This chapter has shown that the claim that lexical items like als/danas/than, behalveexcept/besides, in plaats vaninstead of and laat staanlet alone are (or can be) coordinator-like categories rests mainly on the presupposition that gapping is possible only in coordinate structures; if we reject this presupposition, we can simply analyze these elements as subordinators (e.g. prepositions) which can take a reduced clausal complement. This also solves the problem that the gapping operation in the constructions under discussion differs from gapping in coordinate structures in that it (apparently) applies not only in a forward but also in a backward fashion, by pointing to the fact that the gapped clause is embedded in a clausal constituent that can be topicalized. Thus, the difference between gapping in the constructions discussed in this chapter and gapping in coordinate structures follows from the fact that topicalization of a string such as en-XP in (53) violates the coordinate structure constraint discussed in Section 38.3, sub IIB.

If gapping is possible in subordinate clauses, we expect that it can be applied to all contrastive focus/topic structures, provided that the elided material can be recovered from the immediate linguistic environment. This is a desirable step, as it removes an ad hoc restriction on gapping, and also supports our earlier conclusion in Section 39.2, sub IH, that gapping may be involved in the derivation of fragment clauses. Another possible case is given in (54b) adapted from Van der Heijden & Klein (1995) and Haeseryn et al. (1997: 1605-6).

Conditional constructions of the type in (54a) were analyzed in Section V8.3.3 as regular main clauses (here: dan mag Marie Jan ook kussen) preceded by an extra-sentential conditional clause (here: als Jan Marie mag kussen), which is resumed in the main clause by the proform danthan. It is especially important here that dan is located in the initial position of the main clause and therefore has to be considered as a clausal constituent, which implies that assuming semi-coordinator status for dan is impossible. Example (54b) shows that the conditional clause can nevertheless trigger some kind of gapping in the main clause, which again cannot be easily understood if gapping is restricted to coordinate structures. On the other hand, dropping this restriction opens up new avenues for investigating examples of the sort in (54b).

We would like to mention a final potential problem with the assumption that gapping is involved in the derivation of the constructions discussed in this chapter. Consider the examples in (55), adapted from Van der Heijden (1999:20), who judges them all to be acceptable.

The reduced clauses contained in the bracketed phrases cannot have been derived by gapping, since they all contain a finite verb and finite verbs cannot occur as remnants of gapping. That examples like those in (55) are special is clear from the fact that the bracketed phrases are only possible in embedded clauses; the main clauses in (56) are all degraded compared to the corresponding examples in (55). The examples in (56) become fully acceptable if we replace the finite verb by the full finite clause dat hij zwemtthat he swims.

That the reduced clause cannot be a main clause can be easily demonstrated by example (57a): the fact that the verbal particle must precede the finite verb shows that the verb belt occupies the final position of the reduced clause. Example (57b) again shows that the antecedent clause of the reduced clause following dan cannot be a main clause.

The unacceptability of (56) is to be expected under a gapping account but the acceptability of the examples in (55) remains a mystery, and the acceptability contrast between the two sets of examples is in any case surprising. For the moment, we will tentatively assume that the examples in (55) are not gapping constructions but represent a construction in its own right, the investigation of which we will have to leave to future research.