resolution at the level of the set: the features of the set are the same as the features of each of the members. Thus, in our terms, these constructions generally involve index sharing between the set and the members of the set. We define the ‘appositional’ template in (49), which is associated with each of the daughter constituents in the above phrase structure rule in a non-coordinated juxtaposed construction, as in (50).Footnote 23 This template ensures that the index features of each daughter constituent are shared with the index features of the set (i.e. a set containing two 3sg daughters will likewise have 3sgindex features).
(49) np-appos: (↓ ind)=(↑ ind)
(50)
Thus, the only structural difference between the coordinated phrase in (51) and the non-coordinated phrase in (52), both repeated from above, is that the two members of the coordinated phrase are associated with the np-cnjt template (as in (37)), and the two members of the non-coordinated phrase are associated with the np-appos template (as in (50)), resulting in different index features for the set as a whole. In all other respects, the two are structurally identical.Footnote 24
(51) Pala-nga ngatu jarri-nya-pinti-ngi, mima-nikinyi-yi puluku, kujarra kangkuru-jirri waraja yalapara that-loc stationary inch-nm-ass-loc wait.for-impf-3pl.sub 3du.dat two kangaroo-du one goanna ‘And there, on the finishing line, the two kangaroos and one goanna waited for those two.’
(Sharp Reference Sharp2004: 315 Nyangumarta)
(52) Garidi-ni bungmanyi-ni gin-amany yanybi husband-erg old.man-erg 3sg.m.a-p.twd get‘ (Her) old man husband came and got (her).’
(Nordlinger Reference Nordlinger1998a: 133 Wambaya)
Finally, note that all of the non-coordinated juxtaposed construction types (generic–specfic, nominal–pronoun, etc.) can be also discontinuous, as exemplified with the generic–specific construction in (53).
(53) Ngayika ati-ntji ari-li thuwarr-ku I meat-dat eat-apass snake-dat ‘I'm eating snake.’
(Blake Reference Blake and Simpson2001: 418 Kalkatungu)
This will follow directly from the analysis of discontinuous coordinate constructions discussed in section 3 above. Namely we assume that each element in the c-structure rule is optional, under the Principle of Economy of Expression (Bresnan Reference Bresnan2001: 91), thereby allowing each nominal to constitute an NP on its own in the c-structure, while still contributing to a set at f-structure.
5. Computing meanings: semantic composition
We have argued that both coordinated and non-coordinated juxtapositions are rightfully analysed as structurally identical, with the syntactic differences lying only in the relationship between the index features of the set and those of the member elements, and in the mapping to the semantics. In this section we present a semantic analysis of some of these construction types and show how it integrates with the syntactic analysis presented above. We show how it is possible in LFG to adopt a single syntax and distinguish the construction types in the mapping to the semantics. Our focus is not so much on the details of the semantic analysis per se, about which much more could be said, but rather on the interface itself.Footnote 25
5.1. Coordinate meanings
As we have seen, nominal juxtapositions can have coordinate meanings, involving syntactic feature resolution and the construction of a coordinate semantics. Coordinate agreement can be captured in the template (35), repeated below as (54), following Dalrymple & Kaplan (Reference Dalrymple and Kaplan2000); this will suffice for present purposes as our focus is not on the details of agreement.Footnote 26 This template is associated with each constituent in the phrase structure rule, as in (55), where we have coordinate readings.
(54)
(55)
We turn now to the semantics, and in particular to semantic composition, that is, how meanings are associated with (and derived from) the syntactic structures we are discussing. It is generally assumed in LFG that semantic composition involves reference to f-structures, that is, to representations of syntactic predicate-argument structure, rather than to c-structures. Thus despite the widely varying c-structures that languages have, their semantics is largely invariant, in that the same sorts of meanings are encoded by a diverse range of external syntactic structures.Footnote 27
The predominant approach to semantic composition in LFG, which we adopt here, uses linear logic for meaning assembly. The meanings themselves will generally be represented as simple predicate logic expressions. These can be viewed as an abbreviatory shorthand for more elaborate meaning expressions. Our focus here is primarily on the process of meaning assembly, that is, how meanings are associated with syntactic structures, and in particular on the syntax-semantics interface.
In the glue (linear logic) theory of semantic composition, instructions for combining meanings are stated as premises in a logical deduction. The order of composition is therefore determined only as the logic itself determines it (and thus details of syntactic structure, such as the phrasal composition, are not important). A crucial aspect of the use of glue, or linear logic, for meaning assembly is its resource sensitivity: once a premise is used, it is used up and no longer available for subsequent steps in the deduction. The approach is therefore motivated by the observed resource sensitivity of natural language interpretation. A highly accessible introduction to meaning composition in LFG is provided by Dalrymple (Reference Dalrymple2001: chapter 9), on which this brief outline is based. In this approach meaning constructors are associated primarily with lexical items (though they can also be associated with phrase structure nodes in rules).
Meaning constructors have two parts: they are made up of a meaning on the left-hand side of the colon and a logical formula over semantic structures corresponding to that meaning on the right-hand side of the colon (semantic structures are related to f-structures by the projection σ). Consider as an example the meaning constructor associated with the lexical entry for yawned given in Dalrymple (Reference Dalrymple2001: 233) and shown in (56).
(56) yawned, V: (↑ pred)=‘yawn <subj>’
λX.yawn(X): (↑ subj)σ⊸↑σ
The meaning side of the constructor in (56) gives the meaning of yawn as one-place predicate (of course, other more complex meaning representations can be substituted for the predicate logic expressions used here, if desired). Meaning expressions are typed (so that the constant Kim, for example, is of type e): type information will determine how meaning expressions combine with others. The glue side uses linear implication: (⊸) is an implication which can be read as saying that if the meaning resource for the subj is available then it can be consumed to produce the meaning of the sentence.Footnote 28 Finally, it is standard practice to introduce labels as names of meaning constructors to ease reference to them, a practice we will adopt. Thus the meaning constructor for the individual Kim might be labelled as Kim:
(57) KimKim: ↑σ
Dalrymple (Reference Dalrymple2001) provides an analysis of the semantics of NP coordination which associates the semantic contribution g-and (group-forming and) in (58) with the coordinator and in its lexical entry (60). In the meaning constructor, recall that the left-hand side is a meaning expression and the right-hand side is a glue constructor for resource-sensitive semantic construction using linear implication. On the meaning side, the lambda expression denotes the group-forming function – here, a function from two individuals to the group containing those individuals. On the glue side, g-and consumes the semantics of one conjunct (↑∊)σ and produces a function from the semantics of the other conjunct to the semantics of the coordinate structure as a whole. For more-than-binary coordination, a further semantic contribution g-and2, involving the ! (‘of course’) operator, can be used any number of times (including zero), each time adding an individual into the group (59). Since our primary focus here is on binary coordination, we will have no more to say about coordination involving multiple conjuncts.
(58)
(59)
(60)
Following Dalrymple (Reference Dalrymple2001) and the arguments presented therein, we take NP coordination to be correctly characterised in this way as group formation. However, in the case of asyndetic coordination, there is no coordinator in the structure with which to associate the semantics of g-and.
Notice also that in languages (such as these) with a three-way number distinction (singular, dual and plural), it is not possible simply to associate the use of the group-forming semantics with num resolution to pl, because the syntactic num of a group containing just a pair is du. For present purposes, we restrict ourselves to binary coordination, and define the num resolution as in the template in (61). This captures the generalisation that either the overall number is du (i.e. when two singular nominals are coordinated), or (at least) one of the constituents is non-singular, in which case the overall number is pl. Note that since this makes reference to elements of the (coordinate) f-structure, rather that to c-structure daughters, num resolution will operate correctly in cases of discontinuous coordination as well as cases of contiguous coordination by simple juxtaposition.
(61) binary: {(↑∊index num)≠sg ∧ (↑ index num)=pl}
| (↑ index num)=dual
To complete the interpretation of nominal juxtapositions as coordinate, we need to associate the template binary and the meaning constructor g-and with the phrase structure rule in (55) (restricting our attention to cases of binary coordination). Since there is no coordinator, in asyndetic coordination we arbitrarily associate these with one of the daughter constituents. The phrase structure rule for juxtaposed NPs in coordination is therefore that in (62).Footnote 29
(62)
Our analysis of juxtaposition with coordinate semantics is thus analogous to the analysis of (non-juxtaposed) coordinate constructions in other languages (Dalrymple and Kaplan Reference Dalrymple2001, Dalrymple Reference Dalrymple2001). In the next section we see how this same general approach can also provide an analysis of the Australian ‘appositional’ juxtapositions exemplified in section 2 above.
5.2. Non-coordinate meanings of juxtaposition
We have argued that the entire range of nominal juxtapositions all share the same basic f-structure. In the case of juxtapositions interpreted as close appositions, each constituent is associated with the np-appos template, which ensures identity between the index features of each constituent and the index features of the set as a whole:
(63) np-appos: (↓ ind)=(↑ ind)
(64)
As for the semantics of appositional juxtapositions such as (44), as a first approximation we take this to be basically intersective (applying to property-denoting nominal (rather than NP) meanings). One way of doing this would be to propose a meaning constructor rather comparable to boolean and (as in the joint reading of five linguists and philosophers), taking two sets of properties and intersecting them (see Dalrymple Reference Dalrymple2004):
(65) b-and λX.λY.X⊓Y
An alternative approach, which is the one we will follow here, is to model the semantics of close apposition on the semantics of nominal modification, as follows:
(66) appos λQ.λP.λx.Q(x) ∧ P(x):
[(%nom1σvar) ⊸ (%nom1σrestr)]⊸
[[(%nom2σvar) ⊸ (%nom2σrestr)]
⊸ [(↑σvar) ⊸ (↑σrestr)]]
%nom1∊↑
%nom2∊↑
On the meaning side, this is a function which applies to two nominal (<e, t>) meanings and produces an abstraction over a logical conjunction of predications holding of the given individual (so it takes two nominal meanings and produces a nominal meaning, where nominal meanings are of type <e, t>). On the glue side, the meaning constructor consumes first one nominal contribution and then the other nominal contribution to produce the meaning of the structure as a whole. Note that the meaning which results from this process is a nominal meaning, that is a property or function of type <e, t>, rather than a generalized quantifier or typical DP meaning. This meaning cannot of course be consumed directly by a verbal meaning constructor (given standard assumptions about the latter), but would need to be type-shifted (or the equivalent) to produce a full referential NP meaning. This is consistent with the fact that in these languages a bare nominal may be interpreted predicatively, but may also be interpreted as a full NP in appropriate context.
We can therefore complete our analysis of appositional juxtaposition by arbitrarily associating the appos semantics with some daughter in the appositional phrase structure rule:
(67)
In order to see how this works, consider the nominal apposition in the now-familiar Wambaya example (18). The semantics associated with each of the nominals in this construction is given in (68) and (69).
(68) garidi-ni (husband-erg):
λX.husband(X): (↑σvar) ⊸ (↑σrestr)
(69) bungmanyi-ni (old.man-erg):
λX.old.man(X): (↑σvar) ⊸ (↑σrestr)
The meaning constructor (66), associated with the appositional use of the juxtaposition schema, consumes (68) and (69) to produce another nominal meaning, as follows:
(70) garidi-ni bungmanyi-ni (husband-erg old.man-erg):
λX.old.man(X) ∧ husband(X):
(↑σvar) ⊸ (↑σrestr)
Note that in these languages, a bare nominal such as (68) or (69) (or indeed (70)) may be interpereted predicatively, but may also be given a range of referential (or entity-denoting) NP meanings in context (e.g. ‘the old man’, ‘an old man’, ‘old men’). Pronouns and demonstratives may accompany the nominal in ‘determinizing’ function but are by no means obligatory in the production of full (referential) NP meanings. In the latter cases, where there are no demonstratives or pronouns, we take it that additional meaning constructors (not associated with lexical material) must be available to lift nominals into the appropriate range of NP meanings.Footnote 30
Other construction types expressed by means of juxtaposition and discussed in section 2, such as generic–specific and part–whole constructions, can also be straightforwardly accounted for by the approach suggested in this paper, although there is certainly more to be said on the details of their semantic analysis. These constructions are likewise licensed by the non-coordinate phrase structure rule (67), which is fully consistent with the consensus view in the Australianist tradition that treats such constructions as consisting of apposed nominals (e.g. Heath Reference Heath1978, Blake Reference Blake1987, Evans Reference Evans1995, etc.).
The f-structure corresponding to the juxtaposed (generic–specific) construction in (71) (= (11)) is given in (72).
(71) Dathin-a dangka-a niya wumburung-kuru raa-ja wanku-ya kulkiji-y that-nom man-nom 3sg(nom) spear-prop spear-act elasmobranch-mloc shark-mloc ‘That man speared a shark with a spear.’
(Evans Reference Evans1995: 244 Kayardild)
(72)
Standard (nominal) meaning constructors along the lines of (68) are given for wanku-ya (elasmobranch-mloc) and kulkiji-y (shark-mloc) in (73) and (74). These combine with the appositional meaning constructor to yield (75):
(73) wanku-ya (elasmobranch.mloc):
λx.elasmobranch(x): (↑σvar) ⊸ (↑σrestr)
(74) kulkiji-y (shark.mloc): λx.shark(x): (↑σvar) ⊸ (↑σrestr)
(75) wanku-ya kulkiji-y (elasmobranch-mloc shark-mloc):
λX.elasmobranch – fish(X) ∧ shark(X): (↑σvar) ⊸ (↑σrestr)
In the case of generic–specific constructions, there is an additional relationship between the properties that the nominal predicates introduce, in that one of the nominals (the ‘generic’ term) is (typically) a hypernym whose denotation properly includes that of the other (‘specific’) term. We abstract away from this here, but assume that this further semantic restriction on the construction could be captured by an additional meaning postulate specifying that an appropriate relationship must hold between the two nominal restrictor properties.
6. Templatic mismatch: inclusory constructions
Above we have argued that the range of juxtaposed nominal constructions that we find in many Australian languages – including coordination, close apposition, generic–specific constructions, part–whole constructions and nominal–pronoun combinations – can be accounted for with a modular approach in which these constructions have the same surface syntactic structure as f-structure sets, with the differences resulting from feature resolution or identity, and in the mapping to the semantics. We have proposed a single basic phrase structure rule (76), and two alternative sets of annotations (specifying functional equations and meaning constructors) corresponding to coordinate constructions and non-coordinate constructions respectively, as laid out in (77).
(76) X→ X X
↓∊↑ ↓∊↑
- (77)
(a) Annotate each daughter @np-cnjt and some daughter @binary and g-and; OR
(b) Annotate each daughter @np-appos and some daughter appos
These sets of annotations ensure that the coordinated constructions involve feature resolution and a coordinate semantics, while the non-coordinated constructions involve feature identity between the set and its members, and a close appositional semantics.
However, the existence of a single ‘coordinate’ phrase structure rule and two different templates governing the interaction between the index features of the set and its members allows for the possibility that there may be a mismatch between the templates associated with each constituent of the ‘coordinated’ phrase. In other words, it is possible for one constituent to be associated with the np-cnjt template and the other with the np-appos template. In this section we show that this possibility is, in fact, exactly what we find in inclusory constructions.Footnote 31
As already discussed in section 2 above, the inclusory construction is another type of nominal juxtaposition structure common to many Australian languages. An inclusory construction typically involves two juxtaposed elements, one a pronominal referring to the group as a whole and the other a (pro)nominal picking out a subset of the group. Examples include the following:
(78) Tjirlpi-lu nyupali kati-ku-nti old.man-erg.name 2du.erg take-fut-maybe ‘You and the Old Bloke might take (us).’ (lit. ‘you two, including the Old Bloke, might take (us)’)
(Goddard Reference Goddard1985: 101 Yankunytjatjara)
(79) Nga-rr-a kajakaja warra-ja thaa-th I-du-nom daddy(nom) go-act return-act ‘Daddy and I will go’ (lit. ‘we two, including daddy, will go’)
(Evans Reference Evans1995: 249 Kayardild)
Inclusory constructions (also called ‘plural pronoun constructions’ in the literature) are found in languages from many different families (Schwartz Reference Schwartz, Barlow and Ferguson1988a, Reference Schwartz, Hammond, Moravcsik and Wirthb; McNally Reference McNally1993; Lichtenberk Reference Lichtenberk2000; Bril Reference Bril and Haspelmath2004), and are described for Australian languages in Singer (Reference Singer2001, Reference Singer2005).Footnote 32 The correct analysis of the inclusory construction has been the source of some debate in the literature. Schwartz (Reference Schwartz, Barlow and Ferguson1988a, Reference Schwartz, Hammond, Moravcsik and Wirthb) analyses the inclusory construction as deriving from a coordination construction, as do Hale (Reference Hale1966, Reference Hale, Anderson and Kiparsky1973) and Nash (Reference Nash1986) in their analysis of inclusory constructions in the Australian language Warlpiri. Singer (Reference Singer2001), on the other hand, argues that the inclusory construction in Australian languages is a distinct construction type, albeit similar in some respects to coordination constructions, based on the fact that the inclusory construction is an endocentric construction with the features of one of the elements (i.e. the superset pronominal) being identical to the features of the whole argument. This is shown most clearly in a language with verbal agreement, as in example (80) from Nunggubuyu:
(80) nurru=wang maːgurn nurru we.ex.pl=killed.it Magurn we.ex.pl‘Magurn and us killed it (buffalo).’
(Heath Reference Heath1984: 542 Nunggubuyu)
In terms of their surface syntax, (Type 1) inclusory constructions generally consist of two juxtaposed nominal elements, and are therefore syntactically similar to the other juxtaposed constructions we have discussed above. Indeed, many language descriptions treat them as a type of ‘appositional’ construction similar to part–whole and/or generic–specific constructions (e.g. Dench Reference Dench1995, Evans Reference Evans1995, among others). We therefore propose that they should likewise be analysed as sets at f-structure, therebycapturing their syntactic similarity with apposition and coordination constructions.
Inclusory constructions are a particularly interesting case, as the features of the set overall are identical to the features of one member of the set (the pronominal), in which the features of the other member must be included. Thus, inclusory constructions are a composite of the coordination and appositional schemas presented in (62) and (67) above. The constituent corresponding to the superset pronominal carries the appositional template (specifying that its index features are identical to the index features of the whole) and the constituent corresponding to the subset member carries the coordination template (specifying that its index features must be a subset of the index features of the whole).Footnote 33
(81)
(82)
Thus, our analysis of nominal juxtaposition in Australian languages extends directly to inclusory constructions, correctly capturing the fact that these various juxtaposed construction types are syntactically similar in many Australian languages.Footnote 34 Moreover, this analysis captures the various characteristics of the inclusory construction discussed in the literature. The fact that inclusory constructions are similar to coordination constructions is captured by the fact that, like coordination constructions, inclusory constructions are represented as sets at f-structure. The endocentricity of the inclusory construction (Singer Reference Singer2001), namely the fact that the features of the whole argument are identical to those of the superset pronominal, is captured by having the pronominal carry the np-appos template, which specifies that the features of the individual member are carried up to the set as a whole. The inclusory nature of the construction is captured by the fact that the other element must have features which form a subset of the features of the entire set.Footnote 35
Treating inclusory constructions as sets at f-structure is further supported by the fact that in some Australian languages they can be marked with suffixes that explicitly indicate set membership. In Yidiny the ‘one of a group’ suffix -ba is used in both coordination constructions (83) and inclusory constructions (84):
(83) Darnggidarnggiːba yaburruba galing old.woman:ba(abs) young.girl:ba(abs) go:pres ‘An old woman (being one of a group of people) and a girl (being another member of a group) are going.’
(Dixon Reference Dixon1977: 177 Yidiny)
(84) Nganytyi bunyaːba galing inon-sing woman:ba go:pres ‘A woman and I (and some others) are going.’
(Dixon Reference Dixon1977: 178 Yidiny)
Thus, our analysis of juxtaposition and asyndetic coordination in Australian languages extends naturally to inclusory constructions as well, capturing the similarities amongst these construction types and providing a unified analysis of the NP juxtaposition that is so prevalent in these languages.
7. Conclusion and further implications
We have shown how the range of juxtaposed NP constructions in Australian languages can be accounted for relatively straightforwardly within the constraint-based lexicalist formalism of LFG. We have developed an account in which nominal–nominal sequences all have essentially the same f-structure, but correspond to three different feature association patterns, as in (85)–(87), and map onto a range of different semantics correlated with these three different patterns. In the case of coordinate constructions, the index features of the elements in the set-valued f-structure stand in a subset relation to the index of the set itself. In the case of appositional constructions, the index features of the individual elements are equivalent, and are also equivalent to those of the set itself. Inclusory constructions involve a hybrid of the two, whereby one element of the set has the same index features as the set itself, while the other element in the set stands in a subset relation to the overall index.
(85) coordination – X⊇Y,Z
(86) apposition – X=Y=Z
(87) inclusory – X=Y⊇Z
The flexible architecture of LFG thus provides a unified syntactic account of a range of juxtaposed nominal constructions common in Australian languages, while still capturing their semantic differences. In this paper we have shown how the use of hybrid f-structures can be extended beyond true (semantically) coordinated constructions to generic–specific, part–whole and other types of close appositional constructions, making a distinction between syntactic coordination (hybrid structures) and semantic coordination (corresponding to feature resolution and coordinate semantics) in a simple and intuitive way.
One of the implications of our analysis of Australian juxtaposed nominal constructions is that close appositions are structurally the same as coordinations in the syntax of these languages, which in turn raises the possibility that this might be true of close appositions and coordination in (some) other languages. A further implication of our general approach is that we may expect to find a structural relationship between non-asyndetic coordination and appositional constructions also. In other words, do we find non-asyndetic coordinated structures which can similarly be used with an appositional semantics? In fact, boolean coordination in languages like English may well fit into a continuum between coordination and close apposition in this way. The conjunction of singular nominals generally forms a plural noun phrase in English, under (some form of) group-forming coordination as discussed in section 3, but this is not always the case. As the sg verb agreement in example (88) indicates, ‘vice-president’ and ‘president-elect’ refer here to the same individual. This boolean or joint reading is in contrast to (89), which involves a split or group-forming reading.
(88) The vice-president and president-elect is eating pizza.
(King & Dalrymple Reference Dalrymple2004: 75)
(89) The vice-president and president-elect are eating pizza.
As King & Dalrymple (Reference Dalrymple2004) show (see also Heycock & Zamparelli Reference Heycock, Zamparelli and Dekker1999, Reference Heycock, Zamparelli, Hirotani, Coetzee, Hall and Kim2000) languages differ in the distribution of these readings. Languages such as English permit both joint and split readings (with singular nouns) under a single determiner such as the or this, while other languages permit only joint readings in these circumstances (such as Italian, Portuguese and German). The German example in (90), for instance, has only a joint reading.
(90) mein bester Freund und Mann my.m.sg best friend.m.sg and husband.m.sg ‘my best friend and husband’
(King & Dalrymple Reference King and Dalrymple2004: 93 German)
The distinction between split and joint readings is most evident with singular conjuncts, but is also relevant for the interpretation of coordinate structures with plural conjuncts, as in five philosophers and linguists, which has a number of interpretations including the joint reading under which each individual (of five) is both a linguist and a philosopher.
King & Dalrymple (Reference Dalrymple2004) propose that the distinction between split and joint readings involves two different semantics for and, boolean-and and group-and. Group-forming and (‘normal’ coordination) simply requires a plural index, but King & Dalrymple propose that boolean-and is associated with the syntactic index requirement stated in (91),Footnote 36 which is very similar in effect to the index constraint used in this paper for the cases of appositional juxtaposition.
(91) boolean and (↑ index num)=(↑∊index num)
It is sometimes suggested that a proper account of nominal coordination should involve just one unitary semantics, either in terms of group forming or of boolean coordination (for example, Heycock & Zamparelli (Reference Heycock, Zamparelli, Hirotani, Coetzee, Hall and Kim2000) propose one underlying semantics for and and associate different (semantic) procedures with various (abstract) syntactic features in English and Italian NPs). Similarly, the use of the conjunctive coordinator and to conjoin alternative descriptions of a single individual might seem idiosyncratic. However, our proposal provides support for the existence of both group-forming and boolean and in nominal coordination, by placing the existence of boolean coordination (that is, the joint reading of my friend and colleague) in a rather different crosslinguistic context. We have argued that appositions, inclusories and (standard) conjunctions may all be associated with precisely the same syntactic device, namely of asyndetic coordination or juxtaposition. On our view, then, boolean coordination can be considered to be essentially similar to the (close) appositional juxtapositions we have discussed in Australian languages, the only difference being the presence of an overt coordinator in the former (and note that in English, overt conjunctions, particularly or, may occur in appositions). Boolean coordination is thus syntactically coordinated (having a hybrid f-structure), but semantically appositional (having no feature resolution and appositional (i.e. boolean) semantics).
