This subsection discusses the status of the lexical items alsas and danthan in comparative constructions like those given in (186). Such als/dan-phrases cannot be analyzed as regular PPs with a nominal complement, because this would incorrectly predict that the noun phrases must be assigned objective case. Note in passing that in colloquial speech the examples in (186) occasionally alternate with cases in which the pronouns appear in the object form, just as in English; cf. taaladvies.net/groter-dan-mij-of-ik. However, since such cases occur only if the complement of dan/als can be analyzed as a noun phrase, we will ignore them in our discussion of gapping below, since this phenomenon only occurs in clauses.

The examples in (187) show that the case of the pronouns depends on the noun phrase to which it is compared; the noun phrase in the als/dan-phrase receives nominative case when it is compared to the subject of the main clause, as in (186) and the (a)-examples in (187), whereas it receives accusative case when it is compared to the direct object, as in the (b)-examples in (187).

Since nominative case is usually restricted to subjects of finite clauses, the examples in (186) suggest that the complements of als and dan are clausal in nature. This is also clear from the examples in (188), which contain a finite verb and the (optional) complementizer datthat in the complement of als/dan. It is assumed that examples like these contain an empty adjectival predicate [e], which receives an interpretation from its adjectival correlate in the matrix clause; this is discussed in detail in Section A26.1.3, but since it plays no role in our discussion here, the reader can ignore this in the following discussion.

The examples in (187) can now be derived by eliding everything in the clausal complement of als/dan except the compared noun phrases, as in (189). Note that we have ignored the fact that the non-reduced forms in (188) sound a bit clumsy compared to the more economical reduced version.

The above analysis raises the question of what kind of elision operation we are dealing with. Since the deletion operation in (189) affects the finite verb (as well as the complementizer), we hypothesize that we are dealing with some kind of gapping. This predicts that the complement of the als/dan-phrase can also contain more than one remnant, and the examples in (190) show that this prediction is indeed correct.

The gapping analysis also predicts that the remnants must be contrastively accented. Indeed, this seems to be the case; further support comes from the fact that weak (i.e. unaccented) pronouns do not occur in the reduced constructions in (191), which should be compared with those in (189) and (190).

We also expect that the presumed remnants of gapping cannot be embedded in a clausal constituent (argument, complementive or adverbial), which also seems to be borne out. The acceptability contrast between the examples in (192) shows that while the remnant can be a full prepositional complement, it cannot be its nominal part. Recall that, for our present purposes, the empty position [e] in the reduced clause can be ignored; cf. the remark above example (188).

The discussion so far strongly suggests that the reduction operation affecting the complement of als/dan can indeed be identified as gapping. Hendriks (1995: §2) takes the fact that gapping is possible in comparative constructions as sufficient evidence that alsas and danthan are coordinators; cf. also Haeseryn et. al (1997:1601). If this were true, we would also expect backward conjunction to be possible in comparative constructions, and Hendriks shows that this expectation is fulfilled; cf. also Van der Heijden & Klein (1995). Elision by backward conjunction reduction is indicated in bold to distinguish it from elision by gapping.

Note that backward conjunction reduction is only possible if it follows gapping, because otherwise the finite verb lees in the clause following dan in (193a) would block its application. This rule order follows automatically from our claim in Section 39.3 that gapping and backward conjunction differ in that the former is a syntactic rule while the latter is a post-syntactic rule.

A serious problem for the hypothesis that we are dealing with coordination is that the clausal complement of als/dan in (194) must appear in its embedded form, while the matrix clause is a main clause. That the clauses following dan are non-main clauses is clear from the clause-final placement of the finite verb and the (optional) presence of the complementizer datthat; cf. Van der Heijden & Klein (1995).

For this reason, some researchers consider the elements als and dan as borderline cases: although they behave like coordinators in some respects, they also have properties not found with regular coordinators like enand and ofor; cf. Van der Heijden (1999). An obvious obstacle to the conclusion that als and dan are some kind of mixed category (i.e. with properties of coordinators and subordinators) is that they clearly do not coordinate the strings preceding and following them. Instead, the als/dan-phrases are constituent parts of the adjectival phrases headed by vakermore often and lieverrather (lit. “more gladly”), as evidenced by the fact, shown in (195), that the adjective and the dan/als-phrase can together occupy the initial position of the sentence; cf. Haeseryn et al. (1997:1602). The use of parentheses indicates that this is true for both the non-reduced and the reduced form; this also shows that topicalization does not affect the acceptability of the reduction.

The claim that dan and als are borderline cases between coordinator and subordinator rests mainly on the presupposition that gapping (and backward conjunction reduction) are hallmarks of coordination. However, one could also conclude from the evidence of examples like (194) and (195) that this presupposition is incorrect and that gapping is relevant in a wider range of constructions. And indeed, there is no a priori reason to reject this possibility if we see gapping as a process that can be applied to any construction with contrastive topics/foci, as long as the elided material is recoverable from the linguistic context.

Van der Heijden (1999) has shown that there are other constructions with als and dan with similar properties to the comparative constructions discussed above, but since they can be analyzed in a similar way, there is no real need to go into these cases. However, we would like to conclude this subsection with a brief digression on our earlier discussion of the more or less fixed sequences evenmin/zomin ... als ..., which are given in Haeseryn et al. (1997:544/1518-22) as cases of correlative coordinators: Ik heb [evenmin Jan als Marie] gezienI have seen neither Jan nor Marie. Section 38.4.2, sub IB, has argued that these sequences are not coordinators but comparative adjectival phrases. This is also supported by the fact, illustrated in (196), that the als-phrases show a behavior with respect to reduction that is similar to the als/dan-phrases discussed earlier in this subsection: (196a) shows that pronouns following als can be nominative; the contrast between (196a) and (196b) shows that the case marking of the pronoun depends on its correlate in the main clause; and (196c) shows that we can derive gapping-like structures.

The examples in (196) can only be accounted for if we assume that als takes a clausal complement, but it is still impossible to analyze them as involving coordination. First, the complements of als are embedded clauses, as is clear from their non-reduced counterparts in (197).

Second, the fact illustrated in (198) that the phrase evenmin als ... can be topicalized shows that it functions syntactically as a clausal constituent of the matrix clause. The parentheses again indicate that this is true for both the reduced and for the non-reduced form.

The examples in (196) to (198) thus conclusively show that evenmin/zomin ... als ... cannot be analyzed as a correlative coordinator, which supports the conclusion from Section 38.4.2, sub IB, that we are dealing with a more or less ordinary comparative adjectival construction.

This subsection discusses the status of the lexical element behalveexcept/besides in the examples in (199). These examples show that behalve has two different readings, one restrictive and one additive; cf. Haeseryn et al. (1997:1604). Restrictive behalve typically modifies a (positive or negative) universal quantifier in the main clause, while additive behalve is typically associated with a phrase modified by the focus particle ookalso; cf. Paardekooper (1966). Note that restrictive behalve exhibits more or less the same behavior as uitgezonderdexcept, which is not discussed here.

In the following, we will refer to the italicized string as the behalve-phrase, and to the universal and focused phrases as its associates. These terms should be taken as convenient descriptions, with no intention of saying that behalve-phrases and their associates function as separate clausal constituents. This is because the categorial status of behalve is controversial. Although Komen (1994: §6.1) has shown that the debate dates back to the 18th century, we take Paardekooper (1966/1986: §3.5.1) as the starting point of the debate in modern Dutch linguistics. Paardekooper argues that behalve functions as a coordinator in the sense that behalve-phrases and their associates are initially parts of a coordinate structure: [[iedereen/niemand] behalve [Els]] and [[ook Peter] behalve [Els]]. This is disputed by Landman & Moerdijk (1980), which argues for a prepositional status for behalve; behalve-phrases function as separate clausal constituents with an adverbial function. Since then, the status of behalve has been a recurring theme in the literature; cf. Klein (1985), Kooij (1992), Komen (1994: §6), Van der Heijden & Klein (1995), Haeseryn et al. (1997), Van der Heijden (1999: §1), and others. This subsection will review the key facts that have played a role in the debate about the categorial status of behalve, but we begin with the various analytical stances that have been advocated in the literature.

Judgments about the use of the case forms of the pronouns in examples like those in (224) are not very sharp: they vary from case to case, and most likely from person to person. Our intuition is that the use of the objective form is preferred when the examples are pronounced in a neutral way, while the nominative form requires a contrastive accent.

If this intuition is on the right track, we can conclude that in plaats vaninstead of is a regular (phrasal) preposition that may select either a nominal phrase, which is then assigned objective case, or a reduced clause, in which case nominative can be assigned to the noun phrase in the reduced clause. That reduction may very well be involved in the derivation of the primed examples in (224) can be supported by the fact, illustrated by the primeless examples in (225), that in plaats van can take a finite clause; the primed examples with the nominative pronoun hij can now be derived by gapping, as in the primed examples.

The gapping analysis in the primed examples of (225) is also supported by the fact that nominative case is obligatory in examples with two remnants such as those in (226); cf. Klein (1985).

It would be unwise, however, to take the possibility of gapping as an argument in favor of assuming a coordinator-like status for in plaats van, because in all examples this element can be shown to head a clausal constituent. This is clear from the topicalization constructions in (227); the parentheses are intended to show that both the non-reduced and the reduced variant can be placed in main-clause initial position.

The line of reasoning of the previous three subsections can be repeated for the (lexicalized) phrasal element laat staanlet alone (lit. “let stand”), which occurs only in negative clauses; cf. Haeseryn et al. (1997:566). The primeless examples in (228) make it clear that laat staan can be followed by a finite clause, which means that cases in which it is followed by a noun phrase can in principle be derived by gapping. That gapping is involved is clear from the fact that (nominative or objective) case marking of the pronouns in the primed examples depends on their correlate in the main clause.

The remnants in the primed examples cannot be replaced by a weak pronoun, but unlike the cases discussed in the previous subsections, this cannot be used as an argument for assuming that we are dealing with a gapping-like construction because the same holds for the pronouns in the non-reduced primeless examples. This may be due to the fact that the laat staan-construction is inherently contrastive; the use of the parentheses is again intended to show that the non-reduced and the reduced variant exhibit identical behavior.

Note also that the laat staan-phrases in (228) cannot be topicalized. It is not clear what this shows, since this is a more common feature of phrases that can only occur in negative contexts: cf. the negative polarity item ook maar iets in Niemand heeft ook maar iets gezienNobody has seen anything and *Ook maar iets heeft niemand gezien.