The introduction mentioned that gapping is possible in coordinate structures with enand, ofor and maarbut; examples illustrating this are given in (62a&b). Example (62c) shows that gapping is not possible in e.g. asymmetric coordinate structures with wantbecause or dusso.

The examples in (63a&b) show that the correlative coordinators en ... en ...both ... and ... and of ... of ...either ... or ... can also be used in gapping constructions although some speakers seem to consider cases with en ... en ... to be somewhat marked, as indicated by the percent sign in (63a). Example (63a') shows that the correlative zowel ... als ...both ... and ... cannot be used, which is to be expected because this correlative coordinator cannot be used for the linking of clauses.

The remnants of gapping and their antecedents are characterized by the fact that they (usually) receive a contrastive accent; cf. Hartmann (2000: §4). This requirement immediately explains why elements that cannot be accented, such as the reduced pronouns in (64b), cannot occur as remnants in gapping constructions; cf. De Vries (1992:130).

Because of their contrastive focus reading, gapping constructions such as (64a) are very suitable for giving pair-list answers to questions like Wie las wat?Who read what? or Wat hebben Jan en Marie gelezen?What did Jan and Marie read?. In fact, wh-phrases themselves can also be used as remnants in gapping constructions because they are inherently focused: example (65b) is perhaps a bit laborious because the same question could be asked by the simpler form Wat lazen Jan en Els?What did Jan and Els read?, but otherwise it seems perfectly acceptable. For completeness, example (65c) shows that gapping is also possible in multiple wh-questions.

The fact that the remnants of gapping and their antecedents are assigned a contrastive accent can resolve certain potential ambiguities; cf. Hartmann (2000) and Boone (2014). A gapping construction such as Marie gaf Jan een boek en Els een CDJan gave Marie a book and Els a CD can have the two interpretations indicated by the two structures in (66a&b): Els functions as the subject of the target clause in the (a)-examples and as the indirect object in the (b)-examples. The intonation of the antecedent clause resolves this ambiguity; when the contrastive accent is assigned to the subject Marie of the antecedent clause, Els also functions as the subject of the target clause, but when it is assigned to the indirect object Jan, Els functions as the indirect object of the target clause. The primed examples are added to show that the use of weak pronouns can also help to disambiguate such structures because they cannot be (contrastively) accented.

Haeseryn et al. (1997:1592) notes that the reading in (66b) is usually more salient than that in (66a). If so, this may be related to (i) the fact that subjects are more likely to be interpreted as topics (i.e. discourse-given information) than objects, and (ii) the fact that the use of contrastive accent in the first conjunct is of secondary importance and may even be absent in some cases (cf. Subsection H below).

Gapping is a reduction process prototypically found in clausal coordinate structures. The contrast between the examples in (67a&b) shows that gapping minimally elides the finite verb: even if the finite verb is given a contrastive accent, it cannot be realized overtly (although the string Jan las een boek en Marie schreef is of course acceptable if the verb schrijvento write is interpreted as a pseudo-intransitive verb). The contrast between the examples in (67a&c) also shows that gapping only applies in a forward fashion. Finally, example (67a) shows that gapping differs from the two types of conjunction reduction discussed in Section 39.1 in that the elided element(s) need not be in the periphery of the target coordinand.

The position of the finite verb does not seem to play a role, as can be seen from the fact that gapping is not restricted to main clauses. First, example (68a) shows that gapping is also possible in the embedded counterpart of (67a); Subsection IIA3 will return to the fact that the complementizer datthat of the gapped clause must also be elided. Second, the two (b)-examples in (68) show that the finite verb can also occur in clause-initial position.

As mentioned above, gapping can only apply to clauses, i.e. not to coordinate structures with nominal, adjectival, or adpositional coordinands. This is illustrated in (69) for the inf- and ge-nominalizations of example (67a); the em-dash stands for the (presumed) gapped string voortdurend lezen/voortdurende gelees.

These observations have led to the finite-verb restriction in (70).

The finite-verb restriction predicts that gapping cannot be applied to infinitival clauses. Subsection IIA will show that this prediction is incorrect, but for our present overview we can provisionally accept this restriction, since we will be dealing only with finite clauses.

Besides finite verbs, various kinds of other material can be elided. The elided string need not be a single constituent: in (71), for example, it consists of the finite verb, an indirect object, and two adverbial phrases. The only requirement seems to be that the remnants of gapping have the same syntactic function as their correlates in the antecedent clause.

The examples in (72) further show that the elided material does not have to form a single contiguous string, but can also be scattered throughout the target clause. Example (72a) shows this for the non-finite verb gelezenread, (72b) for a direct object, and (72c) for a prepositional indirect object.

Example (73a) again shows that the elided material in the target clause can consist of more than one (clausal) constituent. However, the elided material has to be locally recoverable: since the adverbial phrase voor haar verjaardag in (73b) has no correlate in the antecedent clause, gapping is excluded (as is clear from the fact that this phrase cannot be present in the interpretation of the target clause).

The examples in (74) further show that it is not easily possible to leave remnants in the target clause that are identical to some constituent in the antecedent clause, i.e. there is a maximization condition on gapping that requires all non-contrastive material to be elided (but see Subsection IID5 for a notable exception). Since the acceptability of example (74h) shows that gapping as such is not obligatory, this shows that gapping is usually an all-or-nothing operation, in the sense that it affects the non-contrastive part of the target clause as a whole, and not the non-contrastive constituents individually.

There is not only a maximization condition on elision, but also on interpretation, in the sense that all non-contrastive material from the antecedent clause must be present in the interpretation of the target clause. This suggests that the target clause in (71) must have the form in (75a) and cannot have the forms in (75b-h). Note that all the examples are acceptable if the elided constituents are overtly realized.

The recoverability and maximization conditions on gapping correctly imply that the remnants of gapping must have the same syntactico-semantic functions as their correlates in the antecedent clause; cf. all the examples discussed so far.

In polyadic coordinate structures, gapping can affect more than one coordinand; the examples in (76a&b) show that (in principle) it can apply to an infinite number of coordinands under identity with elements in the initial coordinand, as long as no coordinand is skipped by gapping. However, (76c) shows that gapping need not apply to all coordinands. Example (76d) further shows that examples in which gapping applies under identity with elements in a non-initial coordinand are degraded: the antecedent clause must be the initial coordinand.

The internal organization of the complex coordinate structure does not affect this: the gapping process simply proceeds from left to right, from one coordinand to the next, without regard to the hierarchical relations between the individual coordinands. This is illustrated by the examples in (77): (77a) shows that in a complex coordinate structure such as [[WP and XP] or [YP and ZP]], WP can trigger gapping in XP, YP, and ZP, while (77b) shows that it is impossible for the complex coordinand [WP and XP] to trigger gapping of the finite verbs in the complex coordinand [YP and ZP]. It is also impossible to construct acceptable examples in which WP triggers gapping of the finite verb in YP, or in which XP triggers gapping of the finite verb in ZP (not shown here).

Example (77a) alternates with (78a), in which the finite verb of YP but not of ZP is overtly realized, but it does not alternate with (78b), in which the finite verb of ZP but not of YP is overtly realized. The acceptability of (78a) is expected if we assume that the two complex coordinands of the disjunction ofor function as separate domains, in which the initial coordinands can independently trigger gapping in the non-initial coordinands: while WP triggers gapping in XP, YP triggers gapping in ZP. The unacceptability of (78b) seems more difficult to explain in light of the acceptability of (76c) above; we may be dealing with some (superficial) parallelism requirement on the two complex coordinands, but we will not push this idea here for lack of conclusive arguments.

In the classic case, gapping remnants are major phrases in the sense of Neijt (1979): they can be clausal constituents (arguments, complementives and adverbials) or specific smaller, non-clausal verbal projections, which we will loosely refer to as “VP” for convenience. Examples of VP-remnants are een film bekijkeninf/bekekenpart watch(ed) a movie in (79).

That the remnants of gapping are typically clausal constituents of the target clause, and thus cannot be more deeply embedded, is illustrated by the acceptability contrast between the primeless and primed examples in (80). The (a)-examples show that while the remnant can be the object clause of the elided finite verb weetknows, it cannot be a constituent embedded in that clause; the (b)-examples similarly show that while the remnant can be the nominal object of the elided verb kocht, it cannot be a phrase embedded in that noun phrase.

Examples with a postnominal van- or over-PP are less suitable for demonstrating this because they are not easy to distinguish from adverbial PPs headed by van or over with a restrictive function; cf. Section N15.2.1, sub VC, for further discussion.

There are a number of exceptions to the generalization that the remnants must be clausal constituents of the elided finite verb of the target clause. This is illustrated in (81): the PP op Peter in (81a) functions as the complement of the adjective boos, but it can still appear as a remnant of gapping; similarly, the measure phrase vier meterfour meters in (81b) functions as a modifier of the adjective diepdeep, but again it can still appear as a remnant of gapping.

The restriction to major phrases and the counterexamples to this restriction will be discussed in more detail in Subsection II; it will be argued that the exceptional cases in (81) are not accidental, and that the acceptability contrast between examples like (80b') and (81a) is related to the contrast in acceptability of focus movement in the primed examples of (82); cf. Subsection IID. The number sign in (82a') indicates that the PP op het plein can function as a place adverbial indicating the location of the shopping event, which is not relevant here.

The examples in the previous subsections all involve cases with exactly two remnants of gapping. However, Neijt (1979) claims that it is also possible to have cases with one remnant, or three or more remnants. However, gapping constructions with more than three remnants sound artificial and quickly deteriorate as the number of remnants increases. This may be related to the fact that they contain so many contrastive accents (cf. Johnson 2017:1745), but at first glance there is no clear reason to assume that they are ungrammatical (although this may in fact follow from the analysis of gapping proposed in Subsection IID).

Examples such as (83a) were discussed in Section 38.3, sub IIB, under the heading of split coordination. There we left open the question whether we are dealing with a kind of fragment clause or with gapping. The gapping analysis is supported by the fact that the remnant in the target clause is usually contrastively accented, but it may be problematic in view of the fact that its correlate in the antecedent clause is usually not contrastive. Thus, if split coordination were analyzed as a case of gapping, the focus requirement should be restricted to the remnants of gapping. Support for this conclusion can be found in the fact that gapping also seems to occur in dialogues such as (84), where speaker B provides additional information to A’s assertion; cf. Neijt (1979:37ff.). The crucial fact is that A’s utterance would normally have a non-contrastive intonation pattern because A does not anticipate B’s reply.

The fact that the examples in (83) and (84) are fully parallel suggests that they should be given a similar account: we will therefore assume that contrastive accent on the antecedents of the remnants of gapping in (83) is indeed of a secondary nature, due to conscious planning by the speaker. Because speaker A clearly does not anticipate B’s reaction in the dialogues in (84), the contrastive accent on the antecedents will not be realized. Finally, note that fragment clauses of the kind in (85) look very much like gapping; this raises the question whether such clauses also involve gapping. We think that they do, and refer the reader to Section 39.4, where it will be argued that the traditional claim that gapping is found only in coordinate structures is incorrect.

Subsection IC discussed the standard assumption that gapping obligatorily elides the finite verb of the target clause. This subsection will show that this generalization correctly predicts that both main and embedded finite clauses can be the target of gapping. It also predicts that target clauses cannot be infinitival because such clauses by definition do not have a finite verb, but this prediction will be shown to be incorrect. This subsection also introduces an additional restriction: the antecedent clause and the target clause must both be immediate constituents of the coordinate structure, i.e. they cannot be embedded in the coordinands.

This subsection takes as its starting point the traditional view that gapping remnants are typically clausal constituents (e.g. arguments, adverbials, or complementives) of the target clause, as expressed by the clausemate restriction in (87a), repeated here as (106). This subsection presents data supporting this restriction; possible counterexamples are discussed in Subsection C.