Syntax of natural and accidental coordination:
Evidence from agreement
Mary Dalrymple ([email protected])Irina Nikolaeva (irina a [email protected])
Centre for Linguistics and PhilologyUniversity of Oxford
September 7, 2005
1
Syntax of natural and accidental coordination:Evidence from agreement
Abstract
Adjective agreement with coordinated nouns in Finnish presents a puzzling pat-tern: in some cases, a plural adjective is required with coordinated singularnouns, while in other cases that seem to be syntactically identical, a pluraladjective is disallowed. The key to this puzzle lies at the syntax-semantics in-terface: plural adjectives are required in cases of natural coordination, wherea salient close relation holds between the conjuncts, but disallowed in cases ofaccidental coordination. The semantic distinction between natural and ac-cidental coordination has a syntactic reflex in Finnish, where only the featuresof the natural coordination structure are compatible with the agreement re-quirements of plural adjectives. We show that the distinction between naturaland accidental coordination is syntactically reflected not only in Finnish, but inother languages as well, including Tundra Nenets, Russian, and Kurdish.
1 Introduction
Adjectives in Finnish show number agreement with the nouns they modify;singular nouns take singular adjectives, and plural nouns take plural adjectives:1
(1) iloinen poikahappy.SG boy‘the/a happy boy’
(2) iloiset pojathappy.PL boy.PL
‘the happy boys’
With conjoined plural nouns, plural adjectives are required. In these examples,the adjective can be interpreted as taking scope over both nouns, as indicatedby the bracketing in the translation.
(3) nuoria tyttoja ja poikiayoung.PART.PL girls.PART.PL & boys.PART.PL
‘young [girls and boys]’
1We use the following abbreviations:
ABL ablative ACC accusative ALL allativeDAT dative DEF definite article DU dualESS essive FEM feminine GEN genitiveILL illative INES inessive IZ izafeINST instrumental LOC locative MASC masculineNEG negation NEUT neuter PART partitivePL plural POSS possessive SG singularSUP superessive
2
(4) protestanttiset arvot ja hyveetProtestant.PL merit.PL & virtue.PL
‘Protestant [merits and virtues]’
The situation is more complicated with conjoined singular nouns. In someinstances, a plural adjective is allowed:
(5) Iloiset mies ja poika lahtivat yhdessa kasi kadessa.happy.PL man & boy left.3PL together hand hand.INES‘The happy [man and boy] left together hand in hand.’
In other instances, however, a plural adjective cannot be used:
(6) * Han osti uudet talon ja auton.he bought.3SG new.ACC.PL house.ACC & car.ACC
‘He bought a new [house and car].’
The purpose of this paper is to explain these patterns. We show that the possi-bility for a plural adjective correlates with a semantic distinction between natu-
ral and accidental coordination (Walchli 2005; Haspelmath 2004a, to appear).Plural adjectives are allowed with natural coordination (example 5), but notwith accidental coordination (example 6). The availability of plural adjectivemodification is explained by assuming that Finnish has a special syntactic struc-ture for natural coordination, whose agreement features are compatible with theagreement requirements of Finnish plural adjectives. In contrast, the acciden-tal coordination structure is not compatible with Finnish adjective agreement.Other languages show a similar contrast: we examine data from Tundra Nenets,Russian, and Kurdish, showing that they also have a special syntactic construc-tion for natural coordination structures, and that the agreement features fornatural and accidental coordination pattern exactly as in Finnish.
Our work shows that the semantic distinction between natural and acciden-tal coordination can be expressed by different syntactic constructions, but thatthese constructions do not necessarily involve different coordination strategiesor different morphosyntactic marking of conjuncts. In fact, the natural coor-dination structures may be superficially completely identical to the accidentalcoordination structure, differing only in the syntactic features that are associ-ated with them. The features of these special natural coordination structuresare much more like features of plural nouns or compounds than of phrase-likecoordination patterns.
2 Adjective agreement with coordinated nouns
In examining a range of Finnish examples in which an adjective modifies singularconjoined nouns, a large amount of variation is found, both across speakers andacross types of examples. Our Finnish data were collected by questionnaire anddiscussion with native speakers. Seven speakers were instructed to translateEnglish sentences into Finnish so that the modifying adjective takes scope over
3
both conjuncts. The questionnaire is provided in the Appendix. Additionalexamples were obtained from the Internet and from the Kielipankki corpus,provided by CSC, the Finnish IT center for science.
2.1 Symmetric modifiers
A certain semantic class of modifiers, which we call symmetric modifiers, re-quires plural and not singular agreement. Semantically, these modifiers specifyreference to two or more entities, and they express a symmetric relation whichholds between these entities. In this class are adjectives like similar, match-ing, and compatible as well as participial modifiers with reciprocal objects likeresembling each other.
Plural adjectives are required by all speakers in the following examples,regardless of the semantic characteristics of the coordinated nouns. A singularadjective is strictly ungrammatical for all speakers.
(7) On vaikeaa suunnitella yhteensopivat/*yhteensopiva yliopisto ja teatteri.is difficult design compatible.PL/compatible.SG university & theater‘It’s difficult to design a compatible [university and theatre].’
(8) En ole koskaan nahnyt toisiaanNEG.1SG have never seen each.other
muistuttavia/*muistuttavaa aitia ja tytarta.resembling.PART.PL/resembling.PART.SG mother.PART & daughter.PART
‘I have never seen a [mother and daughter] resembling each other.’
(9) En ole koskaan nahnyt samannakoisia/*samannakoistaNEG.1SG have never seen similar.looking.PART.PL/similar.looking.PART.SG
aitia ja tytarta.mother.PART & daughter.PART
‘I have never seen a similar-looking [mother and daughter].’
2.2 Closest conjunct agreement
With nonsymmetric modifiers, some but not all speakers allow a singular adjec-tive to take scope over conjoined nouns. The resulting structure is ambiguous:
(10) Sinulla on vanha talo ja auto.you.SUP is old.SG house & car‘You have an old [house and car].’ or‘You have an [old house] and a car.’
This possibility can be shown to be an instance of closest conjunct agreement(Sadler 1999, 2003), in which the adjective agrees not with the coordinate struc-ture as a whole, but with whichever conjunct is closest to the adjective. Closestconjunct agreement is not available for all speakers of Finnish. Those for whomit is possible allow the patterns of agreement shown in (11–12), in which the
4
conjuncts have different number; the noun tuolit ‘chairs’ in (11a) and (12a) isin the plural, and the noun kasvot ‘face’ in (11b) and (12b) is a pluralia tan-tum noun. For those speakers, the adjective shows agreement with the closestconjunct.
(11) a. Han osti vanhan / *vanhat poydan ja tuolit.he bought.3SG old.ACC.SG / *old.ACC.PL table.ACC.SG & chair.ACC.PL
‘He bought the old [table and chairs].’
b. Pidan hanen oudosta / *oudoistalike.1SG his strange.ABL.SG / *strange.ABL.PL
aanestaan ja kasvoistaan.voice.ABL.SG.3POSS & face.ABL.PL.3POSS
‘I like his strange [voice and face].’
(12) a. Han osti vanhat / *vanhan tuolit ja poydan.he bought.3SG old.ACC.PL / *old.ACC.SG chair.ACC.PL & table.ACC.SG
‘He bought the old [chairs and table].’
b. Pidan hanen oudoista / *oudostalike.1SG his strange.ABL.PL / *strange.ABL.SG
kasvoistaan ja aanestaan.face.ABL.PL.3POSS & voice.ABL.SG.3POSS
‘I like his strange [face and voice].’
In the examples in (11), the first conjunct is singular, and a singular and nota plural adjective is allowed. The opposite pattern is found in (12), in whichthe first conjuncts are plural: here, a plural and not a singular adjective mustbe used. For the speakers who do not allow closest conjunct agreement, theconflict can be avoided by repeating the adjective in each conjunct.
Only the speakers for whom closest conjunct agreement is possible have anoption to use a singular adjective when two singular nouns are conjoined. In thefollowing, except where noted, we discuss the judgments of speakers for whomclosest conjunct agreement is not available.
2.3 Nonsymmetric modifiers: definite nouns
For nonsymmetric modification, the possibility for a plural adjective with con-joined singular nouns correlates with semantic features such as definiteness andanimacy. Definiteness seems to play the most important role. Since Finnishhas no definite or indefinite articles, we determined whether a noun phrase wasdefinite or indefinite on the basis of the translations of the examples and the con-texts in which they were given. Word order in Finnish also plays an importantrole in interpretation: preverbal subjects are normally interpreted as definite,while indefinite subjects normally follow the verb. The subject is postverbal in
5
presentational and possessive constructions; in these constructions the verb ex-hibits default (third person singular) agreement, independent of the agreementfeatures of the postverbal noun phrase.2
For definite coordinated nouns, syntactic role is an additional factor inwhether or not a plural adjective is allowed. All speakers agree that with def-inite human subjects, a plural adjective can be used. For speakers who donot allow closest conjunct agreement, a singular adjective in examples (13-14)is ungrammatical.
(13) Iloiset/*iloinen mies ja poika lahtivat yhdessa kasi kadessa.happy.PL/happy.SG man & boy left.3PL together hand hand.INES‘The happy [man and boy] left together hand in hand.’
(14) Tyhmat/*tyhma Mikko ja Juha lahtivat tyopaikoistaan sairaalassa ja pankissa.stupid.PL/stupid.SG M. & J. left.3PL job.ABL.PL.3POSS hospital.INES & bank.INES‘Stupid [Mikko and Juha] left their jobs in a hospital and a bank.’
Plural adjectives can modify either conjoined singular nouns (with a restrictiveinterpretation, as in example 13) or names (with a nonrestrictive interpretation,as in example 14). We know of no differences in the availability of plural mod-ification correlating either with restrictive/nonrestrictive modification or withmodification of nouns as opposed to proper names.
We found many naturally-occurring examples on the Internet in which co-ordinated human definite subjects are modified by a plural adjective. Two suchexamples are given in (15):
(15) a. Loppuun iloiset isa ja poika palaa noutamastafinally cheerful.PL father & son returned fetching.ABL
vapaastilentaviaan lentokentan heinikossa aurinkosena paivana.freely.flying.ACC.PL.3POSS airport.GEN grass.INES sunny.ESS day.ESS
‘Finally on a sunny day the cheerful [father and son] returned fromfetching their freely-flying planes on the airport grass.’
b. Onnelliset isa ja aiti haluavat tarjotahappy.PL father & mother want.3PL to.offer
tyttarelleen vain kaikkein parasta.daughter.ALL.3POSS only everything.ACC best
‘The happy [father and mother] only want to offer their daughter the very best.’
Even for speakers who allow closest conjunct agreement, a plural adjective isstrongly preferred, though a singular adjective is also marginally possible.
On the other hand, not all speakers allow a plural adjective to modify non-human definite coordinated subjects. A plural adjective was acceptable withanimate non-human subjects for only three out of seven speakers, and withinanimate subjects for two of the seven speakers, as shown in (16-17):
2In some works (Toivonen 2002), the non-agreeing postverbal nominative noun phrase inpossessive constructions is not analyzed as a subject, though we will treat it as subject here.
6
(16) Sairaat koira ja hevonen jaavat tanneill.PL dog & horse stay.3PL here‘The ill [dog and horse] will stay here.’ (3 of 7 speakers)
(17) Vanhat poyta ja hylly nayttavat rumilta.old.PL table & shelf look.3PL ugly‘The old [table and shelf] look ugly.’ (2 of 7 speakers)
Definite inanimate subjects modified by a plural adjective occur twice inthe Kielipankki corpus. In both instances, the conjoined nouns refer to uniqueentities, and have something of the status of proper names.
(18) a. ... wienilaiset kuoro ja orkesteri hyokyvat kuin pitaa.... Viennese.PL choir & orchestra surge.3PL as needed‘... Viennese [choir and orchestra] surge as they should.’
b. Tupakkalain valvonnasta vastaa sosiaali- ja terveysministeriotobacco control.ABL is.responsible social - & health.ministry
ja sen alaiset tyosuojeluosasto ja tuotevalvontakeskus.& its subordinate.PL safety.department & product.control.center
‘The social and health ministry and its subordinate [safety department andproduct control center] are responsible for tobacco control.’
Finally, when coordinated definite nouns have a non-subject role, a pluraladjective was judged as marginally acceptable by one speaker if the conjunctsare animate and human.
(19) ? Annoin kirjan onnellisille miehelle ja pojalle.gave.1SG book.ACC happy.ALL.PL man.ALL & boy.ALL
‘I gave the book to the happy [man and boy].’ (1 of 7 speakers)
A plural adjective with inanimate non-subjects is impossible for all speakers.
(20) * En koskaan unohda Pariisin hienoja payvaa ja yota.NEG.1SG never forget Paris.GEN wonderful.PART.PL day.PART & night.PART
‘I will never forget the wonderful [day and night] in Paris.’
There are no instances of a plural adjective modifying definite inanimate non-subject nouns in the Kielipankki corpus.
Thus, human definite subjects require a plural adjective. Plural adjec-tives are acceptable with non-human definite subjects and human definite non-subjects for only some speakers. With non-subject inanimate nouns, a pluraladjective is disallowed.
7
2.4 Nonsymmetric modifiers: indefinite nouns
With indefinite nouns, plural modifiers are considerably less frequent, and forsome speakers they are totally excluded. Indefinite inanimates never occur witha plural adjective, whether they are subjects (21) or non-subjects (22). Thereare no instances of conjoined indefinite inanimate nouns modified by a pluraladjective in the Kielipankki corpus, and all speakers consistently rejected thefollowing examples:
(21) * Tassa kaupungissa on hyvat yliopisto ja teatteri.this.INES city.INES is good.PL university & theater
‘There is a good [university and theatre] in this city.’
(22) * Han osti uudet talon ja auton.he bought.3SG new.ACC.PL house.ACC & car.ACC
‘He bought a new [house and car].’
For all of these examples, a singular adjective is permissible for speakers whoallow closest conjunct agreement. Other than the possibility for closest conjunctagreement, conjoined inanimate indefinite singular nouns cannot be modified byeither a singular or a plural adjective. To convey the meaning intended by (21),the adjective must be repeated:
(23) Tassa kaupungissa on hyva yliopisto ja hyva teatteri.this.INES city.INES is good.SG university & good.SG theater‘There is a good university and a good theatre in this city.’
Animate indefinites do not normally allow a plural adjective either. Thefollowing examples were judged ungrammatical by all speakers we consulted,whether the conjoined nouns are subjects (24) or nonsubjects (25):
(24) a. * Olipa kerran koyhat mies ja nainen.was once poor.PL man & woman
‘Once there was a poor [man and woman].’
b. * Kylaan saapui hyvat opettaja ja laakari.village.ILL arrived good.PL teacher & doctor
‘A good [teacher and doctor] arrived in the village.’
(25) a. * Haluan ostaa vanhat kissan ja koiran.want.1SG to.buy old.ACC.PL cat.ACC & dog.ACC
‘I want to buy an old [cat and dog].’
b. * Nain oudot miehen ja pojan.saw.1SG strange.ACC.PL man.ACC & boy.ACC
‘I saw a strange [man and boy].’
However, a plural adjective modifying indefinite human nouns was acceptablefor one speaker out of seven.
8
(26) Tapasin vanhempainillassa huolestuneet isan ja aidin.met.1SG parent.meeting.INES worried.ACC.PL father.ACC & mother.ACC‘I met a worried [father and mother] at a parents’ meeting.’(1 of 7 speakers)
In sum, a plural adjective is very unusual with indefinite nouns, being onlymarginally acceptable even when the conjoined nouns are animate and human.
2.5 Summary of the data
In summary, all speakers require plural adjectives with symmetric modifiers likesimilar, and prefer plural adjectives with entities that are definite, animate andhuman. Indefinites generally disallow plural adjectives, although human indefi-nites are acceptable for some speakers. For definite nouns, plural modificationis more likely for subjects, and is in fact required if the definite subjects are alsohuman. Non-subject definites pattern with indefinites.
In the next section, we propose a unifying factor for the examples allowingplural adjective modification, i.e. examples with symmetric modifiers and withhighly definite/animate/human conjuncts. These two classes of examples ex-hibit a particular subtype of what has been called natural coordination, inwhich there is a close and salient relation between the entities denoted by theconjuncts.
3 Natural and accidental coordination
Natural coordination is a semantic relation in which two entities are closelyrelated in meaning and form a conceptual unit (Haspelmath 2004a, to appear;Walchli 2005). Walchli (2005) states that natural coordination involves entitiesthat often go together, and are expected to cooccur:
(27) “Natural coordination ... implies, among other things, that the parts ex-press semantically closely associated concepts, such as ‘brother and sister’,‘hands and feet’, ‘eat and drink’, ‘knife and fork’, etc., which are on thesame hierarchical level, and that the whole meaning (‘siblings’, ‘limbs’,etc.) is more general than the meaning of the parts.” (Walchli 2005, p. 1)
In contrast, accidental coordination involves “coordination of items whichare not expected to co-occur, and which do not have a close semantic rela-tionship” (Walchli 2005, p. 5). Obviously, as Walchli (2005, p. 8) notices,natural coordination is highly dependent on culture-specific factors. Entitiesthat are conceptually close in one culture may not be so in another, and there-fore languages exhibit a great deal of variation in what they treat as naturalcoordination.
It is important to note that the term “natural coordination” refers to asemantic rather than a syntactic relation. This relation can be syntacticallyexpressed in a number of ways. Walchli’s study focuses primarily on natural
9
coordination as it is syntactically realized in co-compounds, word-like units con-sisting of two parts such as the Georgian (Kartvelian) da’-dzma ‘sister-brother(= siblings)’ in contrast to accidental coordination with the conjunction da ‘and’(Walchli 2005, p. 46), or Tocharian A (Indo-European) nom-klyu ‘name-glory’.Natural coordination can also be expressed by coordinate structures: coordi-nated nouns, coordinated noun phrases, or coordinated verbs. Here we areconcerned with natural coordination as syntactically expressed in nominal co-ordinate structures, and in particular with the syntax of naturally coordinatednouns with plural modification.
In some languages, the semantic relation of natural coordination is expressedsyntactically by a special coordination strategy. In languages which have aspecial structure for natural coordination, there is generally an asymmetry inthese strategies: accidental coordination cannot be expressed with the naturalcoordination strategy, while natural coordination can be expressed by usingeither the specific structure reserved for it, or the more general structure whichmust be used for accidental coordination. The choice of structure for naturalcoordination depends on how the speaker construes the relation between theconjuncts and what aspects of the relation the speaker wants to emphasize. If aspeaker chooses to emphasize the conceptual closeness between two conjuncts,the natural coordination strategy may be used; otherwise, the speaker can usethe regular accidental coordination pattern.
3.1 Syntactic characteristics of natural coordination
Some languages distinguish natural and accidental coordination by different co-ordination markers. Udihe (Manchu-Tungus) is one such language (Nikolaevaand Tolskaya 2001). It is a ‘with’ language in the sense of Stassen (2000): nom-inal coordination is expressed by an adpositional phrase with the postposition‘with’, so that the conjuncts have different syntactic status. In Udihe, the post-position zuNe is used for accidental coordination, while mule is used for someinstances of natural coordination, normally for pairs of relatives:
(28) Accidental coordination:
bi Sergej zuNe / *mule
I Sergej with / with‘I and Sergej’
(29) Natural cordination:
bi mamasa mule / *zuNe
I wife with / with‘I and my wife’
Natural coordination may also involve phonological reduction of the coor-dination markers, as observed in some Oceanic languages (Moyse-Faurie andLynch 2004). For example, natural coordination is expressed in Lenakel by theconjunction m, a phonologically reduced form of the form m@ne which is used
10
for accidental coordination. This exemplifies what Walchli calls “tight coordi-nation”: a coordination strategy involving syntactic reduction.
Tight coordination may also be characterized by special morphological orsyntactic characteristics of the conjuncts: in particular, single marking of suchfeatures as definiteness, case, number, possession and the like. In German,some instances of natural coordination lack determiners, while this is generallyimpossible with accidental coordination. Lambrecht (1984) refers to examplessuch as (31) as bare binomials:
(30) Accidental coordination:
der Mond und ein Sechserthe moon & a sixpence‘the moon and sixpence’
(31) Natural coordination:
Sonne und Mondsun & moon
‘the sun and the moon’
In Eastern Armenian (Indo-European), a single inflectional marking is usedfor some instances of natural coordination, while this is not possible for acci-dental coordination (Minassian 1980, p. 164, cited in Walchli 2005, p. 49):
(32) Accidental coordination:
dproc’-i-s ev usuc’ic’-ner-i-sschool-DAT-1SG & teacher-PL-DAT-1SG‘for my school and my teachers’
(33) Natural coordination:
ał u hac’-dsalt & bread-2SG‘your salt and bread (= hospitality)’
Walchli (2005) provides more examples of different types of single marking usedin tight coordination structures.
Although tight coordination tends to be used to express the semantic relationof natural coordination, it can also be used to express accidental coordination,particularly in languages which have no overt coordinator and therefore have nocoordination strategy other than tight coordination. However, we have foundno language which uses reduced morphological or syntactic marking for acci-dental coordination as opposed to natural coordination. Haspelmath (2004b,p. 13) proposes a functional explanation for this which is essentially iconic: ifthe relation between conjuncts is predictable, as it is in many cases of naturalcoordination, the marking of the relation between the conjuncts can be reducedor absent. In other words, morphosyntactic closeness between conjuncts mirrorsconceptual closeness between relevant entities.
11
Not all languages make use of this iconic strategy of tight coordination. Ourwork shows that natural coordination can be expressed not only by compound-ing or tight coordination, but also by the use of the more general strategy forcoordination. In fact, in Finnish and some other languages natural coordinationis not associated with any of the special morphosyntactic characteristics thatWalchli discusses. Although natural coordination structures do not involve anyspecial marking strategy, they differ from accidental coordination structure intheir syntactic features, as shown by patterns of plural modification.
3.2 Contextual factors
Often, the relation between the elements of a natural coordination constructionis intrinsic, as in examples like ‘father and mother’ or ‘cup and saucer’ as wellas the examples discussed in the previous subsection. Intrinsic natural coordi-nation may involve not only reduced syntax but also frozen word order (Benorand Levy 2006; see Allan 1987 for further discussion of word order constraints incoordination). However, the natural coordination relation can also be contex-
tually specified – that is, dependent on particular information available fromthe preceding discourse, the utterance situation, or general knowledge of theworld. In fact, inherent natural coordination pairs such as ‘father and mother’are probably best treated as highly conventionalized subtypes of contextuallyestablished natural coordination.
Erz’a-Mordvin (Uralic) shows that co-compounds expressing natural coordi-nation can involve either intrinsic or contextually established natural coordina-tion. Erz’a-Mordvin co-compounds usually consist of two parts, each of whichbears a plural marker: at’a-t baba-t ‘old man.PL and old woman.PL’ (Walchli2005, p. 6). In this example, the conjuncts inherently form a natural pair inde-pendently of the context. However, the same strategy is also used for contex-tually established natural coordination, e.g. Igor-t Natasa-t ‘Igor and Natasha’(Walchli 2005, p. 12). In this example, the individuals Igor and Natasha areperceived as closely related partners. If they did not belong together, the acci-dental coordination strategy with the conjunction di ‘and’ would be used: Igordi Natasa ‘Igor and Natasha’.
Contextually established natural coordination can also be expressed by syn-tactically coordinate structures. Lambrecht (1984, p. 794) shows that for someGerman bare binomials, “a meaningful relation between items is establishedonly by virtue of the context in which they appear”. Heycock and Zamparelli(2003) discuss coordinated bare definites in English and other languages; suchexamples are syntactically reduced, appearing without a determiner, but areproductive and nonidiomatic:
(34) We had to set the table for the queen. We arranged one crystal goblet,one silver spoon, and two antique gold forks. [Goblet and spoon] were seton the right of the plate. (Heycock and Zamparelli 2003, p. 445)
Heycock and Zamparelli show that the bare definites in such examples are either
12
explicitly mentioned in the previous discourse or contextually evoked.3
Walchli (2005, p. 12) presents the Swiss German examples given in (35–36).In some dialects of Swiss German, proper nouns are compatible with determin-ers; as in the Standard German examples in (30–31) above, determiners are usedin accidental coordination, but not in natural coordination. In example (35), theaccidental coordination strategy is used, and it is not presupposed that Simonand Valeri are partners. In contrast, in example (36), the natural coordinationstrategy is used because (or to make it clear that) they are partners.
(35) Hut si dr Simon u d Valeri z Bsuech choo.today is DEF S. & DEF V. at visit come‘Today Simon and Valeri came for a visit [they are not necessarily part-ners].’
(36) Hut si Simon u Valeri z Bsuech choo.today is S. & V. at visit come‘Today Simon and Valeri came for a visit [they must be partners].’
As Walchli shows, the connection between Simon and Valeri is established fromthe situational context. There is no intrinsic or conventional relation that canexplain the use of natural coordination in example (36).
In examples like these, the use of a natural coordination structure dependson subtle pragmatic factors that determine the speaker’s strategy for encodingthe relation between the conjuncts, and we expect to find significant varia-tion in whether and how natural coordination structures are used to expresscontextually-established relations between individuals.
3.3 Semantic types of coordination
We have seen that there are two subtypes of natural coordination: one in whichthe relation between the conjuncts is conventional or intrinsic within a givenculture, and the other in which it is not intrinsic. In the second type, the relationbetween the conjuncts is established from textual or situational context. Thespeaker has a choice whether to use the available natural coordination strategyfor expressing the contextually established relation between the conjuncts. Theuse of this strategy emphasizes that this relation is particularly close.
We propose that this second type of natural coordination subtype is relevantin the analysis of Finnish examples of conjoined nouns modified by plural adjec-tives. In these examples, a close relation between the conjuncts is establishedcontextually. The closeness of this relation is signaled by the use of a partic-ular syntactic construction compatible with modification by a plural adjective,as distinct from the more general structure for coordination which disallows aplural adjective. Only the use of the plural adjective disambiguates between
3Crucial to Heycock and Zamparelli’s analysis is their claim that coordinated bare definiteslike (34) cannot be modified: *valuable [goblet and spoon]. In this way, these examples aresyntactically different from the Finnish examples presented in Section 2.
13
natural and accidental coordination readings, since both types use the samecoordination strategy with the conjunction ja ‘and’.
There are two ways in which this contextually relevant, close semantic re-lation between the conjuncts can be established in Finnish. First, the relationcan be explicitly asserted by the use of a symmetric adjective such as compat-ible or similar, with the conjuncts of the natural conjunction structure as thearguments of the relation.
Second, the natural coordination strategy can be used to establish or empha-size a close relation between the conjuncts, whose naure is not explicitly assertedby the coordinate structure, but is contextually salient. This further impliesthat each of the conjuncts is salient as well: a pragmatically salient relationshipcan only hold between pragmatically salient entities. We therefore expect theconjuncts in the contextually established natural coordination structure to havemany properties of discourse-salient entities. In particular, the conjuncts willshare many properties with topics, which are salient by definition (Lambrecht1994). This is exactly what we observed in Section 2: conjuncts that requireor allow modification by a plural adjective are high in features of what Comrie(2003) calls ‘topic-worthiness’, features such as definiteness, animacy and hu-manness which tend to characterize sentence topics. Conjuncts that are moretopic-worthy are more likely to participate in a contextually salient relation (andmay in fact be topics), to be semantically related by natural coordination, andtherefore to be eligible for modification by a plural adjective.
Such conjuncts also tend to occur in subject position. Given the well knowncorrelation between subjects and topics (Givon 1976, Comrie 1989, and otherworks), this additionally indicates that naturally coordinated conjuncts tend tohave a topical interpretation.
On this view, coordinate structures can involve:
(37) • Accidental coordination, or
• Natural coordination, which encompasses:
Type 1: an intrinsic or conventionalized close relation between con-juncts, or
Type 2: a contextually established, salient, close relation betweenconjuncts, in which
– the relation is explicitly asserted in the same utterance, or
– the relation holds between salient entities in the discourse.
Languages differ in whether and how they encode the semantic distinctionsbetween different types of coordination. Some languages only have one coordi-nate structure used for all types of coordination. In other languages, intrinsicnatural coordination is encoded differently from all other types (Udihe). Finnishpossesses a special syntactic construction to express contextually established
natural coordination. The examples in Section 2 show that the conjuncts co-ordinated by this special construction do not have to be intrinsically related,although they happen to be intrinsically related in some instances.
14
The next section presents an analysis of the syntax of natural coordina-tion structures in Finnish, showing how they differ from accidental coordina-tion structures syntactically. This analysis, together with a particular set ofassumptions about adjective agreement, accounts for the Finnish patterns pre-sented above.
4 Coordination in LFG
In the following, we first provide a brief overview of the basic syntactic assump-tions of our analysis, which is cast in the theory of Lexical Functional Grammar(LFG). In this section we discuss the standard treatment of the syntax of coor-dination and the interaction of agreement features and coordination. We thenpresent our proposal for the syntax of natural coordination in Finnish, and showthat it accounts for the fact that plural adjectives are acceptable with naturalcoordination structures but not with accidental coordination structures. In thenext section, we show that the same patterns are found in a number of otherlanguages as well.
4.1 INDEX and CONCORD features
As in traditional grammatical treatments as well as more recent work withinHead-Driven Phrase Structure Grammar and LFG (Pollard and Sag 1994, Kathol1999, Wechsler and Zlatic 2000, 2003, King and Dalrymple 2004), we assumethat nouns bear two kinds of syntactic agreement features, index and concord.Different agreement processes can refer to either or both of these sets of fea-tures. Wechsler and Zlatic (2003) show that both sets of features are syntacticand cannot be reduced to morphological or semantic features.
index features are closely related to, though not identical to, semantic fea-tures; the plural index features of a phrase like these boys reflect the fact thatthe phrase refers to more than one individual. concord features, in contrast, aremore closely related to morphological properties of phrases. Following Pollardand Sag (1994), Wechsler and Zlatic (2003) show that verb agreement in Englishand many other languages refers to index features and not concord features.Wechsler and Zlatic (2003, p. 51) provide the following example of mismatchbetween concord and index for the Serbian/Croatian noun deca ‘children’:
(38) ta dobra deca dolaze.that.FEM.SG good.FEM.SG children came.3PL‘Those good children came.’
The noun deca has feminine singular concord features, as shown here by thefeminine singular determiner and adjective. Semantically, it refers to more thanone child, and has plural index features, as shown by plural agreement on theverb. For more discussion of concord and index and their relation to mor-phological declension class and semantic interpretation, see Wechsler and Zlatic(2003).
15
Our analysis of Finnish adjective agreement follows the work of King andDalrymple (2004), who show that conjoined singular nouns have singular concord
but plural index. Thus, English conjoined singular nouns require a singular de-terminer (this/*these boy and girl), since English singular determiners specifysingular concord agreement. However, noun phrases with conjoined nounslike this boy and girl require plural verb agreement (this boy and girl are/*islaughing), since English plural verbs require plural index agreement. In the fol-lowing, we propose that Finnish singular adjectives require both singular index
and singular concord, making them incompatible with accidentally coordi-nated singular nouns, whose index is plural. In contrast, the syntactic structurefor naturally coordinated singular nouns has plural index and concord, whichis compatible with the requirements of Finnish plural adjectives.
4.2 Syntactic assumptions
For an overview and introduction to LFG theory, see Bresnan (2001), Dalrymple(2001), and Falk (2001). Our assumptions about the syntax of coordinate struc-tures follow Dalrymple and Kaplan (2000), and we adopt the analysis of Kingand Dalrymple (2004), reviewed below, for modifier agreement with conjoinednouns.
LFG assumes two syntactic structures: the constituent structure or c-structure
is a phrase structure tree encoding phrasal structure and linear precedence, andthe functional structure or f-structure is an attribute-value structure encod-ing syntactic argument structure, grammatical functions, and features such astense, person, number, gender, and case. Ignoring agreement features for themoment, the c-structure and f-structure for the sentence David yawned are:
(39)IP
NP
N
David
I′
VP
V
yawned
pred ‘yawn〈subj〉’
tense past
subj[
pred ‘David’]
Nodes of the c-structure tree are related to f-structures as indicated by thearrows in (39). The NP and N nodes are related to the f-structure for thesubject, and the IP, I′, VP, and V nodes are related to the f-structure for thesentence as a whole.
Since we are mainly concerned with agreement features and the functionalstructure of noun phrases, and in particular with noun phrases headed by con-joined nouns, we usually omit the c-structures for the examples under discussion.We also follow standard LFG conventions in omitting f-structure features suchas person, gender, and case if they are not relevant to the discussion at hand.We assume the following functional structure for the noncoordinate noun phrase
16
this boy:
(40)
spec ‘this’
pred ‘boy’
concord [ num sg ]
index [ num sg ]
This functional structure is constructed on the basis of the lexical entries forthis and boy, given in (41):
(41) this (↑ spec)= ‘this’
(↑ concord num) = sg
boy (↑ pred)= ‘boy’
(↑ concord num) = sg
(↑ index num) = sg
According to these lexical entries, the noun boy has singular concord and index:the value of the num attribute within the concord and index f-structures is sg
(for singular). The English determiner this specifies singular concord, but doesnot constrain the index of its noun phrase. The concord specifications on thedeterminer and the noun are compatible, since both specify singular concord.The sg specification for index is not contradicted by any specifications on thedeterminer, since the determiner does not place any constraints on index. Sincethe agreement features for the determiner and noun are compatible, the phraseis correctly modelled as grammatical.
4.3 Coordinate structures and feature distributivity
Coordinate structures are represented as sets in LFG (Kaplan and Maxwell1988; Dalrymple 2001, chapter 9), reflecting the fact that they may contain anunbounded number of conjuncts as members of the set. A simplified functionalstructure for the coordinate nouns boy and girl is given in (42).
(42)
bg
b :[
pred ‘boy’]
g :[
pred ‘girl’]
Here, the conjuncts labeled b (for boy) and g (for girl) are members of the setlabeled bg, which represents the coordinate phrase.
We have seen that determiners can require the presence of certain agreementfeatures: for example, the lexical entry for this in (41) requires the noun itcombines with to have singular concord. What does it mean when a coordinatestructure — a set of f-structures as in (42) — is required to bear a certainfeature?
It is clear that sets of f-structures – that is, coordinate structures – can havefeatures of their own, different from the features of the conjuncts (Dalrymple and
17
Kaplan 2000, Peterson 2004). For example, a coordinate noun phrase subjectbehaves as a plural phrase, requiring plural verb agreement, even when eachconjunct is singular:
(43) Joe and Fred are/*is sleeping.
In order to address the formal question of how a set of f-structures can have itsown features, we must introduce a distinction between nondistributive anddistributive features (Dalrymple and Kaplan 2000):
(44) a. Nondistributive features are features of the set as a whole.
b. Distributive features are features of each member of the set.
Restating the distinction in more explicit terms:
(45) For any distributive property P and set s, P (s) iff ∀f ∈ s.P (f).For any nondistributive property P and set s, P (s) iff P holds of s itself.(Dalrymple and Kaplan 2000)
This distinction is not made by Peterson (2004), who treats all features (otherthan grammatical functions) as nondistributive. However, the distinction iscrucial, as shown by Dalrymple and Kaplan (2000) and King and Dalrymple(2004); we follow King and Dalrymple (2004) in classifying concord and index
differently in distributivity status:
(46) a. index is a nondistributive feature.
b. concord is a distributive feature.
This means that coordinated nouns or noun phrases have index features thatmay differ from the index features of the conjuncts, but that the concord
feature is determined by examining the concord features of the conjuncts. Aschematic f-structure for conjoined singular nouns is:
(47)
c :
index [ num pl ] (nondistributive feature of c)
[
concord [ num sg ]
index [ num sg ]
]
[
concord [ num sg ]
index [ num sg ]
]
(members of c)
Since coordinate structures lack a nominal head, it is expected that theylack concord features of their own, as concord features are morphosyntacticproperties of lexical heads. As King and Dalrymple (2004) show, the concord
features of coordinated nouns are inherited from the concord features of thenoun conjuncts. According to the definition of distributive features, then, the
18
value of a distributive feature of a coordinate phrase depends on the value ofthat feature in each conjunct:
(48)
index [ num pl ]
concord[
num sg]
index [ num sg ]
concord[
num sg]
index [ num sg ]
In (48), each conjunct is singular, and so the coordinate phrase as a whole in-herits singular concord; the phrase this boy and girl is acceptable, since boyand girl inherits singular concord and satisfies the agreement requirements ofthe determiner. If all noun conjuncts are plural, the coordinate phrase inher-its plural concord: these boys and girls is acceptable because each conjuncthas plural concord, as the plural determiner these requires. If some conjunctsare singular and some are plural, neither a singular nor a plural determiner isacceptable, since the coordinate structure cannot inherit incompatible require-ments from the conjuncts and hence has no concord specification: *this/theseboy and girls.
(49) concord is distributive:
• concord agreement with a coordinate phrase depends on the con-
cord features of each conjunct.
• A coordinate structure with conjoined singular nouns has singularconcord.
In contrast to the behaviour of the concord feature, the index feature isnondistributive. The index feature of a coordinate phrase does not depend onthe index of the conjuncts, but is determined independently for the coordinatephrase:
(50)
index[
num pl]
[
concord [ num sg ]
index [ num sg ]
]
[
concord [ num sg ]
index [ num sg ]
]
19
We assume that it is a general fact about all coordinated phrases with a structurelike (50) that they have plural index, contributed either by the phrase structurerule for coordination, or by the conjunction.4
(51) index is nondistributive:
• index agreement with a coordinate phrase depends on the index fea-tures of the coordinate phrase as a whole.
• A coordinate structure with conjoined singular nouns has plural index.
4.4 Adjective agreement with INDEX and CONCORD
Our proposals for index and concord agreement in adjectival modificationclosely follow the analysis of determiner agreement presented by King and Dal-rymple (2004).
4.4.1 CONCORD agreement: Hindi
In Hindi, singular determiners and adjectives require singular concord, similarto English singular determiners. A masculine singular adjective is used withconjoined masculine singular nouns, and a feminine singular adjective with con-joined feminine singular nouns:
(52) acchaa / *acche laDkaa aur sabziwaalaagood.MASC.SG / good.MASC.PL boy.MASC & vegetable.seller.MASC‘good [boy and vegetable seller]’
(53) wah kaalii / *kaalıı gaay aur billithat black.FEM.SG / black.FEM.PL cow.FEM & cat.FEM‘that black [cow and cat]’
It is not possible to use a plural adjective with conjoined singular nouns, evenif the adjective is a symmetric adjective, as in (55):
(54) kaalaa / *kaale mozaa aur jutaablack.MASC.SG / black.MASC.PL sock.MASC & shoe.MASC‘black [sock and shoe]’
(55) ek jaisaa / *ek jaise thailaa aur jutaasimilar.MASC.SG / similar.MASC.PL bag.MASC & shoe.MASC‘similar [bag and shoe]’
In order to ensure the proper agreement between the adjective and the noun,the singular adjective acchaa is required to combine only with nouns whoseconcord is singular. For conjoined nouns, because concord is a distributive
4More research is needed as to how syntactic and semantic constraints interact to determinethe index num in languages with a richer number system, where certain coordinate structuresare dual, others trial, and still others plural.
20
feature, this requirement entails that each conjunct must have singular concord.The f-structure for example (52) shows that this requirement is satisfied:
(56) acchaa laDkaa aur sabziwaalaagood.MASC.SG boy.MASC & vegetable.seller.MASC‘good [boy and vegetable seller]’
index [ num pl ]
adj{[
pred ‘good’]}
pred ‘boy’
concord[
num sg]
index [ num sg ]
pred ‘vegetable.seller’
concord[
num sg]
index [ num sg ]
Thus, for modifiers which require concord agreement, singular adjectives arerequired and plural adjectives are ruled out for conjoined singular nouns.
4.4.2 INDEX agreement: Hebrew
Hebrew displays a very different pattern of adjective agreement with conjoinednouns. In Hebrew, plural determiners and adjectives are used with conjoinedsingular nouns:
(57) ha-yeled ve ha-yalda ha-ktanim ha-?elethe-boy.SG & the-girl.SG the-small.PL the-these.PL
‘this small [boy and girl]’
Following King and Dalrymple (2004), we propose that Hebrew determinersand adjectives show index and not concord agreement. Recall furthermorethat index is a nondistributive feature, and that the functional structure ofconjoined singular nouns has plural index:
21
(58)
adj[
pred ‘small’]
index[
num pl]
pred ‘boy’
concord [ num sg ]
index [ num sg ]
pred ‘girl’
concord [ num sg ]
index [ num sg ]
Since Hebrew adjectives do not show concord agreement, the singular concord
of the conjuncts does not affect the ability of the plural adjective to appear withconjoined singular nouns. Singular adjectives are ruled out, since the index ofconjoined singular nouns is plural, not singular.
4.5 Natural and accidental coordination in Finnish
We propose that Finnish adjectives require both index and concord agree-ment.5 This means that modification of conjoined singular nouns by a pluraladjective is impossible if index and concord do not match. This is the case ofaccidental coordination.
The functional structure for the accidental coordination structure talo jaauto ‘house and car’ is shown in (59):
5This contradicts King and Dalrymple’s (2004) proposal for determiner agreement inFinnish, which was made on the basis of a smaller amount of data than we examine here.Our data shows that Finnish determiner agreement must be reclassified within King andDalrymple’s system; in their terms, Finnish determiners pattern with German and BrazilianPortuguese determiners in requiring both index and concord agreement, rather than withHindi determiners in requiring only concord agreement.
22
(59) Accidental coordination:
talo ja autohouse & car
‘house and car’
index [ num pl ]
pred ‘house’
concord [ num sg ]
index [ num sg ]
pred ‘car’
concord [ num sg ]
index [ num sg ]
Recall that examples like (60), exemplifying accidental coordination, are un-grammatical with either a singular or plural adjective for speakers who do notallow closest conjunct agreement:
(60) *vanhat / *vanha talo ja autoold.PL / old.SG house & car‘old [house and car]’
Crucially, accidentally coordinated singular nouns have plural index becausethe coordinate phrase as a whole is plural, and singular concord because eachconjunct is singular:
(61) Accidental coordination:
NP
N′
AdjP
Adj
*vanhat/*vanha
N′
N
N
talo
Conj
ja
N
auto
adj{[
‘pred ‘old’]}
index[
num pl]
pred ‘house’
concord[
num sg]
index [ num sg ]
pred ‘car’
concord[
num sg]
index [ num sg ]
23
Finnish adjectives constrain both the index and concord values of the nounsthey combine with: singular vanha requires singular index and concord, whileplural vanhat requires plural index and concord. Neither a singular nor a pluraladjective is allowed for accidental coordination of singular nouns:
• A plural adjective is ruled out because talo ja auto ‘house and car’ hassingular concord (each conjunct has singular concord).
• A singular adjective is ruled out (for speakers who disallow closest con-junct agreement) because ‘house and car’ has plural index (the coordinatestructure as a whole has plural index).
Thus, the agreement requirements for adjectival modification fail to be met forconjoined singular nouns, and no modification is allowed.
As we have seen, however, plural adjectives are in fact allowed for naturalcoordination. This is explained if we assume that Finnish specifies a differentfunctional structure for natural coordination, one which in some respects re-sembles the functional structure that would be associated with a simple pluralnoun:
(62) Natural coordination:
pred ‘conj’
concord [ num pl ]
index [ num pl ]
conj1
pred ‘husband’
concord [ num sg ]
index [ num sg ]
conj2
pred ‘wife’
concord [ num sg ]
index [ num sg ]
The f-structure for natural coordination is different in several ways from theaccidental coordination functional structure. First, the natural coordinationstructure has a bounded number of conjuncts, most often two (for more discus-sion, see Moyse-Faurie and Lynch 2004, p. 450). As we show below, one strategyfor natural coordination in Tundra Nenets is also limited to two conjuncts. Incontrast, the accidental coordination structure has an unbounded number ofconjuncts, for which a representation as a set (with an unbounded number ofelements) is more appropriate. Second, the natural coordination structure hasplural concord as well as plural index, in this respect resembling the f-structurefor a compound or a simple plural noun. As shown below, this pattern holdsfor natural coordination in a number of other languages as well.
A plural adjective is allowed for natural coordination because in this struc-ture, both concord and index are plural. This matches the agreement require-
24
ments of the Finnish plural adjective, and the example is allowed:
(63) Natural coordination:
uskolliset aviomies ja vaimofaithful.PL husband & wife‘the faithful [husband and wife]’
NP
N′
AdjP
Adj
uskolliset
N′
N
N
aviomes
Conj
ja
N
vaimo
pred ‘conj’
concord[
num pl]
index[
num pl]
conj1
pred ‘husband’
concord [ num sg ]
index [ num sg ]
conj2
pred ‘wife’
concord [ num sg ]
index [ num sg ]
adj{[
pred ‘faithful’]}
As shown in (61) and (63), the c-structure for the natural and accidental coordi-nation examples is the same; we know of no morphological or phrasal differencesbetween natural and accidental coordination structures in Finnish. The differ-ence is at functional structure, where constraints on adjective agreement arerelevant. This explains the distribution of adjectives with coordinated singularnouns in Finnish and, as we will show, in other languages as well.
We know of no other current theory of coordination and agreement thatwould naturally give rise to these predictions. The architecture of LFG, withits separation between c-structure and f-structure, facilitates our analysis ofFinnish, according to which accidental and natural coordination structures donot differ at c-structure – there is no morphological or phrasal distinctionbetween the two types of coordination – but are syntactically very differentat f-structure. Our analysis also crucially relies on the LFG distinction be-tween distributive and nondistributive features to produce a mismatch in index
and concord features for accidentally coordinated singular nouns. Classifyingconcord as a distributive feature means that the concord features of a coor-dinate phrase are inherited from the concord features of the conjuncts. Clas-sifying index as nondistributive means that the index features of a coordinatephrase may differ from the index features of the conjuncts, as required.
Of course, any theory of syntax can be reworked or augmented to take careof these facts. However, there is a clear sense in which the modifications neededwould go against fairly fundamental assumptions of at least some other theories.
25
For example, a basic assumption of Principles and Parameters-based analysesof coordination (Moltmann 1992, Johannessen 1998) is that all syntactic infor-mation is encoded in terms of phrase structure trees. This makes it difficultto reconcile the morphological and phrasal similarity of natural and accidentalcoordination structures in Finnish with the differences in adjective agreementthat we have treated at functional structure. It is also not clear how the dif-ference between nondistributive index and distributive concord features canbe modeled in this framework, since feature resolution in coordination has notbeen a primary focus of work within the theory.
The idea that nominal coordinations can lack the concord feature speci-fied in lexical nouns, but that this feature can be inherited from the concord
features of the conjuncts in certain circumstances, seems to go against HPSG’srequirement for total well-typing of syntactic structures (Carpenter 1992), whichstates that syntactic structures must be specified for all features appropriate fortheir type. According to our analysis, coordinations with singular noun con-juncts have singular concord, coordinations with plural noun conjuncts haveplural concord, and coordinations in which some nouns are singular and someare plural do not have concord features (see King and Dalrymple 2004 for morediscussion). Sag (2000) hints that the requirement for total well-typing mightbe relaxed in coordinate structures, and this might allow for concord to bespecified in some coordinate structures but not in others, but this possibilityremains to be explored in detail.
5 Natural coordination in other languages
The patterns of agreement that we have observed for Finnish are found in otherlanguages as well, and we believe that our findings generalize to other languageswhich express natural coordination with a special syntactic construction. As wehave seen, the accidental coordination structure exhibits a mismatch betweenconcord and index with coordinated singular nouns, since the singular concord
value is determined by the individual conjuncts, while the plural index value isdetermined by the coordinate structure as a whole. The natural coordinationstructure does not exhibit this mismatch: natural coordination structures withsingular conjuncts have plural concord as well as plural index.
5.1 Tundra Nenets
Tundra Nenets (Uralic) displays two types of coordinate structure: acciden-tal coordination involves either juxtaposition or an explicit conjunction, whilenatural coordination is expressed with the ‘double dual’ construction. Example(64) exemplifies the accidental coordination construction, with either an explicitconjunction nyabyi or juxtaposition of the conjuncts:
(64) lyekaro (nyabyi) toxolkoda too-xøh.doctor & teacher arrived-3DU
‘A doctor and teacher arrived.’
26
In the double dual construction, which expresses natural coordination, eachconjunct is marked as dual:
(65) lyekarø-xoh toxolkoda-xøh too-xøh.doctor-DU teacher-DU arrived-3DU
‘A doctor and teacher arrived.’
Despite the dual marking on each conjunct, example (65) means that two peo-ple arrived, one doctor and one teacher, as shown by the dual agreement inthe verb.6 The double dual construction is fully productive; as example (65)indicates, the relation between the conjuncts can be contextually established,and need not be inherent or conventional. (64) does not presuppose that thedoctor and teacher arrived together, although it can be interpreted in this way,while in (65) they are most likely to have arrived together or are otherwiseclosely associated. Unlike the accidental coordination structure, which allowsan unbounded number of conjuncts, the double dual construction requires twoconjuncts.
As in Finnish, the natural coordination structure is affected by saliency ofthe conjuncts and features of topic-worthiness: animacy, definiteness, appearingin subject position, and so on. In (66), the double dual structure appears as thesubject, and the conjuncts are definite:
(66) xabto-xøh yaxodyeyo-xøh tyukoxøna xødirnga-xøh.reindeer.buck-DU reindeer.doe-DU here graze-3DU‘The reindeer buck and reindeer doe are grazing here.’
In contrast, the double dual structure cannot appear as a nonsubject with indef-inite conjuncts, as shown in (67). Instead, the accidental coordination structurein (68) must be used.
(67) * xabto-xøh yaxodyeyo-xøh temtaødom.reindeer.buck-DU reindeer.doe-DU bought.1SG
‘I bought a reindeer buck and a reindeer doe.’
(68) xabto-m yaxodyeyo-m temtaødom.reindeer.buck-ACC reindeer.doe-ACC bought.1SG‘I bought a reindeer buck and a reindeer doe.’
We now turn to an examination of adjective agreement in Tundra Nenets.Closest conjunct agreement is not allowed for the speakers we consulted. This isseen in (69), in which the adjective cannot agree only with the first dual-markedconjunct:
(69) * søwa-xøh lyekarø-xoh toxolkoda too-q.good-DU doctor-DU teacher arrived-3PL
‘The good [(two) doctors and teacher] arrived.’
6The subject noun phrase in (65) could in principle be an accidental coordination structurereferring to two doctors and two teachers, which would require plural and not dual agreementon the verb.
27
The only way of conveying the meaning intended by (68) is to repeat the adjec-tive:
(70) søwa lyekaro (nyabyi) søwa toxolkoda too-xøh.good doctor & good teacher arrived-3DU‘A [good doctor] and a [good teacher] arrived.’
As in Finnish, adjectival modification is not possible for an accidental coordi-nation structure with conjoined singular nouns:
(71) *søwa / *søwa-xøh lyekaro (nyabyi) toxolkoda too-xøh.good / good-DU doctor & teacher arrived-3DU‘A good [doctor and teacher] arrived.’
The explanation for the ungrammaticality of (71) exactly mirrors the Finnishfacts: the Tundra Nenets adjective requires both concord and index agreement,and cannot modify accidental coordination structures with singular conjuncts,since the concord and index features in those structures do not match.
In contrast, and again like Finnish, it is possible to modify the double dualnatural coordination structure with a dual adjective:
(72) søwa-xøh / *søwa lyekarø-xoh toxolkoda-xøh too-xøh.good-DU / good doctor-DU teacher-DU arrived-3DU‘A good [doctor and teacher] arrived.’
Thus, the same explanation holds for the Tundra Nenets data as for the Finnishdata: Tundra Nenets dual adjectives require dual concord and index, andthe double dual natural coordination structure in Tundra Nenets has matchingconcord and index features.
5.2 Russian
Corbett (1983, 2000) points out a correlation between animacy and adjectivalagreement with coordinated nouns in Russian: as with Finnish, plural modifiersare preferred with animate definite conjoined nouns.7 Example (73), adaptedfrom Corbett (1983, p. 20), exemplifies a contextually-established natural co-ordination relation, which holds between salient entities that are likely to beanimate and definite.
(73) Marija zadumalas’ ob ostavlennyx / *ostavlennomMaria thought about left.behind.LOC.PL / left.behind.LOC.MASC.SG
muze i doceri.husband.LOC.SG & daughter.LOC.SG
‘... Maria thought about her [husband and daughter] (who had been) left behind.’
7King and Dalrymple (2004) classify Russian determiner agreement as involving onlyindex agreement, thus patterning with the Hebrew data described above. We believe thatthis is a misclassification, based on examining only instances of natural coordination, and wereclassify Russian determiners and adjectives as outlined in this section.
28
As with other cases of natural coordination, the plural adjective is acceptable,and a singular adjective is impossible.
Singular modifiers are required in structures involving accidental coordina-tion, which correlates with inanimacy and indefiniteness. As in Finnish, this isin fact closest conjunct agreement, as seen in the following two examples, basedon Corbett (1983, p. 160). The possessive modifier shows feminine singularagreement with the first conjunct in (74) and neuter singular agreement withthe first conjunct in (75):
(74) so svoej / *svoim / *svoimiwith his.own.INST.FEM.SG / his.own.INST.MASC.SG / his.own.INST.PL
zenoj i domom.wife.INST.FEM & house.INST.MASC
‘... with his own [wife and house].’
(75) nazovite vase / *vasu / *vasigive your.ACC.NEUT.SG / your.ACC.FEM.SG / your.ACC.PL
imja i familiju.first.name.ACC.NEUT & surname.ACC.FEM
‘... give your [first name and surname].’
The modifier cannot agree with the second conjunct, and it cannot be plural.A contextually-established natural coordination relation is more likely for a
coordinate structure with inanimate conjuncts which appears in subject posi-tion. In such cases, plural modification is possible. Compare (75) with (76), inwhich the coordinate structure appears in subject position:
(76) Vase / vasi imja i familija krasivye.your.NEUT.SG / your.PL first.name.NEUT & surname.FEM beautiful.PL‘Your [first name and surname] are beautiful.’
In (76), the speaker has a choice to construe the subject either as natural or asaccidental coordination. Natural coordination requires a plural adjective, andaccidental coordination requires closest conjunct agreement.
These facts are easily explained if, as in Finnish and Nenets, Russian adjec-tives require both concord and index agreement, and contextually-establishednatural coordination is expressed in terms of a structure which has matchingconcord and index features.
5.3 Bahdinani Kurdish
In Bahdinani Kurdish (Indo-European), the distinction between natural and ac-cidental coordination is marked explicitly. This language requires izafe markingto be present on the head noun when it has a possessor or an adjective (Jardine1922).8 The izafe marking expresses the gender and number of the head noun.
8We cite examples from Jardine’s grammar in his transcription, while examples providedby our informant Newzad Hirori follow a different transcription convention.
29
(77) a. mer-e pırman-IZ.MASC.SG old‘the old man’ (Jardine 1922)
b. mal-et gundhouse-IZ.PL village‘the houses of the village’ (Jardine 1922)
In the following, we will not discuss gender agreement with izafe marking.With conjoined nouns, the izafe marking is attached to the second noun. In
accidental coordination with inanimate abstract nouns, the izafe marking mustbe singular:
(78) dilovani u ciwani-ya wıkindness & beauty-IZ.SG (*IZ.PL) his‘his [kindness and beauty]’ (N. Hirori, p.c.)
With intrinsic natural coordination, where the nouns form a natural or conven-tional pair, the izafe must be plural:
(79) deyk u bab-et wımother & father-IZ.PL (*IZ.SG) his‘’his [mother and father] (= his parents)’ (N. Hirori, p.c.)
But for many nouns, the izafe can be either singular or plural, depending onthe semantico-pragmatic construal. This is an example of contextually inducednatural coordination. In most instances, the izafe is singular, as in (80). How-ever, if the nouns are conceived as being closely associated with each other, theplural izafe is used. According to the intuitions of our informant, example (81)represents a closer relation between the hat and coat than is expressed in (80):
(80) kulav u cakit-e wıhat & coat-IZ.SG hi‘his hat and coat’ (N. Hirori, p.c.)
(81) kulav u cakit-et wıhat & coat-IZ.PL his‘his hat and coat’ (N. Hirori, p.c.)
From these examples, it is clear that izafe marking represents the concord
num of the coordinated nouns: singular for accidental coordination structureswhere each noun is singular, and plural for natural coordination structures,where the natural coordination structure has plural concord. The izafe mark-ing does not represent the index num, since in both natural and accidentalcoordination the index is plural. No data is available from modifier agreement,since adjectives do not show number agreement in Bahdinani Kurdish. Verbagreement patterns are complicated, and it is not always clear how it is deter-mined; in some examples, closest conjunct verb agreement seems to be required,while in other examples, the verb shows the same agreement as the izafe markerand should perhaps be analyzed as concord agreement.
30
6 Conclusion
We have seen that syntactically coordinate structures can express two differentsemantic relations: accidental coordination and natural coordination. Naturalcoordination involves a salient relation between two entities which can be intrin-sic, as with conventionalized pairings like father and mother or sun and moon,or can be established contextually. Recent research has shown that in many lan-guages, natural and accidental coordination structures are distinguished mor-phosyntactically, at the level of constituent structure, with natural coordinationtending to be less morphosyntactically marked than accidental coordination.
Our work shows that in Finnish and some of the other languages we haveexamined here, there is no difference in constituent structure between the twotypes of coordination. Rather, the difference is apparent at functional structure,and is shown by patterns of modifier agreement: natural coordination struc-tures are represented as a single functional structure, reflecting the fact thatthey contain a bounded number of conjuncts, usually two. Furthermore, in thenatural coordination structure, the concord and index features match, allowingmodification in languages in which both concord and index requirements mustbe satisfied. The functional structure of natural coordination structures thatwe propose is more like that of a compound or plural lexical noun than acci-dental coordination structures; this is as expected, since as shown by Walchli(2005), the semantic relation of natural coordination is often expressed as aco-compound.
In contrast, accidental coordination structures are formally most appropri-ately represented as sets of functional structures, reflecting the fact that ac-cidental coordination structures can have an unbounded number of conjuncts.The index and concord values of these structures exhibit a mismatch when theconjuncts are singular nouns: the concord value is singular, while the index
is plural. This entails that there is no possibility for modification in languageswhere modifiers must agree both in index and concord.
Acknowledgments
The research reported here is supported by a grant from the Arts and Human-ities Research Council, UK (MRG-AN10939/APN17606). The Tundra Nenetsdata were collected in fieldwork supported by the Endangered Languages Doc-umentation Programme (ELDP), SOAS, London. We are grateful to our infor-mants, especially Riho Grunthal, Johanna Laakso, Marja Leinonen, and TapaniSalminen for Finnish, Galina Koreneva and Anna Lamdo for Nenets, DevyaniSharma for Hindi, Jonathan Ginzburg for Hebrew, and Newzad Hirori for Bah-dinani Kurdish.
For very helpful discussion and comments, we thank Greville Corbett, DavidCram, Bill Davies, Lauren Delfs, Jonathan Ginzburg, Brian Joseph, Tracy King,Johanna Laakso, Philomen Probert, Kevin Ryan, Louisa Sadler, Tapani Salmi-nen, Devyani Sharma, Ida Toivonen, Aline Villavicencio, Bernhard Walchli, andaudiences at the 2004 meeting of the Linguistic Association of Great Britain,
31
the University of Oxford, and the York-Essex Morphology Meeting 2 (2004) atthe University of Essex.
Appendix: Finnish questionnaire
Below is the questionnaire that was used to collect the Finnish data presentedin Section 2. The sentences were presented to the speakers in a random order.Here, the questions are grouped according to the responses that were obtained:first those for which the answers were unanimous, then those that producedvariation. The number following the sentence represents the number of speakerswho produced a plural adjective, out of the seven who were consulted.
It is difficult to design a compatible theatre and university. 7You can rarely meet a resembling (each other) brother and sister. 7I have never seen a similar-looking mother and daughter. 7It is difficult to find a compatible cat and dog. 7I have never seen a mother and daughter resembling each other. 7I bought a matching skirt and blouse. 7The lucky brother and sister won £100. 7Thank God, the unfriendly husband and wife moved out. 7Stupid Mikko and Juha left their jobs in a hospital and a bank. 7The happy man and boy left together holding hands. 7
I bought a new skirt and blouse. 0A good teacher and doctor arrived in the village. 0I want to buy an old cat and dog. 0Once upon a time a poor rabbit and chicken lived together in a small house. 0You own a beautiful house and car. 0You can rarely meet a faithful husband and wife. 0I discovered an ill dog and horse in the barn. 0I want to buy a small rabbit and chicken. 0There is a good university and theatre in this city. 0He cannot find a clean fork and spoon. 0I saw a strange man and boy who walked on opposite sides of the street 0I will never forget the wonderful day and night in Paris. 0He was building a wooden boat and ship. 0You can see the beautiful moon and stars tonight. 0I like her strange face and voice. 0He bought a new house and car with a week in between. 0Once there was a poor man and woman. 0I only had a fresh orange and pear for breakfast. 0
I met a worried (single) mother and (single) father at a parents’ meeting. 1Her amazing kindness and beauty attracts people. 1I gave the book to the happy man and boy. 1The new table and chairs look nice. 2
32
I don’t like my furniture: the old table and shelf look ugly. 2The good university and theatre located in this city attract many people. 2The clean fork and knife are already on the table. 2I will take the healthy animals with me, but the ill dog and horse will stay here. 3My animals are quiet: the old cat and dog always sleep. 3The new chairs and table look nice. 4You can see the beautiful stars and moon tonight. 4
References
Allan, Keith. 1987. Hierarchies and the choice of left conjuncts (with particularattention to English). Journal of Linguistics 23(1), pp. 51–77.
Benor, Sarah Bunin and Roger Levy. 2006. The chicken or the egg? A proba-bilistic analysis of English binomials. Language 82. To appear.
Bresnan, Joan. 2001. Lexical-Functional Syntax . Oxford: Blackwell Publishers.
Carpenter, B. 1992. The Logic of Typed Feature Structures. Cambridge, UK:Cambridge University Press.
Comrie, Bernard. 1989. Language universals and linguistic typology. Chicago:The University of Chicago Press, second edition.
Comrie, Bernard. 2003. When agreement gets trigger-happy. Transactions ofthe Philological Society 101(2).
Corbett, Greville G. 1983. Hierarchies, Targets, and Controllers: AgreementPatterns in Slavic. London: Croom Helm.
Corbett, Greville G. 2000. Number . Cambridge, UK: Cambridge UniversityPress.
Dalrymple, Mary. 2001. Lexical Functional Grammar , volume 34 of Syntax andSemantics . New York, NY: Academic Press.
Dalrymple, Mary and Ronald M. Kaplan. 2000. Feature indeterminacy andfeature resolution. Language 76(4), pp. 759–798.
Dalrymple, Mary, Ronald M. Kaplan, John T. Maxwell, III, and Annie Zaenen(editors). 1995. Formal Issues in Lexical-Functional Grammar . Stanford,CA: CSLI Publications.
Falk, Yehuda N. 2001. Lexical-Functional Grammar: An Introduction to Par-allel Constraint-Based Syntax . Stanford, CA: CSLI Publications.
Givon, Talmy. 1976. Topic, pronoun, and grammatical agreement. In CharlesLi (editor), Subject and Topic, pp. 57–98. New York: Academic Press.
Haspelmath, Martin (editor). 2004a. Coordinating Constructions . Amsterdam:John Benjamins.
33
Haspelmath, Martin. 2004b. Coordinating constructions: An overview. In Mar-tin Haspelmath (editor), Coordinating Constructions . Amsterdam: John Ben-jamins.
Haspelmath, Martin. to appear. Coordination. In Timothy Shopen (editor),Language typology and linguistic description. Cambridge, UK: CambridgeUniversity Press, second edition.
Heycock, Caroline and Roberto Zamparelli. 2003. Coordinated bare definites.Linguistic Inquiry 34(3), pp. 443–469.
Jardine, R. F. 1922. Bahdinan Kurmanji: a grammar of the Kurmanji of theKurds of Mosul Division and surrounding districts of Kurdistan. Baghdad:Government Press.
Johannessen, Janne Bondi. 1998. Coordination. Oxford: Oxford UniversityPress.
Kaplan, Ronald M. and John T. Maxwell, III. 1988. Constituent coordinationin Lexical-Functional Grammar. In Proceedings of the 12th InternationalConference on Computational Linguistics (COLING88), Budapest, volume 1,pp. 303–305. Reprinted in Dalrymple et al. (1995, pp. 199–210).
Kathol, Andreas. 1999. Agreement and the syntax-morphology interface inHPSG. In Robert Levine and Georgia Green (editors), Studies in CurrentPhrase Structure Grammar , pp. 223–274. Cambridge, UK: Cambridge Uni-versity Press.
King, Tracy Holloway and Mary Dalrymple. 2004. Determiner agreement andnoun conjunction. Journal of Linguistics 40(1), pp. 69–104.
Lambrecht, Knud. 1984. Formulaicity, frame semantics and pragmatics in Ger-man binomial expressions. Language 60(4), pp. 753–796.
Lambrecht, Knud. 1994. Information structure and sentence form. Cambridge,UK: Cambridge University Press.
Minassian, Martiros. 1980. Grammaire d’armenien oriental . Delmar: Caravan.
Moltmann, Friederike. 1992. Coordination and Comparatives . Ph.D. thesis,MIT.
Moyse-Faurie, Claire and John Lynch. 2004. Coordination in Oceanic languagesand Proto Oceanic. In Martin Haspelmath (editor), Coordinating Construc-tions . Amsterdam: John Benjamins.
Nikolaeva, Irina and Maria Tolskaya. 2001. A Grammar of Udihe. Berlin:Mouton de Gruyter.
Peterson, Peter G. 2004. Coordination: Consequences of a Lexical Functionalaccount. Natural Language and Linguistic Theory 22(3), pp. 643–679.
34
Pollard, Carl and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar .Chicago: The University of Chicago Press.
Sadler, Louisa. 1999. Non-distributive features and coordina-tion in Welsh. In Miriam Butt and Tracy Holloway King (ed-itors), On-line Proceedings of the LFG99 Conference. URLhttp://csli-publications.stanford.edu/LFG/4/lfg99.html.
Sadler, Louisa. 2003. Asymmetric coordination and agreement in LFG. InMiriam Butt and Tracy Holloway King (editors), Nominals: Inside and Out .Stanford, CA: CSLI Publications.
Sag, Ivan A. 2000. Coordination and underspecification. In Jong-Bok Kim andStephen Wechsler (editors), On-line Proceedings of the HPSG ’01 Conference.URL http://csli-publications.stanford.edu/HPSG/3/hpsg02.html.
Stassen, Leon. 2000. And-languages and with-languages. Linguistic Typology4(1).
Toivonen, Ida. 2002. The relationship between agreement andgrammatical functions in Inari Sami and Finnish. URLhttp://www-cmll.concordia.ca/linguistics/toivonen/inari-www/inari.html.ms, University of Canterbury.
Walchli, Bernhard. 2005. Co-compounds and natural coordination. Oxford:Oxford University Press.
Wechsler, Stephen and Larisa Zlatic. 2000. A theory of agreement and itsapplication to Serbo-Croatian. Language 76(4), pp. 759–798.
Wechsler, Stephen and Larisa Zlatic. 2003. The Many Faces of Agreement .Stanford, CA: CSLI Publications.
35
