Coordination

Coordination

Coordination 183 Brown P (1990). ‘Gender, politeness, and confrontation in Tenejapa.’ Discourse Processes 13, 123–141. Brown P & Levinson S C (1987). ...

250KB Sizes 4 Downloads 262 Views

Coordination 183 Brown P (1990). ‘Gender, politeness, and confrontation in Tenejapa.’ Discourse Processes 13, 123–141. Brown P & Levinson S C (1987). Politeness: some universals in language use. Cambridge: Cambridge University Press. Cameron D (1985). Feminism and linguistic theory. New York: St. Martins. Chomsky N (1975). Reflections on language. New York: Pantheon. Cooper M M (1982). ‘Context as vehicle: implicatures in writing.’ In Nystrand M (ed.) What writers know: The language, process, and structure of written discourse. New York: Academic Press. 105–128. Cooper M M (1984). ‘The pragmatics of form: how do writers discover what to do when?’ In Beach R & Bridwell L S (eds.) New directions in composition research. New York: The Guildford Press. 109–126. Edwards D & Mercer N (1987). Common knowledge: the development of understanding in the classroom. London: Routledge. Fraser B (1990). ‘Perspectives on politeness.’ Journal of Pragmatics 14, 219–236. Gautam K & Sharma M (1986). ‘Dialogue in Waiting for Godot and Grice’s concept of implicature.’ Modern Drama 29, 580–586. Grandy R & Warner R (eds.) (1986). Philosophical grounds of rationality: intentions, categories, ends. Oxford: Clarendon. Grice H P (1975). ‘Logic and conversation.’ In Cole P & Morgan J L (eds.) Syntax and semantics: Speech acts 3. New York: Academic Press. 58–85. Grice H P (1989). Studies in the way of words. Cambridge: Harvard University Press. Huang Y (1991). ‘A neo-Gricean pragmatic theory of anaphora.’ Journal of Linguistics 27, 301–335. Kasher A (1976). ‘Conversational maxims and rationality.’ In Kasher A (ed.) Language in focus: foundations, methods and system. Reidel: Dordrecht. 197–216. Kasher A (1977). ‘Foundations of philosophical pragmatics.’ In Butts R E & Hintikka J (eds.) Basic problems in methodology and linguistics: part three of

the proceedings of the Fifth International Congress of Logic, Methodology and Philosophy of Science, London, Ontario, Canada-1975. Reidel: Dordrecht. 225–242. Kleifgen J A (1990). ‘Prekindergarten children’s second discourse learning.’ Discourse Processes 13, 225–242. Lehnert W (1978). The process of question answering. Hillsdale, New Jersey: Erlbaum. Lindblom K (2001). ‘Cooperating with Grice: a crossdiscplinary metaperspective on uses of Grice’s Cooperative Principle.’ Journal of Pragmatics 33, 1601–1623. Lovejoy K B (1987). ‘The Gricean model: A revising rubric.’ Journal of Teaching Writing 6, 9–18. McCarthy L P (1987). ‘A stranger in strange lands: a college student writing across the curriculum.’ Research in the Teaching of English 21, 233–265. Mey J (2002). ‘To Grice or not to Grice.’ Journal of Pragmatics 34, 911. Michell G (1984). ‘Women and lying: a pragmatic and semantic analysis of ‘‘telling it slant.’’’ Women’s Studies International Forum 7, 375–383. Ostwald M (1969). Nomos and the beginnings of the Athenian democracy. Oxford: Oxford University Press. Pratt M L (1977). Toward a speech act theory of literary discourse. Bloomington: Indiana University Press. Rundquist S (1992). ‘Indirectness: a gender study of flouting Grice’s maxims.’ Journal of Pragmatics 18, 431–449. Singer M (1990). ‘Answering questions about discourse.’ Discourse Processes 13, 261–277. Sperber D & Wilson D (1986). Relevance: communication and cognition (2nd edn. 1995). Oxford: Basil Blackwell. Suppes P (1986). ‘The primacy of utterer’s meaning.’ In Grandy R & Warner R (eds.) 109–130. Tannen D (1986). That’s not what I meant: how conversational style makes or breaks your relations with others. New York: Morrow. Tsohatzidis S (ed.) (1994). Foundations of speech act theory: philosophical and linguistic perspectives. London: Routledge.

Coordination B Crysmann, DFKI, Saarbru¨ken, Germany ß 2006 Elsevier Ltd. All rights reserved.

The term ‘coordination’ refers to the combination of like or similar syntactic units into some larger group of the same category or status, typically involving the use of a coordinating conjunction, such as and or or, to name just two. The units grouped together

by means of a coordinating conjunction are usually referred to as conjuncts (or conjoints). Within syntactic theory, coordination enjoys a special status, for several reasons: first, the properties of a coordinate structure are not projected from a single syntactic head, unlike most major syntactic phrases, which generally conform to some notion of X0 -syntax (Jackendoff, 1977). Instead, syntactic properties of the complex structure are almost always symmetrically

184 Coordination

determined by all conjuncts. Second, coordination is characterized by a syntax–semantics mismatch: in syntactic terms, it is the conjuncts that should be regarded as (multiple) heads of the construction, determining, inter alia, the syntactic category of the complex structure, whereas semantically, it is the coordinating conjunction that acts as a functor, regulating how the semantic contribution of the individual conjuncts is composed into the meaning of the complex coordinate expression. Unsurprisingly, these two characteristics pose quite a strong challenge for syntactic theories such as Dependency Grammar, which postulates that head-dependent structures are always binary, featuring a unique head. Third, the treatment of coordination also becomes nontrivial for almost any syntactic theory to date, once we leave the domain of simple constituent coordination and investigate structures where conjuncts do not appear to be classical constituents (nonconstituent coordination) or else the units to be coordinated fail to observe strong notions of similarity. The study of coordination phenomena has exerted a considerable influence on syntactic theory construction over the last three decades. In transformational syntax, the work on extraction islands pioneered by Ross (1967) has generated a good deal of interest in coordinate structures as well. Starting with Generalized Phrase Structure Grammar (GPSG), the perspective adopted by current syntactic theories, in particular Categorial Grammar (CG), Lexical Functional Grammar (LFG), and Head-driven Phrase Structure Grammar (HPSG) with respect to coodination phenomena, has shifted considerably, regarding it now as a core, rather than peripheral phenomenon of natural language that demands principled explanation. In these frameworks, the study of nonconstituent coordination and coordination of unlikes has attracted particular interest over the past few years.

Basic Properties It is a distinguishing property of coordination that it can be applied to a wide range of syntactic categories: put differently, coordinating conjunctions do not appear to be highly selective as to the syntactic categories they combine with. As illustrated by the examples in (1), a coordinating conjunct such as and can be used to combine sentences, or constituents of any major category, such as VPs, NPs, PPs, or APs. (1a) [[The moon goes round the earth]S and [the earth goes round the sun]S]. (Grover, 1994)

(1b) Jack [[fell down]VP and [broke his crown]VP]. (Grover, 1994) (1c) She promised to send [either [a letter]NP or [a postcard]NP]. (Grover, 1994) (1d) A [[very sleek]AP but [rather fat]AP] cat appeared in the garden. (Grover, 1994) (1e) The journey took them [[across desert]PP and [through jungles]PP]. (Grover, 1994)

Similarly, coordination does not seem to be confined to a particular level of projection: thus, besides clausal and phrasal categories, coordination of lexical categories is equally well attested. (2a) We will [[attack]V and [beat]V] the enemy. (Grover, 1994) (2b) Fred was [both [angry]A and [upset]A] about the incident. (Grover, 1994)

Coordination of unsaturated constituents, i.e., constituents with open valencies, is subject to the additional requirement that the conjuncts show the same degree of saturation. (3) *We will [[attack the enemy]VP and [beat]V]

While it is clear that all major lexical categories can undergo coordination, this is not always possible for all members of functional categories, such as determiners or complementizers. While coordinations of almost purely functional determiners such as a and the are hardly possible in English, other determiners do allow coordination, as in some or all people. In languages such as German, where determiners also encode number and gender information, coordination of definitites is easily possible: der oder die Lehrer ‘the(m.sg) or the(pl) teacher(s)’. Coordinating conjunctions, just like other conjunctions, are closed-class items, forming a very small set. Coordinating conjunctions can be further subdivided into noncorrelative conjunctions, such as and, or, but, and correlative conjunctions, which necessarily come in pairs: neither . . . nor, either . . . or, both . . . and. In English (and German), the first part of the correlative pair is used to mark the first conjunct, whereas the second part will be used to mark any subsequent conjuncts. In other languages, the markers for first and subsequent conjuncts may not be distinct, e.g., French ni . . . ni ‘neither . . . nor’. Minimally, a coordinated structure is binary, i.e., it must contain at least two conjuncts. There is, however, no general upper bound on the number of conjuncts, the conjunctions both . . . and or but being notable exceptions. The marking of such multiple coordinations differs between correlative and noncorrelative conjunctions: with correlative conjunctions,

Coordination 185

every conjunct needs to be marked with an appropriate conjunction, whereas with noncorrelative conjunctions, either all conjuncts or only the final conjunct will be marked. In English and German, the latter constitutes the unmarked choice. As reported in Dixon (1972), there are also languages, such as Dyirbal, where coordinate structures are not marked by a conjunction at all. In English, such conjunctionless coordinations are also possible, although this appears to be a more marked option. It is an often-made assumption that coordination involves the two central notions of likeness and constituency. Typically, the conjuncts in a coordinate structure are constituents of the same basic category and level of projection. With respect to the likeness condition, morphosyntactic features do not seem to pattern alike: in the nominal domain, it is only case features that need to be identical between all conjuncts and the entire coordinate structure: all other common inflectional features, such as person, number, or gender, can assume different values in each of the conjuncts. (4a) *der Vater und den Sohn the.nom.sg.m father and the.acc.sg.m son (4b) der Vater und die So¨ hne the.nom.sg.m father and the.nom.pl.m sons (4c) [der Vater] und [die Tochter] the.nom.sg.m father and the.nom.sg.f daughter

Similarly, in the verbal domain, basic distinctions regarding verb form such as finiteness are required to be identical across conjuncts, whereas tense, mode, and aspect distinctions may well differ: (5) He [[arrived yesterday] and [will leave again tomorrow]].

This different status of morphosyntactic features with respect to categorial identity is, of course, related to the primary function of coordination, namely that of creating aggregates of individuals or events: it is therefore a necessity that, for example, the number specification of a coordinated NP may differ from those of its conjuncts. Exemption of person and number features from the identity requirement gives rise to interesting interaction with agreement relations: depending on the choice of coordinating conjunction, namely whether it is ‘conjunctive’ like and or ‘disjunctive’ like or, a coordination of singular NPs may show either plural or singular agreement. In languages such as German or Spanish, person agreement between a coordinated subject NP and the verb depends on whether one of the conjuncts refers to the speaker or hearer: if one of the conjuncts refers to the speaker or hearer, the

verb inflects for 1st or 2nd person plural, respectively. Otherwise, it is inflected for 3rd person plural. (6a) [Jose´ i yo] hablamos / *habla´ is / *hablan Jose and I speak-1pl/speak-2pl/speak-3pl (Dalrymple and Kaplan, 2000) (6b) [Jose´ i tu] *hablamos/ habla´ is/ *hablan Jose and you speak-1pl/speak-2pl/speak-3pl (Dalrymple and Kaplan, 2000) (6c) [Jose´ i Maria] *hablamos/ *habla´ is / hablan Jose and you speak-1pl/speak-2pl/speak-3pl

In German, agreement patterns with ‘disjunctive’ coordinating conjunctions depend on the linear order of the conjuncts: as illustrated by the examples in (7), subject-verb agreement is determined by the conjunct that appears closest to the verb. (7a) Gestern yesterday oder du or you (7b) Gestern yesterday oder ich or I

habe / *hast have-1sg / have-2sg geschnarcht. snored *habe / hast have-1sg / have-2sg geschnarcht. snored

entweder either

ich I

entweder either

du you

Phrase structure grammar approaches (Chomsky, 1957; Jackendoff, 1977; Gazdar et al., 1985; Kaplan and Maxwell, 1988) typically invoke special phrase structure rules to license coordinate structures, e.g., (8), where XP is interpreted as a variable ranging over (major) syntactic categories: (8) XP ! XP conj XP

In HPSG, likeness of category and level of projection between conjuncts is enforced by the Coordination Principle (Pollard and Sag, 1994: 202), which requires sharing of category information, including valence and head features, between the mother and all conjuncts. HPSG draws a principled distinction between head features such as CASE, which are part of the categorial information (CAT), and INDEX features such as person, number, and gender, which are part of semantics (CONT). As a result, the different behavior of number and case features in coordinate structures is directly related to an independently motivated feature geometry in HPSG. LFG uses phrase structure rules akin to (8) to model the constituent structure of coordination constructions. The f-structure representation of the entire coordinate structure is a set, consisting of the f-structures of the individual conjuncts. Following Kaplan and Maxwell (1988), constraints imposed on the f-structure of the mother node, e.g., external case requirements, will get distributed across all the element f-structures. In order to capture the apparent contrast between case and number or gender, features

186 Coordination

are declared explicitly as either distributive or nondistributive. Grammatical functions and case are normally taken as distributive features; person and number gender features are taken as nondistributive. While feature-based approaches to grammar have been successful in providing solutions for the agreement patterns in (6) (Dalrymple and Kaplan, 2000; Sag, 2003), the linear effects observed with German ‘disjunctive’ coordination still await treatment in the context of phrase structure grammar.

Coordination and Extraction Coordination also witnesses theoretically interesting interactions with the syntax of extraction. Since Ross (1967), it is a quite well-known fact that extraction out of coordinate structures is subject to strong restrictions, as illustrated by the data in (10). (9a) Nuclear physicsi, I have never understood i. (Grover, 1994) (9b) Whoi did Fay send a letter to i. (Grover, 1994) (10a) *Nuclear physicsi, I have never understood [[organic chemistry] and [i]]. (Grover, 1994) (10b) *Whoi did Fay [[send a letter to i] and [forget all about it]]. (Grover, 1994)

While extraction of conjuncts is completely unacceptable, extraction out of conjuncts is only possible if it applies across the board (ATB), i.e., if the filler corresponds to a gap in every conjunct, as in (11). (11a) Nuclear physicsi, [[Fay wants to study i] and [May wants to drop i]]. (Grover, 1994) (11b) Whoi did Fay send a letter to i and receive a reply from i. (Grover, 1994)

There exist, at least in English, a few examples that might serve to question the validity of the ATB constraint on extraction from coordinate structures, superficially suggesting asymmetric extraction from the right conjunct only. (12) Nuclear physicsi, Fay wants to try and study i. (Grover, 1994)

Many researchers, however, agree that in these examples – termed asymmetric coordination or even pseudocoordination – the conjunction actually functions more like a complementizer. See, however, Lakoff (1986) for more harmful exceptions to the ATB constraint. These restrictions, with the exception of ATB extraction, are captured by Ross’s (1967) Conjunct Structure Constraint (CSC):

In a coordinate structure, a. no conjunct may be moved b. nor may any element contained in a conjunct be moved out of that conjunct.

This constraint obviously consists of two parts, essentially capturing independent observations: first, an absolute ban on the extraction of the conjuncts themselves, and second, an extension of the similarity requirement to the gaps contained in the conjuncts. The two individual clauses of the CSC are referred to as Conjunct Constraint and Element Constraint, respectively, a distinction introduced by Grosu (1973). In contrast to other constraints on extraction, Transformational Grammar has long failed to provide a more principled account of the CSC and ATB extraction, as observed by Roberts (1997). Featurebased frameworks, however, such as GPSG, where information about extracted constituents is encoded by means of a SLASH feature, have been quite successful in deriving the Element Constraint as just another instance of categorial similarity (Gazdar et al., 1985). The approach to constituent coordination taken by HPSG is essentially identical, requiring sharing of local syntactic features, including category, open valencies, and case, alongside sharing of nonlocal features (Pollard and Sag, 1994: 202). It is of note that Categorial Grammar, despite technical differences, arrives at a conceptually highly comparable solution, requiring identity of complex categories (Steedman, 1985, 1990). The extended notion of categorial similarity that encompasses properties of the gap can also straightforwardly capture the apparent requirement that gaps in ATB extraction must always be of the same type in each conjunct, as witnessed by the unacceptability of examples, such as *On whom does Fay rely _ and Kim dislike_. Although an extended notion of similarity including nonlocal features proves quite successful in deriving the Element Constraint and ATB restriction, it does not yet derive all instances of the Conjunct Constraint. (13) *John, I saw [ _ and _ ]

An interesting explanation for the impossibility of coordinating gaps has been offered by Pollard and Sag (1994): they argue that an account of the Conjunct Constraint follows naturally from traceless theories of extraction. In these theories, nonlocal dependencies are launched directly by the subcategorizing head, instead of being mediated by a phonetically empty syntactic element. Since none of the conjuncts selects the other, and since external heads do not subcategorize for individual conjuncts of the

Coordination 187

coordinate structure, the Conjunct Constraint is merely a correlate of traceless theories of extraction. In LFG, where extraction is treated by way of functional uncertainty, the ATB constraint is also a correlate of a more general theory of constituent coordination: since the f-structure of coordinated constituents is a set, and constraints imposed on this set are distributed over the element f-structures (Kaplan and Maxwell, 1988), the functional uncertainty constraint that relates the filler to the grammatical function governed by some predicate will have to be satisfied by all the f-structures in this set.

Nonconstituent Coordination The coordination phenomena reviewed so far were mostly of a quite simple type, namely coordination of like constituents, be they lexical, phrasal, or clausal, which are composed into a complex constituent of the same type. By and large, constituent coordination does not pose any serious challenge for contemporary formal syntactic theories. Human language, however, also witnesses cases of coordination that appear to involve units that are generally not considered constituents, hence the cover term nonconstituent coordination (NCC). (14a) Bill gave the girls spades and the boys recorders. (CR; Maxwell and Manning, 1996) (14b) Bill likes and Joe is thought to like cigars from Cuba. (RNR; Maxwell and Manning, 1996) (14c) Bill gave a rhino to Fred, and Sue a camera to Marjorie. (Gapping; Maxwell and Manning, 1996)

There are different ways of understanding NCC data in descriptive terms: one way is to think of NCC as involving not coordination of constituents, but rather coordination of sequences of constituents. An alternative view is to regard NCC as a subcase of constituent coordination, where identical material is either shared between the two conjuncts, or deleted, under identity, an interpretation that gives one of the conjuncts the status of an incomplete or ‘elliptical’ expression. (15a) Bill gave the girls spades and (Bill gave) the boys recorders. (15b) Bill likes (cigars from Cuba) and Joe is thought to like cigars from Cuba. (15c) Bill gave a rhino to Fred, and Sue (gave) a camera to Marjorie.

There is, however, no consent yet as to whether all types of NCC should be given some unified treatment, or whether we are actually dealing with only

superficially similar phenomena that call for independent solutions: while both Conjunction Reduction (CR) or VP-gapping (14a) and Right-Node Raising (RNR; [14b]), can be interpreted as sharing of peripheral material, it is central material that gets shared or elided in gapping constructions. The well-formedness of coordination involving peripheral sharing is often related to the fact that the well-formedness of a string ‘A [X and Y] B’ can be directly related to that of the strings ‘A X B’ and ‘A Y B.’ Gapping, on the other hand, does not appear to follow this general pattern. Maxwell and Manning (1996) argue that verb medial gapping (or just gapping; [14c]) also differs from both CR and RNR in that it does not feature the same type of strict identity and parallelism requirements as peripheral sharing. (16a) A bandicoot escaped today and more will tommorow. (Blevins, 1994; taken from Maxwell and Manning, 1996) (16b) A lot of this material can be presented in a fairly informal and accessible fashion, and often I do. (Chomsky, 1982; taken from Maxwell and Manning, 1996)

As shown by the data in (16) above, the elided material in the second conjunct cannot be replaced verbatim with the corresponding string from the first conjunct. Similar observations can be made for true ellipsis. Maxwell and Manning (1996) suggest that gapping phenomena should be regarded outside the scope of syntactic theory and should be treated at the level of semantics, using more powerful devices such as higher-order unification, as suggested by Dalrymple et al. (1992) for the treatment of ellipsis. Peripheral Sharing

According to Milward (1994), approaches to NCC can be assigned to one of two types: deletion accounts, or shared structure approaches. Contrary to what the terms may suggest, the main difference between these two types of approaches does not lie with the deletion vs. sharing or merging opposition – a more or less metaphorical issue – but rather depends on what role syntactic structure should play to restrict NCC, in addition to identity of peripheral strings of lexical items. It is clear that pure surface identity cannot suffice to provide a restrictive theory: (17a) [John will drive] and [Mary built the drive]. (Milward, 1994) (17b) *[John will] and [Mary built the] drive. (Milward, 1994)

Minimally, identity of syntactic category must be taken into account as well, to rule out examples

188 Coordination

such as (17) above. Most probably, identity requirements will also need to make reference to lexical meaning, if one wants to account for the marginality of (18), which are only acceptable as a rhetorical figure known as ‘zeugma.’ (18a) *John bored [the new hole] and [his fellow workers]. (Milward, 1994) (18b) *Mary came in [a hurry] and [a taxi]. (Milward, 1994)

Identity of reference, however, appears to be too strong a requirement, at least in the domain of events. In a sentence like (19), the shared verb gave clearly denotes two independent giving events. (19) John gave Mary a book on Wednesday and several records on Saturday.

Similarly, even at the level of individuals, there is reason to doubt that sharing (or deletion under identity) presupposes identity of reference. (20a) [Johni bought] and [Billj hired] a suit for his{i,j},k wedding. (20b) Three men died [in Bagdad on Thurday] and [in Tikrit on Friday night]. (Beavers and Sag, 2004)

While it is quite uncontroversial that surface form, syntactic category, as well as lexical meaning constitute the minimal identity requirements for peripheral sharing at the word level, it is not fully clear to what degree structural or functional properties are necessary to ensure well-formedness of CR or RNR constructions. At least for RNR, identity of shared material is not reducible to mere identity of surface strings, but should also take into account the internal structural representation: (21) *Sue sawi the manj [through the telescope]i and [with the troublesome kid]j. (Milward, 1994) (22a) I found [[a friend [of [Mary’s]]] handbag]. (22b) I found [the manufacturer [of [[Mary’s] handbag]]]. (22c) *I found a friend of and the manufacturer of Mary’s handbag. (Milward, 1994)

With CR, acceptability also appears to degrade once structures fail to exhibit a certain degree of structural or functional parallelism: speakers of English seem generally divided as to the acceptability of examples like (23), where one conjunct exhibits dative shift, but the other one does not. (23) (*) John gave Mary a book and a record to Peter. (Crysmann, 2003)

Although it may be tempting to relate the unacceptability of (23) to a mismatch in subcategorization

requirements of the shared verb give, as suggested by Maxwell and Manning (1996), heavy NP shift patterns just the same for these speakers, even though subcategorization frames are identical across the two conjuncts. This is further supported by the observation that acceptability of (23) improves, once Dative Shift and Heavy NP Shift are combined, such that surface order of theme and goal arguments are again in parallel. (24) (*) John gave a book to Mary and to Peter a very old and famous early Beatles record featuring vocals by Ringo. (Doug Arnold; Crysmann, 2003) (25) John gave Mary a book, and to Peter, a record. (Robert Levine; Crysmann, 2003)

As argued by Whitman (2002), acceptability of the nonparallel cases also improves, once a suitable context is provided. (26) Q: So Neal, did you give the dog that toy like you were planning to do? (27) A: No, actually I changed my mind. I gave the dog a bone, and the toy to the cat.

Here, nonparallelism regarding syntactic form and the linear order of goal and theme arguments is apparantly motivated by considerations of information structure: in terms of given–new partitioning, both conjuncts are actually fully parallel. To summarize, NCC imposes identity requirements on shared material, both at the lexical and at the structural level. Lexically, identity is restricted to surface form, syntactic category and core lexical meaning, but presumably excludes identity of reference. At the syntactic level, constituency of shared material appears to be a minimal criterion. As a result, Deletion Accounts, where identity is restricted to the lexical level, must be discarded (Milward, 1994). The question as to whether likeness of argument structure should be included as a necessary criterion is still not entirely settled. It appears, though, that strict identity of subcategorization frames may be violated, if some sort of parallelism between conjuncts is preserved, either by means of linear order of semantic role bearers, or else parallel information and structural partitioning. The perspective on CR and RNR as sharing of peripheral material across two conjuncts – Milward (1994)’s Shared Structure approaches – is quite a popular approach, which can be found in different technical incarnations across frameworks, including multidimensional approaches, such as Goodall (1987) or Moltmann (1992), coordination of partial constituents (Maxwell and Manning, 1996), or

Coordination 189

linearization-based accounts (Crysmann, in press; Yatabe, 2000, 2003). What is common to all these approaches is that they make rather conventional assumptions about constituent structure and rely on extraneous machinery to perform the necessary sharing or reconstruction of surface material. On the other side of the spectrum, we find the kind of analysis advanced within the framework of Categorial Grammar, in particular (Steedman, 1985; Dowty, 1988), which employs a somewhat weaker notion of constituency. In Categorial Grammar, the central problem of NCC can be resolved, e.g., by means of an operation such as type raising, i.e., rules capable of converting a (sequence of) arguments into a functor. In these approaches, a sequence of two NPs such as [the girls] [spades], or [the boys] [recorders], respectively, will be type raised to a VP constituent lacking the verb. Such type-raised constituents can then be coordinated in the usual fashion. What makes this solution particularly attractive is that type raising is independently motivated in CG for the treatment of unbounded dependencies. Even though type raising has proven to be a convenient tool to provide simple and elegant accounts of NCC, this approach has repeatedly been criticized (Milward, 1994; Maxwell and Manning, 1996; Beavers and Sag, 2004), mainly for its lack of restrictiveness: as detailed in Milward (1994) and Maxwell and Manning (1996), the unconstrained application of type raising essentially conflates differences regarding the internal structure of the conjuncts: as a net result, the ungrammaticality of sentences like (22) remains unaccounted for, similar to Deletion accounts. Beavers and Sag (2004) argue that the apparent attractiveness of having a single operation for extraction and NCC makes typologically wrong predictions, precluding languages, such as Hausa, which feature long-distance extraction but fail to allow RNR (Davis, 1993). Phrase structure-based approaches to syntax have long failed to provide a satisfactory treatment of NCC phenomena in their entirety. Over the past decade, however, considerable progress has been made: owing to the fact that the commitment to classical constituents is somewhat stronger in LFG and HPSG than it is in CG, approaches to NCC in phrase structure grammar uniformly address the issue by ‘‘moving coordination up a level while maintaining classical notions of constituency’’ (Maxwell and Manning, 1996: 7). In LFG, the treatment of NCC most crucially depends on an extension of the phrase structure rule formalism: given the two c-structure rules in (28), Maxwell and Manning (1996) suggest that the phrase structure component should be generalized in such

a way as to also permit coordination of partial constituents, to be licensed by rules such as (29). (28a) VP ! VP and VP (28b) VP ! V (NP) (NP) PP* (29) VP ! V [[NP PP] and [NP PP]] PP

To achieve this, Maxwell and Manning (1996) suggest that the expansion of c-structure rules may be suspended at any point to permit coordination of partial constituents. Thus, c-structure node labels are augmented with states (stacks) that encode the current level of expansion, such that expansion can later be resumed. Thus, a node label VP-x will encode suspension of the VP rule in (28b) at some point, e.g., after the verb, whereas x-VP will correspond to resumption/continuation of the expansion. The coordination rule is then extended to permit coordination of partially expanded rules, provided their suspension and resumption states will match. In essence, the NCC constructions will receive a schematic c-structure representation as in (30) (30) XP ! XP-x [x-XP-y Conj x-XP-y] y-XP

Expansion states associated with the nodes are stacks, such that NCC involving more deeply embedded partial constituents will become possible, as witnessed by the example in (31), where expansion of both the VP rule and the PP rule is suspended. (31) John flew to London on Monday and Paris on Tuesday. (Maxwell and Manning, 1996) (32) [flew [to]PPy] VPx [xVP [yPP[L.] [on Mon]] and xVP[yPP[P.] [on Tue]]].

These stacks of expansion states must match up. Maxwell and Manning (1996) then relate the unacceptability of (22) to the simple fact that right node raising in the first conjunct involves suspension of two rules, viz., of the outer NP and PP embedded in the prenominal possessive, whereas in the second conjunct, it only involves suspension of the complement PP. HPSG approaches to NCC (Yatabe, 2000, 2003; Crysmann, 2003, in press; Beavers and Sag, 2004) uniformly build on the notion of complex order domains (Reape, 1994; Kathol, 1999), which was originally developed for the treatment of linearization in free word order languages. Complex order domains provide a flat, linear representation of the main constituents of a sentence that is partially independent of immediate constituency, and which therefore facilitates the expression of word order regularities in considerably larger domains (33). The basic intuition behind HPSG accounts of NCC is that identical peripheral elements in an order domain can be collapsed in coordinate structures,

190 Coordination

ence identity. Approaches differ, though, as to the way disjoint information coming from different conjuncts is resolved on the order domain of the coordinated structure: while Yatabe (2003) suggests composing conflicting features into disjunctive or distributive representations, Crysmann (2003, in press and Beavers and Sag (2004) require asymmetric sharing of entire domain objects between the mother and one of the conjuncts, and selectively equate across all conjuncts only those feature paths that are known to be necessary to ensure well-formedness of peripheral sharing, e.g., HEAD, PHON, and PRED. Crysmann (2003, in press) and Yatabe (2000) agree, however, that the entire nonshared rest is composed into a single domain object whose syntactosemantic information is equated with that of the entire construction. Following Crysmann (2003), a sentence like (14a) will receive a representation as in (35).

where identity includes basic syntactic category, or HEAD-values, surface form (PHON), the semantic key relation, as well as (partial) bracketing. Somewhat simplifying, a phrase structure schema for coordinate structures with peripheral sharing can roughly be represented as in (34). As mentioned above, the sharing of entire domain objects across the conjuncts is oversimplifying: since domain objects do contain full syntactosemantic information, including individual and event variables, sharing entire domain objects would make empirically wrong predictions, e.g., with respect to refer-

Although none of the published HPSG work on NCC mentions this explicitly, sharing nonparallel structure as in (22) can be ruled out, because order domains are partially structured representations. Multidimensional (or 3D) approaches, such as Goodall (1987) or Moltmann (1992), are another subtype of Shared Structure Accounts, where shared phrase markers are merged between (sentential) conjuncts, yet phrase markers dominating nonshared material continue to reside on different planes. As detailed in Milward (1994), these approaches are well-equipped to ensure structural parallelism, as

Coordination 191

in (22), but run into problems once NCC involves different numbers of adjuncts within each conjunct, as in the example (36) below. (36) You can call me [directly] or [[after 3 P.M.] [through my secretary]].

To summarize, approaches to NCC generally agree that what we are dealing with is not really coordination of arbitrary nonconstituents, but rather a special case of constituent coordination that either involves coordination at some higher structural level (VP or S in LFG and HPSG), or a more abstract notion of constituency (CG). The mere existence of NCC also has repercussions on the usefulness of coordination data as a constituency test: given that apparent nonconstituents can be combined by means of some coordinating conjunction, mere conjoinability of two syntactic units will not tell us anything about their status. The converse, however, is not true: failure to coordinate can still be regarded as an indicator of nonconstituency (or dissimilarity). Gapping

In contrast to peripheral sharing, gapping constructions, where the shared material is central, not peripheral, are less well studied in Phrase Structure Grammar. As mentioned above, Maxwell and Manning (1996) argue to exclude it from the domain of syntax altogether. In the framework of HPSG,

gapping phenomena have not received any detailed attention either, although it is highly conceivable to extend existing approaches to peripheral sharing in such a way as to permit sharing of head domain objects. In Categorial Grammar (Steedman, 1990), gapping phenomena, just like peripheral sharing, is treated on a par with constituent coordination, building crucially on an extended notion of constituency, including ‘nonstandard constituents’ (Steedman, 1990) which result from the application of type raising and functional composition. (37) Harry will buy bread, and Barry, potatoes. (Steedman, 1990)

In Steedman’s account, the constituents of the gapped conjunct (Barry potatoes) are type raised to a sentence lacking the verb (¼S/((S/NP)/NP)). The nongapped conjunct of category S, however, is decomposed by a special rule (the Left Conjunct Revealing Rule) into the verbal functor ((S/NP)/NP) and the remainder, again a sentence lacking the verb (¼S/((S/NP)/NP)). Since these two categories are identical, they can coordinate to yield a constituent of the same category, which then recombines with the verb to produce a sentence. The theory of gapping as advanced by Steedman (1990) also establishes a link between the direction of functional application and the linear order of gapped vs. nongapped conjuncts. As observed as early as

192 Coordination

Ross (1970), there is a strong correlation in the languages of the world between basic word order (SVO/ VSO vs. SOV) and the position of the gapped conjunct: in consistently head-initial languages such as English (SVO) or Irish (VSO), the gapped conjunct is noninitial, just like (most) arguments are, whereas head-final languages, such as Japanese (SOV), have the gapped conjunct precede the nongapped one. (38) [watakusi wa sakana o] [Biru wa I NOM fish ACC Bill NOM gohan o] tabeta rice ACC ate ‘I ate fish, and Bill, rice’ (Ross, 1970: 251)

Note, however, that gapping in strict SOV languages is not easy to distinguish from RNR. Languages such as German, which is somewhat intermediate between a head-final and a head-initial language, do not seem to behave in typologically clear-cut ways: in subordinate clauses, where word order is strictly verb-final, the gapped conjunct may either precede the nongapped one (39a) or vice-versa (39b). In verb-second clauses, however, the gapped conjunct must follow the nongapped one, just like in English. (39a) Ich vermute daß Peter Wein, und I presume that Peter wine and Marie Bier getrunken hat. Marie beer drunk has ‘I presume that Peter drank wine and Marie beer.’ (39b) Ich vermute daß Peter Wein getrunken I presume that Peter wine drunk hat, und Marie Bier. has and Marie beer ‘I presume that Peter drank wine and Marie beer.’ (39c) Peter trank Wein, und Marie Bier. Peter drank wine and Marie beer ‘Peter drank wine, and Marie beer.’ (39d) *Peter Wein, und Marie trank Bier. Peter wine and Marie drank beer

Since (39a) may as well be interpreted as RNR, German appears to pattern with head-initial languages, despite its verb-final order. SGF Coordination

Another NCC phenomenon that cannot be reduced to peripheral sharing is so-called SGF coordination (Subject Gaps in Finite/Frontal clauses), a term originally from Ho¨ hle (1983) and Wunderlich (1988). In this construction, which is typically found in languages such as German or Dutch, a subject internal to the first conjunct gets elided in the second. In English,

a similar construction exists, involving Locative Inversion (40c). (40a) [In den Wald ging der Ja¨ ger] into the woods went the hunter und [fing einen Hasen]. and caught a rabbit ‘Into the woods went the hunter and caught a rabbit.’ (40b) [Wenn du in ein Kaufhaus if you into a department store gehst] und [hast kein Geld], ... go and have no money then ‘If you go to a department store and you do not have any money, . . .’ (40c) [Into the room strode Robin] and [sat down on a chair].

The main challenge of SGF coordination is similar to that posed by gapping, i.e., that the shared subject is properly contained within one of the conjuncts. It is a distinguishing property of this construction that the second conjunct must be a verb-first (V1) structure, whereas the first conjunct may either be verb-second (V2, see [40a]), verb-first, or a complementizerintroduced verb-final structure; see (40b). The order of the two conjuncts is fixed. Furthermore, it is generally agreed that SGF coordination exhibits a strong subject–object asymmetry: neither can an object be elided in the second conjunct, nor can a nonsubject in the first conjunct be interpreted as coreferent with the missing subject. Principle and Parameter approaches to the phenomenon typically invoke asymmetrical syntactic structures to model the properties of the construction (Ho¨ hle, personal communication; Heycock and Kroch, 1993; Wunderlich, 1988). See Kathol (1999) for a detailed synopsis. On the basis of data from Dutch, Steedman (1990) illustrates how his Categorial Grammar account of gapping can straightforwardly derive SGF coordination as well. His approach, however, has been criticized by Kathol (1995) for overgeneration, being able to equally derive sharing of conjunct-internal objects. Kathol (1995, 1999) suggests instead of reducing SGF to coordination at the VP level, and derives the differences in word order (V2 vs. V1) by means of word order constraints operating on complex order domains. Frank (2002) presents an LFG analysis of the construction that crucially exploits LFG’s notion of grammaticalized discourse functions to account for the subject–object asymmetry and the word order effects in a symmetrical way.

Coordination of Unlikes In our discussion of constituent coordination, we have made the rather common assumption that

Coordination 193

conjuncts need to be identical to a certain degree in order to undergo coordination. Inter alia, we have assumed identity of syntactic category as a prerequisite for conjoinability. Work in GPSG (Gazdar et al., 1985; Sag et al., 1985), however, has devoted quite a lot of attention to examples where this fairly basic requirement appears to be violated: (41) Kim is [[a Republican]NP[PRD þ] and [proud of it]AP[PRD þ]].

The solution GPSG provided for what is now known as ‘coordination of unlikes’ involved the assumption that the featural specification of the mother node in a coordinate structure should be the intersection (or generalization) of the feature specification of the conjunct daughters. As shown, however, by Jacobson (1987), data such as (42) illustrate the systematic limitations of a generalization-based approach, like GPSGs: (42a) (42b) (42c) (42d) (42e)

Kim grew wealthy. *Kim grew a Republican. Kim grew and remained wealthy. *Kim grew and remained a Republican. *Kim grew and remained wealthy and a Republican.

If grow only subcategorizes for APs [N þ, Vþ], and remain subcategorizes for APs or NPs ([N þ]), coordination of the two verbs should yield the intersection of the two feature specifications, namely [N þ]. But this weakened feature specification cannot account for the ungrammaticality of (42e), where the coordinated phrase [wealthy and a Republican] would carry exactly this feature specification. Bayer and Johnson (1995) propose a solution to the above problem within the framework of Lambek’s categorial grammar where functors correspond to logical implication. Thus, a functor ‘‘a/b can only combine with an argument b0 to form if b0 is logically stronger than b’’ (Bayer and Johnson, 1995: 70), or in other words, if the argument is subsumed by the antecedent of the functor category. As a result, the category of remained, which is VP/(NP _ AP), can be strengthened to VP/AP, which allows it to be coordinated with grew. By the same token, an NP or AP can be weakened to NP _ AP, which permits coordination of unlike categories, as in (41). Sentences like (42e) are then ruled out for the simple reason that the category NP _ AP of the coordinate structure is logically weaker than the antecedent of the conjoined functor category VP/AP. Bayer and Johnson (1995) relate the success of their approach to the fact that subsumption operates on predicate– argument relations, in contrast to unification-based approaches, where it is associated with coordination.

An analysis of Coordination of Unlikes that is highly similar to this approach has recently been proposed by Sag (2003) for HPSG. Work in unification-based frameworks such as LFG (Maxwell and Manning, 1996) or HPSG (Crysmann, in press; Beavers and Sag, 2004) has provided an alternative approach to the puzzle of coordination of unlikes, by way of assimilating it to NCC. Instead of constituent coordination of unlike categories, they propose to analyze these phenomena as left-peripheral sharing, at the level of VP or S: thus, a coordination of unlikes as in (41) is well formed, just because the shared verb can combine, individually, with an AP or NP constituent in each of the VP or S conjuncts. Feature Neutrality

An issue related to Coordination of Unlikes is feature neutrality, or feature indeterminacy, as witnessed by the following example from German (Pullum and Zwicky, 1986): (43a) Er [findet[OBJ acc] und hilft[OBJ dat]] he finds and helps ‘He finds and helps women.’ (43b) *Er [findet[OBJ acc] und hilft[OBJ dat] he finds and helps (43c) *Er [findet[OBJ acc] und hilft[OBJ dat] he finds and helps

Frauenacc/dat. women Kinderndat. children Ma¨ nneracc. men

The two verbs finden ‘find’ and helfen ‘help’ impose different restrictions on the case value of their respective objects. Still, they can coordinate at the word (or X0) level and share an object, if this object is morphologically neutral between the conflicting case requirements. Since Zaenen and Karttunnen (1984), it has been established that feature neutrality cannot be reduced to ambiguity in unification-based frameworks: if we assume that Frauen has a case specification acc _ dat, unification with the selectional requirements of finden will disambiguate it to acc, so it would no longer be compatible with the requirement imposed by helfen, and vice versa. It has been argued by Ingria (1990), that feature indeterminacy, besides Coordination of Unlikes, therefore poses a major challenge for unification-based grammar. Note also that a generalization-based approach to the coordination of finden and helfen will run into exactly the same problems as the GPSG account of Coordination of Unlikes discussed above, equally permitting the ungrammatical (43b) and (43c). Bayer and Johnson’s (1995) approach to feature indeterminacy is entirely analogous to their treatment of Coordination of Unlikes: at the heart of their account lies a principled distinction between neutral

194 Coordination

or overspecified categories, like np ^ acc ^ dat, assigned to the neutral Frauen and underspecified categories, like np ^ (acc _ dat) that they take to be the category of a coordination of case-distinct NPs, such as Ma¨ nneracc und Kinderndat. Functor categories like vp/np ^ acc or findet or vp/np ^ dat for hilft can be strengthened to vp/np ^ acc ^ dat, which can then be coordinated into a category of the same type. While neutral NPs like Frauen are exactly of this overspecified type, and hence can function as an argument to yield a VP, non-neutral NPs like Ma¨ nner or Kindern cannot, since their respective categories (np ^ acc or np ^ dat) are not subsumed by np ^ acc ^ dat. Within the context of unification-based grammar, solutions similar to Bayer and Johnson (1995) have recently been developed for LFG (Dalrymple and Kaplan, 2000) and HPSG (Daniels, 2001; Levy and Pollard, 2001). Dalrymple and Kaplan (2000) suggest replacing equality-based checking of agreement features with a setbased approach: non-neutral NPs such as Ma¨ nner or Kindern will be specified as ("CASE) ¼ {ACC} or ("CASE) ¼ {DAT}, respectively, whereas neutral NPs such as Frauen will specify both cases as elements of their CASE set ("CASE) ¼ {ACC,DAT}. Case checking by verbs will then be expressed as constraints on set membership, e.g., ACC 2 ("OBJ CASE) for finden, or DAT 2 ("OBJ CASE) for helfen. The f-structure of the V0 coordination will simply contain the conjunction (or unification) of these set constraints, requiring the direct object’s CASE set to contain both an ACC and a DAT element. Within HPSG, Daniels (2001) has shown how the basic insight of the Bayer and Johnson approach to feature neutrality can be incorporated by refining the type hierarchies of case values (see also Levy and Pollard, 2001 for technically different, though conceptually identical approaches).

Building on an earlier proposal by Levine et al. (2001), he suggests distinguishing traditional case requirements such as acc or dat from the maximal (or ‘pure’) cases to which they may resolve, enabling him to include a neutral type p-acc-dat alongside the non-neutral types p-acc and p-dat. While the case values of Ma¨ nner and Kindern will be the nonneutral types p-acc and p-dat, Frauen will be lexically specified as [CASE p-acc-dat]. This augmentation of the type hierarchy makes it possible to reduce feature neutrality to standard notions of unification: since

finden will subcategorize for acc and helfen for dat, the unification of these requirements can only be satisfied by the case-neutral type p-acc-dat. A drawback of a type-based approach such as Daniels’s is that the hierarchies needed to provide all necessary neutralizing types may grow factorial to the number of case distinctions. An alternative has been proposed by Sag (2003), who advocates subsumption checks, thereby likening the HPSG approach to that of Bayer and Johnson (1995) and Dalrymple and Kaplan (2000). To conclude, current solutions to the puzzle of Coordination of Unlikes unanimously agree that likeness of constituents should still be regarded a defining property of coordinate structures, but that the definition of likeness is subject to further refinement. Solutions differ, though, as to whether resolution is sought at the level of the ‘unlike’ constituents themselves, using subsumption checks, or else at some higher level, likening the treatment of Coordination of Unlikes to that of NCC. See also: Combinatory Categorial Grammar; Constituent

Structure; Declarative Models of Syntax; Dependency Grammar; Ellipsis; Head-Driven Phrase Structure Grammar; Island Constraints; Lexical Functional Grammar; Long-Distance Dependencies; Right Node Raising; X-Bar Theory.

Bibliography Bayer S & Johnson M (1995). ‘Features and agreement.’ In Proceedings of the 33rd Annual Meeting of the ACL. 70–76. Beavers J & Sag I A (2004). ‘Coordinate ellipsis and apparent non-constituent coordination.’ In Mu¨ ller S (ed.) Proceedings of the 11th International Conference on HPSG, Aug 3–6, 2004, Leuven. Stanford: CSLICSLI Online Proceedings. Chomsky N (1957). Syntactic structures. Den Haag: Mouton. Chomsky N A (1982). Linguistic Inquiry Monographs: Some concepts and consequences of the theory of Government and Binding (vol. 6). Cambridge MIT Press. Crysmann B (2003). Constraint-based coanalysis. Portuguese cliticisati on and morphology–syntax interaction in HPSG. Saarbru¨ cken Dissertations in Computational Linguistics and Language Technology, No. 15. Saarbru¨ cken: Computational Linguistics, Saarland University and DFKI LT Lab. Crysmann B (in press). ‘An asymmetric theory of peripheral sharing in HPSG: conjunction reduction and coordination of unlikes.’ In Jaeger G, Monachesi P, Penn G, & Wintner S (eds.) Proceedings of FGVienna: The 8th Conference on Formal Grammar, Aug 16–17 2003, Vienna. Stanford: CSLI Publications.

Coordination 195 Dalrymple M & Kaplan R (2000). ‘Feature indeterminacy and feature resolution.’ Language 76(4), 759–798. Dalrymple M, Shieber S & Pereira F (1992). ‘Ellipsis and higher-order unification.’ Linguistics and Philosophy 14, 399–452. Daniels M (2001). ‘On a type-based analysis of feature neutrality and the coordination of unlikes.’ In Proceedings of the 8th International Conference on Head-Driven Phrase Structure Grammar. CSLI Online Proceedings. Stanford: CSLI Publications. 137–147. Davis A (1993). ‘Empty heads and missing subjects: underspecification in Hausa VPs.’ In Proceedings of the Chicago Linguistics Society 28. Dixon R (1972). The Dyirbal language of North Queensland. Cambridge: Cambridge University Press. Dowty D (1988). ‘Type raising, functional composition, and non-constituent conjunction.’ In Oehrle R et al. (eds.) Categorial grammars and natural language structures. Dordrecht: Reidel. Frank A (2002). ‘A (discourse) functional analysis of asymmetric coordination.’ In Butt M & Holloway King T (eds.) Proceedings of the LFG 2002 Conference. Athens: CSLI Online Publications. 174–196. Gazdar G, Klein E, Pullum G K & Sag I (1985). Generalized Phrase Structure Grammar. Oxford: Basil Blackwell. Goodall G (1987). Parallel structures in syntax: coordination, causatives and restructuring. Cambridge: Cambridge University Press. Grosu A (1973). ‘On the nonunitary nature of the coordinate structure.’ Linguistic Inquiry 4, 88–92. Grover C (1994). The encyclopedia of language and linguistics (vol. 2). Chapter Coordination. Elsevier. 762–768. Heycock C & Kroch A (1993). ‘Verb movement and the status of subjects: implications for the theory of licensing.’ The Linguistic Review 11, 257–283. Ho¨ hle T (1990). ‘Assumptions about asymmetric coordination in German.’ In Mascaro´ J & Nespor M (eds.) Grammar in progress. Dordrecht: Foris. 221–235. Ingria R J P (1990). ‘The limits of unification.’ Proceedings of the 28th Annual Meeting of the (ACL). 194–204. Jackendoff R (1977). X-Syntax. Cambridge: MIT Press. Jacobson P (1987). ‘Review of Gerald Gazdar, Ewan Klein, Geoffrey Pullum, and Ivan A. Sag (1985): Generalised Phrase Structure Grammar.’ Linguistics and Philosophy. 10, 389–426. Kaplan R M & Maxwell J T (1988). ‘Constituent coordination in Lexical-Functional Grammar.’ In Proceedings of COLING 88. COLING 88, Budapest. 303–305. Kathol A (1995). Linearization-based German syntax. Ph.D. diss., Ohio State University. Kathol A (1999). ‘Linearization versus phrase structure in German coordinate constructions.’ Cognitive Linguistics 10, 303–342. Lakoff G (1986). ‘Frame semantic control of the coordinate structure constraint.’ In Farley A, Farley P & McCullogh K E (eds.) Papers from the parasession on pragmatics and grammatical theory at the 22nd Regional Meeting, Chicago Linguistic Society. 152–167.

Levy R & Pollard C (2001). ‘Coordination and neutralization in HPSG.’ In Proceedings of the 8th International Conference on Head-Driven Phrase Structure Grammar. CSLI Online Proceedings. Stanford: CSLI Publications. 221–234. Levine R, Hukari T & Calcagno M (2001). ‘Prasitic gaps in English: some overlooked cases and their theoretical implications.’ In Culicover P & Postal P (eds.) Parasitic gaps. Cambridge: MIT Press. 181–222. Maxwell J T & Manning C (1996). ‘A theory of nonconstituent coordination based on finite-state rules.’ In Butt M & Holloway King T (eds.) Proceedings of the LFG96 Conference, Rank Xerox, Grenoble. CSLI Online Proceedings, Stanford: CSLI Publications. Milward D (1994). ‘Non-constituent coordination: theory and practice.’ In Proceedings of the Fifteenth International Conference on Computational Linguistics (COLING94). 935–941. Moltmann F (1992). ‘Coordination and comparatives.’ Ph.D. diss., MIT. Pollard C & Sag I (1994). Head–Driven Phrase Structure Grammar. Stanford: CSLI and University of Chicago Press. Pullum G & Zwicky A (1986). ‘Phonological resolution of syntactic feature conflict.’ Language 62, 751–773. Reape M (1994). ‘Domain union and word order variation in German.’ In Nerbonne J, Netter K & Pollard C (eds.) German in Head-Driven Phrase Structure Grammar. Lecture Notes, No. 46. Stanford: CSLI Publications. 151–197. Roberts I (1997). Comparative syntax. New York: Edward Arnold. Ross J R (1967). ‘Constraints on variables in syntax.’ Ph.D. diss., MIT. Ross J R (1970). ‘Gapping and the order of constituents.’ In Bierwisch M & Heidolph K E (eds.) Progress in linguistics. The Hague: Mouton. Sag I, Gazdar G, Wasow T & Weisler S (1985). ‘Coordination and how to distinguish categories.’ Natural Language and Linguistic Theory 3, 117–171. Sag I A (2003). ‘Coordination and underspecification.’ In Kim J-B & Wechsler S (eds.) Proceedings of the 9th International Conference on Head-driven Phrase Structure Grammar, Kyung Hee University, Seoul, 5–7 August, 2002. Stanford: CSLI Publications. 267–291. Steedman M (1985). ‘Dependency and coordination in the grammar of Dutch and English.’ Language 61, 523–568. Steedman M (1990). ‘Gapping as constituent coordination.’ Linguistics and Philosophy 13, 207–264. Whitman N (2002). ‘Category neutrality: a type-logical investigation.’ Ph.D. diss., Ohio State University. Wunderlich D (1988). ‘Some problems of coordination in German.’ In Reyle U & Rohrer C (eds.) Natural language processing and linguistic theories. Dordrecht: Reidel. Yatabe S (2000). ‘The syntax and semantics of left-node raising in Japanese.’ In Proceedings of the 7th International Conference on Head-Driven Phrase Structure Grammar. CSLI Online Proceedings. Stanford: CSLI Publications. 325–344.

196 Coordination Yatabe S (2003). ‘A linearization-based theory of summative agreement in peripheral-node raising constructions.’ In Kim J-B & Wechsler S (eds.) Proceedings of the 9th International Conference on Head-driven Phrase Structure Grammar, Kyung Hee University, Seoul, 5–7 August, 2002. Stanford: CSLI Publications. 391–411.

Zaenen A & Karttunnen L (1984). ‘Morphological nondistinctiveness and coordination.’ In Proceedings of the First Eastern States Conference on Linguistics (ESCOL). 309–320.

Copenhagen School J Rischel, Birkerød, Denmark ß 2006 Elsevier Ltd. All rights reserved.

The general understanding of the term ‘the Copenhagen School’ is that it refers to the scholars, mostly Danes, who – together with Louis Hjelmslev and Hans-Jørgen Uldall – can be identified with ‘glossematics,’ one of the leading trends in European structuralism from the early 1930s until the early 1950s. Their forum of discussion was the ‘Linguistic Circle of Copenhagen,’ which was founded in 1931 (inspired by the Cercle Linguistique de Prague). The theory that eventually came to be called ‘glossematics’ was developed by Hjelmslev and Uldall in the second half of the 1930s. Their Outline of Glossematics monograph was due to appear in 1937, but never appeared in full, and a comprehensive formal statement appeared only posthumously. Therefore, the theory became influential mainly through its basic claims. A conspicuous feature of glossematic theory is its recognition of a full duality of linguistic structure: there is a ‘content form’ as well as an ‘expression form.’ The claim is that these forms are organized in similar ways, so that the categories of content and of expression can be defined in terms of the same kinds of relations. This was a further development of the Saussurean view of language as a semiotic comprising a signifiant and a signifie´ . Glossematics was intended as a definitive and complete theory of language. Accordingly, Hjelmslev’s contemporaries in the 1940s and 1950s made more or less successful attempts to apply it to descriptive linguistic work. The most extensive contribution was made by Danish dialectologists such as Børge Andersen, Anders Bjerrum, Poul Andersen, and Inger Ejskjær, who mostly concentrated on the expression side of language. Danish theoretical linguistics around the World War II should, however, not be seen as a ‘school’ defined by wholesale acceptance of a particular approach. Rather, the milieu was characterized by lively debates on a high intellectual level.

Eli Fischer-Jørgensen, then a very young member of the group, had always taken a critical attitude towards Hjelmslev’s theories, but with her expert knowledge she has been instrumental in getting much of the glossematic work of that period published, often posthumously. She wrote the best presentations of glossematic theory to date. Knud Togeby was strongly inspired by Hjelmslev but did not follow his ideas strictly. Paul Diderichsen was even more eclectic; he made an original and lasting contribution by inventing a position-based scheme for syntactic analysis, which for half a century has played a role in Nordic grammar work. Some other very prominent scholars worked entirely outside the Hjelmslevian framework. Otto Jespersen had developed his own grammatical approach, and Viggo Brøndal established himself as an important figure within several fields of theoretical linguistics. Danish linguists of later generations have seen the glossematic legacy mainly as an inspiration for distinguishing clearly between form and substance, and for basing syntactic analysis explicitly on meaning and function. The Copenhagen version of Functional Grammar, an offshoot of Simon Dik’s theory – with inspiration both from glossematics and cognitive grammar – is perhaps the most distinctive trend in contemporary Danish linguistics. Although the glossematic framework of linguistic analysis has become obsolete, Hjelmslev’s ideas about semiotics continue to attract the attention of linguists and philosophers worldwide, notably in Italy and in Japan. See also: Fischer-Jørgensen, Eli (b. 1911); Glossematics;

Hjelmslev, Louis Trolle (1899–1965); Uldall, Hans-Jørgen (1907–1957).

Bibliography Fischer-Jørgensen E (1997). ‘Hjelmslev et le Cercle Linguistique de Copenhague.’ In Hjelmslev aujourd’hui (¼ Semiotic and Cognitive Studies, 5). Turnhout: Brepols. 27–36. Rischel J (2001). ‘The Cercle linguistique de Copenhague and Glossematics.’ In History of the Language Sciences. 2.2. Berlin//New York: Walter de Gruyter. 1790–1806.