1   Coordination

Martin Haspelmath

0      Introduction

The term coordination refers to syntactic constructions in which two or more units of the same type are combined into a larger unit and still have the same semantic relations with other surrounding elements. The units may be words (e.g. verbs (1a)), phrases (e.g. noun phrases (1b)), subordinate clauses (e.g. (1c)) or full sentences (e.g. (1d)).

All languages appear to possess coordination constructions (or coordinate constructions) of some kind, but there is a lot of cross-linguistic variation. Individual languages may possess a wealth of different coordinate constructions that relate to each other in complex ways. It is the purpose of this chapter to introduce and discuss a wide range of conceptual distinctions that are useful for describing the cross-linguistic and language-internal variation. This entails the use of a large number of technical terms (printed in boldface on first occurrence), each of which is explained and illustrated as it is introduced. Terminological issues are discussed further in an appendix.

   The particle or affix that serves to link the units of a coordinate construction is called the coordinator. In (1) and in the other numbered examples in this chapter, the coordinator is printed in boldface. By far the most frequently occurring coordinator is ‘and’ (i.e. English and and its equivalents in other languages), but coordinate constructions can also involve various other semantic types of linkers, such as ‘or’, ‘but’ and ‘for’. ‘And’-coordination is also called conjunctive coordination (or conjunction), ‘or’-coordination is also called disjunctive coordination (or disjunction), ‘but’-coordination is called adversative coordination, and ‘for’-coordination is called causal coordination. Examples of each of these four types are given in (2).

   The units combined in a conjunctive coordination are called conjuncts, and, more generally, the units of any coordination will be called coordinands here. Adversative coordination is always binary, i.e. it must consist of two coordinands. Ternary or other multiple coordinations are impossible here. This is illustrated in (3).

By contrast, conjunctions and disjunctions can consist of an indefinite number of coordinands. The examples in (4) show six coordinands each.

   Languages differ with respect to the number and the position of the coordinators used in coordinate constructions. For instance, while English generally shows the pattern A co-B (where co stands for coordinator), Kannada (a Dravidian language of southern India) shows the pattern A-co B-co:

The patterns of coordinator placement and the types of linkers are discussed further in Section 1.

   Many languages have several alternative patterns for a given semantic type of coordination, as illustrated in the English examples (6a,b). Coordination with the two-part coordinator both . . . and describes the coordinands as contrasting in some way: (6a) is appropriate, for instance, if the hearer expects only one of them to make the trip. This construction will be called emphatic coordination in this chapter.

Moreover, many languages have special coordinators for negative contexts, as in the English example (7a). This sentence is roughly equivalent semantically with (7b), but again it has a more emphatic flavour. The construction in (7a) will be called emphatic negative coordination.

Emphatic and negative coordinate constructions are discussed further in Section.

   We saw in (1) above that a coordinate construction can consist of different types of coordinands: words, phrases, clauses or sentences. But as the definition of coordination says, each coordinand must be of the same type within a coordinate construction. Thus, (8b) and (9b) are ungrammatical, because the coordinands are syntactically different (NP vs PP in 8b) or at least semantically different (manner vs comitative in 9b).

Different languages may require different coordinators depending on the syntactic type of the coordinands. For example, Yapese (an Austronesian language of Micronesia) has ngea ‘and’ for NP conjunction (10a), but ma ‘and’ for sentential conjunction (10b) (Jensen (1977:311–12)):

‘I will go to Donguch, and Tamag will go to Nimgil’

Types of coordinands and their relevance for the structure of coordination are discussed further in Section.

   In addition to the major semantic distinctions that we saw in (2), numerous more fine-grained distinctions can be made. For example, many languages distinguish between two types of disjunction: interrogative disjunction and standard disjunction. Mandarin Chinese uses two different coordinators for these two cases, háishi and huòzhe (both translate as ‘or’) (Li and Thompson (1981:654)):

More fine-grained semantic distinctions such as these are discussed further in Section.

   Next, I discuss some special types of conjunction. Since conjunction is the most frequent kind of coordination, it exhibits the greatest formal diversity, and some of these patterns are examined in Section 5. The most prominent ‘special type’ of conjunction involves the use of a comitative marker (i.e. a marker expressing accompaniment), as in Hausa, where da means both ‘with’ (12a) and ‘and’ (12b) (Schwartz (1989:32, 36)):

   In addition to coordinations in which each coordinand is a regular syntactic constituent (e.g. an NP, a VP, or a clause), many languages allow non-constituent coordination, as illustrated in (13). For the sake of clarity, the coordinands are enclosed in square brackets in these examples.

In (13a) and (13b), the first coordinand is an ordinary constituent (a sentence and a VP, respectively), but the second coordinand is not. In (13c), only the second coordinand is an ordinary constituent. In order to assimilate non-constituent coordinations to patterns found elsewhere in the grammar, linguists have often described them in terms of ellipsis (or coordination reduction). That is, abstract underlying structures such as those in (14a–c) are posited, which show ordinary constituent coordination. In a second step, a rule of ellipsis of identical elements deletes the words underlined in (14), resulting in the surface patterns in (13).

Non-constituent coordination and ellipsis are discussed further in Section 6.

   Finally, in Section 7 I discuss ways of distinguishing coordination from less grammaticalized constructions and, perhaps most importantly, from subordination and dependency. The latter two notions will be discussed briefly here. The primary contrast is that between coordination and dependency. In a coordination structure of the type A(-link-)B, A and B are structurally symmetrical in some sense, whereas in a dependency structure of the type X(-link-)Y, X and Y are not symmetrical, but either X or Y is the head and the other element is a dependent. When the dependent element is a clause, it is called subordinate clause.

   Although the distinction between coordination and dependency is, of course, fundamental, it is sometimes not evident whether a construction exhibits a coordination relation or a dependency relation. The best-known distinctive property of coordinate structures is that they obey the coordinate structure constraint (J. R. Ross (1986)), which prohibits the application of certain rules, such as extraction of interrogative words from coordinate structures. This is illustrated in (15–16), where the (ⅰ) sentences show the basic structure, and the (ⅱ) sentences show fronting of who. As the examples make clear, only the dependency structures allow extraction (15a(ii) and 16a(ii)).¹

(15)
a.	dependency (subordination)
	(ⅰ) (basic sentence)	You talked to someone before Joan arrived
	(ⅱ) (who extraction)	Who did you talk to _ before Joan arrived?
b.	coordination
	(ⅰ) (basic sentence)	You talked to someone and then Joan arrived
	(ⅱ) (who extraction)	∗Who did you talk to _ and then Joan arrived?
(16)
a.	dependency
	(ⅰ) (basic sentence)	You saw Marvin with someone
	(ⅱ) (who extraction)	Who did you see Marvin with _?
b.	coordination
	(ⅰ) (basic sentence)	You saw Marvin and someone
	(ⅱ) (who extraction)	∗Who did you see Marvin and _?

Obeying the coordinate structure constraint is a formal property of constructions that is sometimes taken as the decisive criterion for coordinate status. In this chapter, by contrast, I will work with a primarily semantic definition of coordination, as given at the beginning of this section. The reason for this is that only semantically based notions can be applied cross-linguistically – formal criteria are generally too language-particular (for instance, not all languages have extraction constructions that would show the effect of the coordinate structure constraint).

Types and positions of coordinators

Coordinate constructions may lack an overt coordinator (asyndetic coordination) or have some overt linking device (syndetic coordination). So far in this chapter, all examples have shown syndetic coordination. If we restrict ourselves for the moment to binary coordinations, syndetic coordinations may have either a single coordinator (monosyndetic) or two coordinators (bisyndetic). Monosyndetic coordination is illustrated by Franz and Sisi (cf. (6b)), and bisyndetic coordination is illustrated by both Franz and Sisi (cf. (6a)). Coordinators may be prepositive (preceding the coordinand) or postpositive (following the coordinand). In English, all coordinators are prepositive, but we saw an example of the postpositive coordinator -u: in Kannada earlier (example (5)).

   The logical possibilities for binary coordination are shown schematically in (17) (the two coordinands are represented as A and B, and the coordinator is represented as co).

As we will see below, with one exception (co-A B), all these possibilities occur in languages. However, not all of them are equally common.

Asyndetic coordination

Coordination without an overt linker occurs widely in the world’s languages, and, although in European languages monosyndesis of the type A co-B is the norm, asyndesis (also called juxtaposition) also occurs commonly, especially with the meaning of conjunction:

In European languages, asyndesis occurs mostly with modifying phrases such as adverbials and adjectives, or with clauses. Asyndetic coordination of NPs is more restricted and quite impossible in many cases (cf. ??I met Niko, Sandra ‘I met Niko and Sandra’). Many non-European languages have no such restrictions, and asyndetic coordination is very wide-spread in the world’s languages. The following examples are from Sarcee (an Athapaskan language of Alberta, Canada), Maricopa (a Yuman language of Arizona), and Kayardild (a Tangkic language of northern Australia).

In asyndesis, intonation is the only means by which the coordinated structure can be indicated, and it is probably not an accident that languages with a long written tradition tend to have a strong preference for syndesis: intonation is not visible in writing (see Mithun (1988)). Languages that lack writing (or lacked it until recently) often lack indigenous coordinators and now use coordinators borrowed from prestige languages such as Spanish, English, Arabic, and Russian. Asyndesis is often preferred in natural conjunction, i.e. when the two conjuncts habitually go together and form some kind of conceptual unit (see Section below).

Monosyndetic coordination

There are three occurring patterns of monosyndetic coordination: A co-B, A-co B, and A B-co, which are illustrated in (20–22). The logically possible type co-A B is unattested (this fact will be explained below).

   The two types A co-B (medial prepositive) and A-co B (medial postpositive) can be distinguished on the basis of evidence for different constituency divisions: [A] [co B] vs [A co] [B]. Relevant constituency tests include:

   Intonation: in certain cases, English and forms an intonation group with the following phrase, not with the preceding phrase (Joan, and Marvin, and their baby, not ∗Joan and, Marvin and, their baby; here commas represent intonation breaks). Of course, this test does not apply in the simplest cases: a construction such as Joan and Marvin forms a single intonation group.

   Pauses: in English, it is much more natural to pause before and (Joan . . . and Marvin) than after and (??Joan and . . . Marvin).

   Discontinuous order: in special circumstances, the coordinands may be separated by other material, as when a coordinand is added as an afterthought. In English, the coordinator must be next to the second coordinand (e.g. My uncle will come tomorrow, or my aunt, not ∗My uncle or will come tomorrow, my aunt).

   (Morpho)phonological alternations: when the coordinator or one of the coordinands undergoes (morpho)phonological alternations in the construction, this is evidence that they form a constituent together. For instance, in Biblical Hebrew the coordinator wə ‘and’ has the alternant ū when the first syllable of the following phrase has a schwa vowel (e.g. wə-æræx ‘and (a) way’, ū-ðərx-īm ‘and ways’). In Latin, the element -que and the preceding conjunct form a single domain for stress assignment (e.g. pópulus ‘the people’, populús-que ‘and the people’).

   In principle, one could imagine cases in which none of these criteria yields a clear asymmetry, so that one would have a symmetrical pattern A-co-B in addition to prepositive A co-B and postpositive A-co B. But no case of a language that requires such an analysis has come to my attention. Monosyndetic coordination seems to be universally asymmetric.

   When the coordinator is linked by phonological processes to its coordinand (see (iv) above), it is generally regarded as a clitic or affix rather than an independent word. (Criteria for clitic or affix status are largely language-particular and cannot be discussed further here.)² Due to the universal preference for suffixation over prefixation, postpositive coordinators are typically suffixed and thus written as one word with the coordinand to which they are attached. Prepositive coordinators, by contrast, are rarely prefixed and written together with the coordinand. Thus, when a language has a coordinate construction of the form A co B, where co is not an affix on A or B, it is likely that constituency tests will show co to be a prepositive coordinator, like English and.

   Postpositive coordinators may follow the complete phrase, or they may en- clitically follow the first word of the coordinand. The latter is illustrated by Turkish postpositive de in (23).

   As is noted in Stassen (2000), the order of the coordinator correlates with other word order patterns of the language, in particular verb–argument order: languages with a postpositive coordinator (such as Latin and Classical Tibetan) tend to have verb-final word order, whereas verb-initial languages tend to have a prepositive coordinator. However, Stassen’s generalizations are based exclusively on conjunctive coordinators. Disjunctive coordinators may conform to different ordering patterns. For instance, Kanuri (a verb-final Nilo-Saharan language of northern Nigeria) has (bisyndetic) postpositive conjunctive coordinators (-a . . . -a, see (24a)), but a (monosyndetic) prepositive disjunctive coordinator râ ((see 24b)). A similar asymmetry is found, for instance, in Lezgian (a verb-final Daghestanian language; Haspelmath (1993:327, 331)).

   As I noted above, the pattern co-A B is unattested and seems to be non-existent, at least for conjunction (see Stassen (2000), who examined a sample of 260 languages). This generalization can be explained diachronically if the two main diachronic sources of conjunction constructions are (ⅰ) a comitative modifying construction of the type ‘A with B’ (see Section ), and (ⅱ) a construction with an additive focus particle of the type ‘A, also B’. An example of a comitative-derived construction is Lango cây kèdè càk ‘tea and milk’ (cf. (20)), which comes from a dependency construction in which kèdè càk is a modifier meaning ‘with milk’ (in fact, the phrase can still have this meaning; Noonan (1992:163)). Since languages with modifier–noun order tend to have postpositions and languages with noun–modifier order tend to have prepositions (cf. Greenberg (1963); Dryer (1992)), the patterns A-co B and A co-B are the most expected ones from the comitative source. The focus-particle source of conjunction always has the marker on the second conjunct: ‘A, also B’, or ‘A, B too’. When the focus particle is postpositive (like too), this yields A B-co, and when the focus particle is prepositive (like also), this yields A co-B. There is thus apparently no common diachronic source for the pattern co-A B, whose non-existence or extreme rarity is thereby explained.

Bisyndetic coordination

When there are two coordinators in the binary coordination, there are again four logically possible patterns, but in this case, all four patterns are attested (see (25–28)). However, the mixed patterns (27–28) seem to be extremely rare. In the non-mixed patterns (co-A co-B, A-co B-co), both coordinators generally have the same shape, whereas this is not the case in the mixed patterns.

   Stassen (2000) finds that, for conjunctive coordination, postpositive bisyndesis (A-co B-co) is fairly widely attested, especially in the Caucasus, northeastern Africa, Australia, New Guinea and southern India. By contrast, prepositive bisyndesis (co-A co-B) is only found as an emphatic variant of prepositive monosyndesis. Thus, besides (25), Yoruba also has the non-emphatic monosyndetic pattern èmi àti Khìndé ‘I and Kehinde’, and several European languages have similar patterns (e.g. French (et) Jean et Marie ‘(both) Jean and Marie’, Russian (ⅰ) Nina i Miša ‘Nina and Misha’) (see further Section).