It has been suggested that the coordinands must be of the same syntactic category. This would explain the contrast between the (110c) and (112c): while coordination of two PP-modifiers is possible, coordination of a postnominal PP and a relative clause is not, even though they both clearly function as modifiers of the noun phrases.

However, the examples in (113) show that categorial identity is not sufficient for coordination, since it is not possible to coordinate two postnominal modifiers of the same category if their meaning contributions differ: while the PP-modifier over Hitler specifies the subject matter of the books, the van-PP refers to their author or possessor, and coordination is usually highly questionable.

The difference in meaning contribution of the two PP-modifiers in (113) is also reflected in the fact that they can co-occur. Example (114a) first shows that the PP-modifiers in (110), which have an identical semantic relation to the modified noun, cannot co-occur as independent phrases within a single noun phrase. Example (114b), on the other hand, shows that PP-modifiers such as the over- and van-PP in (113), which have different meaning contributions, can co-occur as independent phrases within a single noun phrase; this suggests that they have different syntactic functions and that it is this fact that blocks coordination. If so, we can give a similar account of the unacceptability of (112c), in which a PP-modifier and a relative clause are coordinated, because (114c) shows that they can also co-occur as independent phrases within a single noun phrase.

The examples in (115) show that categorial identity is not only insufficient, but actually not necessary for coordination to be possible: coordination of an adjectival and an adpositional complementive is possible; and the same holds for an adjectival and a prepositional adverbial phrase.

It has been suggested that the acceptability of (115a''&b'') can be reconciled with the categorial identity requirement by assuming that we are dealing with coordination of predicative phrases (i.e. PredPs) in (115a'') and adverbial phrases (i.e. AdvPs) in (115b''). This may be undesirable, since it seems to be based on mixing the notions of syntactic category and syntactic/semantic function. Moreover, it is not necessary, since the simpler alternative in (116) presents itself (assuming that the notion of adverbial function is used to cover a set of different syntactic functions, as is claimed in Chapter V8 on independent grounds); see also Dik (1968:25), Haeseryn et al. (1997:1450), Hendriks (2001b), and many others.

Generalization (116) predicts linking of adjectival and nominal predicates also to be possible. This is indeed confirmed, although there seem to be additional restrictions that are not fully understood. For instance, the examples in (117) have been taken to show that such linking is possible only if the nominal predicate is in some sense “gradable”; cf. Goodall (1987:45).

The examples in (118) suggest, however, that the evaluative nature of the nominal predicates may also play a role: the use of the adversative coordinator maar is possible with the nominal predicate een leugenaara liar because its negative connotation contrasts with the positive property denoted by the adjective aardigkind, but impossible with the nominal predicate een steuna support because this predicate has no negative connotation.

The examples in (119) show that [+human] nouns can easily be coordinated more generally with adjectives denoting some property typically (dis)associated with them: if the property is expected for the persons denoted by the noun, the coordinator en is often followed by the adverbial dustherefore, but if the adjective denotes an unexpected property, the adversative coordinator maarbut is used.

Since we have little more to say about the factors affecting the acceptability of linking of adjectival and nominal predicates, we leave this issue for future research; for the moment, it suffices to note that the acceptable examples in (117)-(119) show that such linking is syntactically allowed.

An apt illustration of the co-occurrence restriction on coordinands in (116) is given in (120); although the verb wegento weigh can select a referential noun phrase such as de appelsthe apples in (120a) or a non-referential noun phrase such as 70 kilo ’70 kilos’ in (120b), the coordinate structure in (120c) is impossible. This is probably due to the fact that the two noun phrases have different syntactic functions. The referential noun phrase clearly has the function of a direct object, as can be seen from the fact that it allows passivization: De appels worden gewogenthe apples are weighed. The non-referential noun phrase does not function as a direct object because it does not allow passivization: *Er wordt/worden 70 kilo gewogen. Example (120c) is therefore correctly excluded by restriction (116).

Assuming that the notion of adverbial function is an umbrella term, the formulation in (116) also correctly predicts that, for example, place and time adverbials cannot be coordinated. This accounts for the acceptability contrast in the (a)-examples in (121), but leaves puzzling why examples such as (121b) are acceptable; cf. Schachter (1977:91). We will assume that the latter case is a fixed collocation because this example refers to a single (prospective) eventuality at a certain time and place, whereas clauses with coordinated time or place adverbials usually refer to different eventualities: cf. Ik ontmoet Jan morgen en volgende week maandagI will meet Jan tomorrow and Monday next week. The claim that we are dealing with a collocation can perhaps also be supported by the fact that the order of the two coordinands is more or less fixed, since at least some speakers prefer the order in (121b) to the order in (121b').

The co-occurrence restriction on coordinands in (116), which states that the coordinands must be able to have the same syntactic function as the coordinate structure as a whole, differs from the categorial identity requirement in that it correctly allows for examples such as Jan is [[AP moe] en [PP in de war]] Jan is tired and confused in (115). In other respects, however, it seems to cover more or less the same ground, since many syntactic functions are prototypically expressed by phrases of a particular syntactic category. Nevertheless, the formulation in (116) still seems to be inadequate, since it wrongly predicts that nominal and clausal phrases can be coordinated in the function of subject or object; the primed examples in (122) are therefore wrongly predicted to be fully acceptable.

That coordinate structures consisting of a noun phrase and a clause do not often occur as a subject or object is not surprising: in the prototypical case, the coordinands arguably occupy different positions when used in the same function. The primeless examples in (123) show that argument clauses are usually preceded by the anticipatory pronoun hetit and appear in postverbal position, i.e. in a position where a nominal subject cannot appear. The primed examples show that argument clauses are not easily possible in the preverbal positions normally occupied by the nominal arguments.

As nominal and clausal arguments occupy different positions in the clause we may account for the markedness of primed examples in (122) by adding to the generalization in (116) that the coordinands should be able to occur in the same syntactic positions as the coordinate structure as a whole.

The formulation in (124) predicts that the coordinands of a coordinate structure that functions as an argument can all be nominal or all be clausal, as in as (125a-b), but cannot be easily mixed: example (122a'), repeated as (125c), is predicted to be at best equally acceptable as (123a') with a subject clause.

The same holds for object clauses; example (122b'), repeated as (126c), is predicted to be at best equally acceptable as (123b') with an object clause.

Note that the co-occurrence restrictions on coordinands in (124) do not categorically exclude the coordination of nominal and clausal constituents. For example, Section V5.1.2.3 has shown that factive verbs such as betreurento regret do allow object clauses in preverbal position, and they do indeed allow coordination of a nominal and clausal object.

Another “mixed” case allowed by the formulation (124) is given in (128): the first two examples show that nominal and clausal adverbial phrases of time can both occur in preverbal position, and the third example shows that they can also be coordinated.

The literature mentions a number of apparent counterexamples to the co-occurrence restrictions on coordinands in (124), which we will discuss next; see Zhang (2010:§3.3.1) for an overview and references. Consider the English examples in (129), which are taken to show that “mixed” coordinate structures of the form [NP and clause] can be the complement of a preposition, although it is not possible for a preposition to take a clausal complement. Example (129d) shows that mixed coordinate structures in which the clause comes first are also unacceptable.

It is relevant to note here that all the examples of mixed coordinate structures provided by Zhang are sentence-final; this allows an analysis according to which we are dealing with an extraposed PP-complement. To see why this is relevant, we need to consider the Dutch counterparts of these examples. First, consider the Dutch (130a), which shows that complement-PPs with a nominal complement can precede or follow the verb(s) in clause-final position. The (b)-examples show that PPs with a clausal complement are marked in both positions, and that the complement-PP is usually realized instead by a clause in postverbal position preceded by the anticipatory pronominal PP eropon it in preverbal position.

However, we should not account for the markedness of (130b) by prohibiting that complement-PPs may contain a clausal complement, since this is at least marginally possible in topicalization constructions such as (131a). The contrast in acceptability between (130b) above and (131a) below seems to be related to the fact, illustrated in (131b), that topicalization is excluded when the anticipatory pronominal PP eropon it is present; cf. Haslinger (2007).

Now consider the Dutch counterparts of the English examples in (129c&d), which are given in (132). The acceptability of (132a) is expected, since it is consistent with the co-occurrence restrictions on coordinands in (124), as well as with the earlier observation that clauses can occur as complements of PP-complements when the anticipatory pronominal PP eropon it cannot be used for some reason or other. The unacceptability of (132b) is not expected on syntactic grounds, but it may be degraded for the rather superficial reason that the P-NP sequences are more frequent than P-clause sequences and are therefore easier to parse. If English PP-complements behave in the same way as their Dutch counterparts, the acceptability pattern found in (129) is accounted for.

Note that examples such as (133) are also fairly acceptable in Dutch. If examples of this kind are to be considered grammatical, we may be dealing with a case of “split” coordination; Subsection IIB will analyze what looks like the extraposed part of such splits as a reduced clause.

A similar “split” coordination analysis can also account for the other mixed examples provided by Zhang. We illustrate this for example (134a), where an object DP and an adverbial PP seem to have entered into a single coordinate structure, which would be rendered as (134b) in Dutch; we provide the clause in its embedded form in order to show that we are dealing with “split’ coordination.

Our tentative conclusion from this brief discussion is that the English data in Zhang has simply received an incorrect analysis and that the generalization in (124) can be maintained in full force. This conclusion is desirable, since it is fully consistent with the hypothesis found in the semantic literature that the coordinands in a coordinate structure must be of the same semantic type (cf. Section 38.1, sub IVF), since syntactic functions can be reformulated in terms of semantic types in the prototypical case. There are, however, some problems of a theory-internal nature for the semantic approach that do not arise with (124). For example, proper nouns and definite descriptions are often taken to be semantically different: while proper nouns are usually taken to be entities (type e), definite descriptions are taken to denote sets of properties (type <<e,t>,t>). If coordinands must be of the same semantic type, we need a type-changing operation in order to allow for coordinate structures such as (135c); cf. Winter (2001a). Note that we do not present this remark as an argument against the semantic approach, since this type-changing operation may be needed for independent reasons. Indeed, it is to be expected that syntactic and semantic approaches will converge on this issue.

It is important to note that generalization (124) does not claim that a coordinate structure can always be replaced by its individual coordinands, since there may be various interfering factors unrelated to coordination that may affect the acceptability of the resulting structure. One of these factors is subject-verb agreement: while the coordinate structure in (135c) above can be replaced by its second coordinand, which would result in (135b), it is not possible to replace it by its first coordinand just like that, as this would result in an agreement mismatch: *Jan hebben gewandeld. Another factor may be the semantic selection restrictions imposed by the predicate on its arguments: a verb such as zich verspreidento spread in (136) selects a plural subject, and consequently it is impossible to omit the second coordinand even if we adjust the form of the finite verb.

The examples in (137) show that there are also inverse cases, i.e. cases where a coordinate structure cannot be used in a position where its coordinands are possible as independent phrases. This is related to the fact that example (137a) involves the fixed collocation een foto nemento take a picture, while een koekje nemento take a cookie is (137b) is fully compositional. The dollar sign indicates that, under the interpretation intended here, example (137c) would normally be interpreted as a linguistic pun, and regarded as marked otherwise.

Another idiomatic case is given in (138) with the expression de geest gevento die; the nominal expression de geest cannot be coordinated with a nominal expression functioning as a referential theme argument of gevento give; of course, the verbal expression as a whole can be coordinated with the compositional verbal projection het geld geven, as is illustrated in (138b).

A third case, adapted from Dik (1997:200ff.), is illustrated in the examples in (139): while the simple clauses in (139a&b) are both fully acceptable, the coordinate structure in (139c) is severely degraded.

Dik claims that the two verbal complements are of a different semantic type and attributes the unacceptability to the selection restrictions imposed by the perception verb ziento see. Alternative accounts might appeal to the fact that the infinitival construction in (139a) has certain special syntactic properties that the construction in (139b) lacks: for instance, the subject of the infinitival clause depends on the perception verb zien for case assignment and the verbal head of the infinitival clause and the perception verb form a verb cluster; cf. the discussion of perception verbs in Section V5.2.3.3. It is not so easy to choose between the available options: an appeal to selection restrictions would predict that examples such as (140) are unacceptable, while e.g. the verb-cluster approach would predict that such examples are possible. However, examples like those in (140) seem to have an intermediate status; they are marked but certainly less degraded than example (139c).

Whatever the correct analysis of the degraded status of (139c), this case also illustrates that the co-occurrence restrictions on coordinands in (124) do not imply that a coordinate structure can always be used in a position where its coordinands are possible as independent phrases, since there may be interfering factors unrelated to coordination that may affect the acceptability of the resulting structure.

A final case worth mentioning is illustrated in (141). These examples show that the selection restrictions imposed by the predicative coordinands on their subjects must be similar; while the individual-level predicate zijn zoogdieren triggers a generic reading on the bare plural subject, the stage-level predicate op dit moment buiten blaffen triggers an indefinite reading. The examples in (142) show that we find essentially the same for the singular noun phrase een honda dog.

As far as Dutch is concerned, one might ask whether it is really necessary to assume that coordinated predicates must impose similar selection restrictions on their subject, given that generic and indefinite subjects clearly occupy different positions in the (a)- and (b)-examples: the co-occurrence restriction on coordinands in (116) is thus sufficient to exclude the (c)-examples. Furthermore, the examples in (143) show that predicates selecting a cumulative and a distributive subject, respectively, can be coordinated.

Since the English renderings of (141c) and (142c) are also unacceptable (cf. Zhang 2010:188), despite the fact that the generic and indefinite subjects seem to occupy the same position, this may suggest that we are minimally dealing with a restriction on the semantic type of predicative coordinands: “mixed” cases of individual and stage-level predicates cannot be coordinated.

The previous subsections have shown that coordinate structures must satisfy the co-occurrence restrictions on coordinands in (144). We have found only a limited number of remarks on restrictions concerning the semantic nature of the coordinands in the literature we have consulted; more research seems to be needed to determine whether any finer distinctions go beyond the individual and stage-level distinction. We have not extensively discussed the hypothesis found in the semantic literature that the coordinands should be of the same semantic type (entity, predicate, etc.), because this hypothesis is more or less equivalent to the first part of restriction (144a); see Subsection B for a brief discussion of this.

It should be noted that there are a number of co-occurrence restrictions that are not (immediately) related to those given in (144). Given the difference in the syntactic function of the two PPs in (145a), it follows directly from (144a) that the coordinate structure in (145b) is impossible (on the intended interpretation), but it may not be immediately clear why (145c) is also unacceptable, since this is only indirectly related to syntactic function; here the PP cannot be interpreted as an argument and a place adverbial at the same time.

The (c)-examples in (146) and (147) show that coordination is sometimes also sensitive to the thematic role of the coordinands: while the coordinands can all be used as subjects they cannot be coordinated due to the fact that they have different semantic roles: agent versus instrument/means.

Although this does not seem to follow from (144a) at first sight, it can be derived from it in a straightforward way. While Section V9.5 has argued that agents (i.e. the external argument of V) are arguably base-generated in the highest specifier of the lexical part of the verbal projection (vP), it seems plausible that instruments/means are generated in some other position (e.g. corresponding to that of the met-PP); if this is indeed the case, the unacceptability of the (c)-examples follows from (144a) on the plausible (and often tacitly adopted) assumption that coordinate structures are formed before they are inserted into some larger syntactic structure.

Coordinate structures exhibit island effects in the sense that it is usually not possible to extract a constituent from them. This generalization is known as the coordinate structure constraint; the version in (148) is adapted from Ross (1967 (4.84)).

The two instances mentioned in (148) are illustrated in (149) by wh-movement of (a part of) a direct object: while (149a) shows that it is possible to wh-move an object as a whole, the (b)-examples in (149) show that it is not possible to wh-move the first (or second) coordinand of a coordinate structure functioning as a direct object, and the (c)-examples show that it is not possible to wh-move an object if the verbal projection including it functions as a coordinand in a coordinate structure.

Note in passing that the primed examples are also unacceptable when the wh-phrase remains in situ, as is clear from the fact that example (150a), in which the coordinate structure remains in its base position, and example (150b), in which the coordinate structure is wh-moved into the clause-initial position, are both unacceptable. This shows that coordination of interrogative and non-interrogative phrases is always unacceptable unless we are dealing with an echo-interpretation: cf. Jan heeft jouw boek en wat\`1welk artikel gelezen? Jan has read your book and what/which article?.

A possible counterexample to the coordinate structure constraint in (148) is given in (151), which suggests that in some cases it is possible to split coordinate structures by placing the coordinator and the following coordinand (henceforth: [en/of/maar XP]) after the verbs in clause-final position. Note that some speakers prefer to repeat the preposition metwith in examples such as (151b') and that this actually applies more generally to adverbial phrases of e.g. place and time: cf. ze hebben na het ontbijt gewandeld en ??(na) het avondeten They walked after breakfast and after dinner.

The coordinate structure constraint predicts that the split pattern cannot be derived by movement and the following two subsections will show that there are indeed independent reasons for assuming that a movement approach is not viable. We will therefore argue in favor of an ellipsis approach, i.e. that we are dealing here with clausal coordinands with ellipsis in the second coordinand: example (151a'), for instance, is assigned the structure [[dat ik gisteren Marie ontmoette] en [dat ik gisteren Jan ontmoette]], in which strikethrough indicates non-pronunciation.

It has been claimed that in certain forms of so-called asymmetric coordination it is possible to violate the part of the coordinate structure constraint that prohibits extraction of a phrase from a single coordinand. A prototypical case for English is given in (167b), which apparently allows wh-movement from the second (but not from the first) coordinand.

Example (168b) shows that similar examples are definitely impossible in Dutch. This strongly suggests that Ross (1967:§4.2.3) and Schmerling (1975) are correct in arguing that the examples in (167) are cases of “fake” coordination, although this has remained a topic of debate; we refer the reader to Zhang (2010:§5.3) for a concise review and references.

Section 38.4.1, sub IC, will show, however, that there are several other types of asymmetric coordination in Dutch which express a special semantic (causal, concessive, etc.) relation between the first and the second coordinand. Van der Heijden (1999:65) has claimed that some of these cases also allow extraction of a phrase from a single coordinand. An example (adapted from her work) is (169b); the percent sign in (169b) is used to indicate that some speakers (including ourselves) do not find this example acceptable.

It is not obvious that examples such as (169b) should be taken as counterexamples to the coordinate structure constraint, because their internal structure is not clear. First, the presupposition that we are dealing with some kind of VP-coordination is problematic, because it does not seem to be possible to construe the root (deontic) modal kunnento be able in the examples in (169) with the second coordinand, as is clear from the fact that the modal verb in (170a) preferably receives an epistemic reading. Second, De Vries (2005) notes that extraction is only possible from the first coordinand: the fact that examples such as (170b) cannot be derived from (169a) would be surprising if the coordinate structure constraint did not apply to asymmetric coordinate structures.

De Vries (2005) suggests a third type of possible counterexample, which is based on a construction known in the Dutch literature as balansschikking, which will be discussed in more detail in Section 38.4.1, sub IID. Consider example (171a), in which the first coordinand modifies the second conjunct in the way indicated by the English translation. De Vries takes the controversial, but possibly correct, position that we are dealing with normal disjunction and claims that wh-extraction from the first coordinand is acceptable. For the sake of the argument, we will assume that (171b) is grammatical although some informants (including ourselves) reject such examples, as is indicated by the percent sign.

It should be noted, however, that example (171b) is not a counterexample to the coordinate structure constraint because we are dealing with coordination of two main clauses and wh-movement targets the first position of the first main clause. Or, to put it differently, we are dealing with a disjunction of an interrogative and a declarative main clause (which may be the reason why some informants reject this example; cf. Section 38.4.1, sub IIA). The structure of (171b) is thus as given in (172), which does not violate the coordinate structure constraint.

This analysis also accounts for the fact noted by De Vries that a similar “extraction” from the second coordinand is impossible: the only way to derive *Wati was Jan nog niet thuis of hij gaf Marie ti from (173a) is by moving the wh-object of the second clausal coordinand clause into the initial position of the first clausal coordinand, which is not allowed for independent reasons.

This subsection has discussed some possible violations of the coordinate structure constraint involving asymmetric coordination. The first case can only be demonstrated for English, and it has been argued that it involves “fake” coordination. The second case can be found in both English and Dutch (although not all Dutch speakers accept it), but it is unclear whether it is really run-of-the-mill coordination. The third case (also rejected by some speakers) is found in so-called balansschikking constructions, but has been shown to be irrelevant, since it seems to be based on the false assumption that the movement targets a position external to the first coordinand. We therefore conclude that wh-movement in asymmetric coordination does not provide conclusive evidence against the coordinate structure constraint.

A well-known exception to the coordinate structure constraint is so-called across-the-board movement: extraction from a coordinate structure is possible if the movement is applied in such as a way that it affects the same type of constituent in all coordinands. A characteristic example is given in (174a), where the wh-phrase welk boekwhich book is related to the two interpretative gaps indicated by ti, which function as the direct object of the first and the second coordinand, respectively. The (b)-examples give the corresponding examples in which the movement is not applied in an across-the-board fashion for comparison. Detailed discussions of across-the-board movement can be found in Ross (1967), Williams (1978), and De Vries (2017), and the references cited therein.

Note that across-the-board movement of the full coordinands themselves, as in (175b), is not possible. However, this should not be seen as an exception to across-the-board movement, since it may be due to the independently established fact that the coordinands in a coordinate structure cannot be identical in the prototypical case; examples such as (175a) are impossible if the two proper nouns refer to the same person. The unacceptability of (175b) is thus related to the fact that the answer to this question will provide a noun phase identifying a unique girl, which also functions as the “filler” of the two interpretative gaps, so that providing the answer Marie will result in a representation similar to the one assigned to (175a).

This account of the unacceptability of (175b) is supported by the fact that question (174a) presupposes that there is a unique book that was read by Jan and studied by Els. It should be noted, however, that De Vries (2017) claims that the presupposition of uniqueness need not always be present in the case of across-the-board movement. This is especially true for examples such as (176), where the moved phrase contains an anaphor like zichzelfhim/herself. Speakers who accept such examples require a so-called sloppy reading of the reflexive zichzelf: zichzelf is construed as co-referential with Jan and Marie, respectively. Since the acceptability judgments on examples such as (176) are not very clear and seem to vary from speaker to speaker, we will not discuss them any further.

The following subsections will examine across-the-board movement in greater detail but first it should be noted that this type of movement is not restricted to coordinate structures with two coordinands. The examples used for illustration will generally involve coordinate structures with two coordinands for reasons of simplicity, but the reader should keep in mind that across-the-board movement can also apply to coordinate structures with three (or more) coordinands; we illustrate this in example (177).

We will follow the general practice of restricting our attention mainly to conjunctions with enand: again, the reader should keep in mind that across-the-board movement is also possible in disjunctions with ofor and in adversative conjunctions with maarbut; this is illustrated in (178).