One problem with a deletion analysis is that it does not explain why Q-er must be present; example (264) in Section 20.2.4, sub I, has already shown that N-ellipsis is possible without Q-er (see also Section A28.4). In fact, the acceptability contrast between the two examples in (324) shows that Q-er is blocked in N-ellipsis contexts; the number sign in (324b) shows that this example is acceptable if er is assigned a locative interpretation, which is not relevant for the present discussion. For this reason we should probably reject this analysis. -

Barbiers (2017) has proposed a second argument against the deletion analysis. First, the examples in (325) show that elided nouns in N-ellipsis constructions retain their gender and number features: the article het and the relative pronoun dat in (325a) show that the elided noun is neuter and singular; (325b) shows that changing the number of the elided noun (the interpretation is, of course, determined by the context) triggers changes in both the article and the relative pronoun.

Barbiers claims that speakers of the northern varieties of Dutch do not make a similar distinction in constructions with Q-er: the (a)-examples in (326) show that such speakers can use the relative pronoun die both when the interpretive gap is interpreted as huishouse, which is neuter in Dutch, and when it is interpreted as autocar, which is non-neuter.

Note that we have placed northern between quotation marks to indicate that some of our northern informants have the southern judgments, which suggests that the stratification of the distinction needs further investigation.

According to the pro-analysis proposed in Bennis (1986) and Kester (1996), the presence of Q-er is required to license the phonetically empty (and base-generated) pronominal element pro: eri ... [Num/Q [ proi ]]. Assuming that the licensing relation involves binding, the analysis can be used to explain at least some of the basic properties of the relation between Q-er and its nominal associate. For example, since binding requires the binder to c-command the pronominal element, we correctly predict that er must precede its associate noun phrase (when both are in the middle field of the clause).

Example (328a) further shows that the noun phrase can be moved into the clause-initial position, and this also follows from the present analysis, as the same holds for reflexive pronouns; the topicalized phrase is “reconstructed” into its original position (indicated by the trace tj) as far as its binding properties are concerned.

Finally, if we assume that the binding relation between Q-er and its nominal associate is local in the same sense as the binding relation between a reflexive pronoun and its antecedent, we also correctly predict that er and its nominal associate must be part of the same clause: example (329) shows that placing er in some higher clause leads to unacceptability (cf. Bennis 1986).

Despite its descriptive success, the proposed analysis has at least two potential flaws. First, it is not clear why the antecedent of pro must be er and cannot be some more meaningful element that could also identify the semantic content of pro. Second, it is not clear how Q-er itself is licensed; usually all elements in the clause are licensed by being in a selection or modification relation with some other elements in the clause, but this does not seem to hold for er in the proposed analysis; it is not assigned a thematic role, nor does it have an obvious modification function.

The two problems mentioned for the pro-analyses do not arise in the movement analysis, proposed by Coppen (1991), Barbiers (2009/2017) and Kranendonk (2010), according to which Q-er pronominalizes a certain part of the nominal structure. This means that er is base-generated as part of the noun phrase and subsequently moved to a NP-external position: eri ... [Num/Q [ti]]. That the movement is obligatory can be attributed to the more general properties of discourse-linked pronouns. For instance, the examples in (330) show that weak definite pronouns like ’m are obligatorily scrambled to the left of clause adverbs such as waarschijnlijkprobably; cf. Section 22.1.3, sub IB, for further discussion.

The movement analysis is also similar to the pro-analysis in that it accounts for the fact that Q-er must precede the associate noun phrase when the latter is in the middle field of the clause. Because er is extracted from the noun phrase, the unacceptability of (331b) follows from the independently motivated restriction that a phrase from which an element has been extracted is frozen in its middle-field position; cf. Jan heeft eri lang [ti op] gewacht Jan has waited a long time for it versus *Jan heeft [ti op]j eri lang tj gewacht.

As before, the acceptability of topicalization of the remnant of the noun phrase in (328) requires an appeal to the notion of reconstruction. The clause-boundedness of the relation between Q-er and its nominal associate, illustrated in (329), now follows from the fact that scrambling is clause-bound (just like anaphor binding), which may actually be an improvement over the pro-analysis in the previous section, since it is not at all obvious that pro should be analyzed as an anaphor.

The claim that Q-er is pronominal and thus has referential properties is supported by the examples in (332). Example (332a) first shows that Q-er associated with a direct object can easily precede an indefinite indirect object, while (332b) shows that this is excluded if the indirect object contains a noun phrase denoting the same entities as the quantified noun phrase; cf. Coppen (1991:§2.4) and Kranendonk (2010:§2). This would follow from binding condition C if er binds the noun phrase embedded in the indirect object. The unacceptability of (332b) thus supports the claim that Q-er is referential.

Coppen and Kranendonk also provide a number of strong arguments for the movement analysis. They note that the restrictions on the relation between Q-er and its nominal associate are similar to those between a moved element and its trace. First, the examples in (333b&c) show that Q-er cannot be associated with a single noun phrase in a coordinate structure, while (333d) shows that er is possible if it is associated with both noun phrases. Section C38.3, sub II, shows that similar facts have been described for movement; cf. coordinate structure constraint and across-the-board movement.

Second, the relation between Q-er and its nominal associate seems to be sensitive to islands for extraction. Consider the examples in (334), which show that R-extraction is possible from PP-complements but not from time adverbials.

The examples in (335) show that we find the same thing with Q-er; er can be associated with a noun phrase in a PP-complement, but not in a time adverbial. The percentage sign indicates that some speakers of the southern variety of Dutch do not accept this construction; cf. Smessaert (2014:86).

The examples in (336) further show that both R-extraction and the association of Q-er require that the PP-complement precede the verb in clause-final position. Note that Ik heb er lang gepiekerd over twee may be marginally acceptable with a locational reading of er, but this is not relevant for our present discussion.

A third and final argument is that Q-er constructions exhibit the typical A'-movement property of licensing parasitic gaps; cf. Section V11.3.7 for a detailed discussion. Movement of de boeken across the infinitival adjunct clause in (337a) can license the parasitic gap e; example (337b') shows that the assumed extraction of Q-er results in a configuration similar to that in (337a) and can also license a second, parasitic gap; (337b'') is added to show that the gap e does indeed depend on Q-er extraction from the noun phrase.

The discussion above has shown that the movement analysis has much to commend itself, but it should be noted that there are also problems. The most important one is related to structure (335a'), which assumes that Q-er can be extracted from the nominal complement of a PP. However, there are reasons to think that this is usually not possible. First, consider the examples in (338), which are given only to show that R-extraction from the voor-PP is possible.

Now consider example (339a), which differs from (338a) in that the preposition voor takes the PP-complement bij de koffie instead of the nominal phrase de lunch. Example (339b) shows that R-extraction from the voor-PP is not possible.

Note that R-extraction from the voor-PP leads to a configuration similar to that proposed for the Q-er construction in (336b), in the sense that a moved element is related to a trace within a complement of a preposition. The difference in acceptability between (339b) and (335a') therefore casts some doubt on the movement analysis of Q-er, unless it can be reduced to some independent reason. One approach would be to assume that the unacceptability of (339b) is due to the fact that the embedded bij-phrase is an island for extraction, but this does not seem to be the case since example (340) shows that R-extraction from the bij-phrase is possible as long as the R-word remains within the voor-PP.

Since we have no further insight to offer here, we will leave this problem for future research.

So far, this section has shown that Q-er can be associated with an interpretive gap within a noun phrase modified by a cardinal numeral or a quantifier; we have identified the gap as the trace of Q-er. There are also cases where an NP-internal gap is not associated with Q-er. Some examples are given in (341); adding Q-er to the second conjuncts leads to unacceptable results.

The second conjuncts in (341) are reduced clauses derived by gapping, so one might assume that the gapping operation also triggers obligatory deletion of Q-er. The prediction would then be that Q-er should appear in the non-gapped counterparts of these sentences. The examples in (342) show that this prediction is correct for the cases with a numeral or a quantifier, but not for the case with a demonstrative pronoun.

The unacceptability of (342c) is not surprising, since noun phrases with a demonstrative never occur with Q-er. It might then be attractive to assume that the demonstrative pronoun die in the (c)-examples is not a modifier of a larger noun phrase but an independently used argument; cf. Section 19.2.3.1. However, the examples in (343) show that this solution is not viable either, because the alleged demonstrative argument can be followed by an attributively used adjective.

Nominal constructions of the type in the second conjuncts of the examples in (343), consisting of a determiner followed by an attributive adjective, are generally assumed to be derived by so-called N-ellipsis, i.e. [e] represents a noun that can be elided because its content is recoverable from the context. If this is a viable assumption, there is no a priori reason not to apply a similar analysis to the cases in (342a&b). We are then faced with two competing analyses for these examples: we must leave the question of which is the better one to future research. We refer the reader to A5.4 for further discussion of N-ellipsis.