Support for the assumption that the wat-voor string is a complex modifier comes from the fact that een does not behave like a regular indefinite article, which is clear from the fact, illustrated in (207a), that it can precede both singular and plural N2s, whereas indefinite articles preceding a plural noun usually have a null form. As a matter of fact, it may be that the null form can also occur in the wat-voor construction (alternatively, of course, one can assume that there is no article present at all), but the data in (207b) then shows that this null form is not restricted to plural noun phrases, as would normally be the case.

It is not entirely clear whether een can also precede N2 if the latter is an existential pronoun (which would normally be excluded: *een iets/iemand). Our intuition is that this is impossible if N2 is the [-human] pronoun ietssomething, but at least marginally possible if it is the [+human] pronoun iemandsomeone. This intuition seems to be confirmed by a Google search (June 2008): while the search for the string [wat voor een iets] yielded only 3 wat-voor constructions, the search for [wat voor een iemand] yielded 17 results. It should also be noted that in most of these cases the wat-voor phrase was used as the predicate in copular constructions like wat voor een iets is dat?what kind of thing is that? and Wat voor een iemand ben jij?What kind of person are you?

Another argument for the idea that een is a spurious article is that it cannot be replaced by any other determiner or other element that may occur in the left periphery of the noun phrase; for example, replacing een with a definite article or a numeral leads to an unacceptable result.

Note, however, that there is one seeming counterexample to the claim that N2 cannot be preceded by a numeral, viz. constructions in which an empty N2 is licensed by quantitative er, as in (210). Een, which is normally pronounced with a schwa, must be pronounced like the numeral éénone, /e:n/. However, since één cannot be replaced by a numeral like drie, it seems plausible that the occurrence of één in (210) is due to the fact that the empty noun must be preceded by an element carrying accent. Note that examples such as (210a) also occur without er: we found several various uses of Wat voor een wil je (hebben)? on the internet.

Some speakers also allow examples such as (210a) without the presence of een, as shown in (211a). The split pattern in (211b), on the other hand, is consistently judged unacceptable, which may be related to the fact that the phonetic string in (211b) has a more prominent reading, in which er ... voor functions as a pronominal PP: Wat wil jij ervoor hebben?What do you want for it?. Examples such as (211a) also occur without er: we found several instances of Wat voor wil je (hebben)? on the internet.

The discussion in Subsection A suggests that een is a spurious indefinite article. Similarly, the preposition voor may not be a true preposition, which is suggested by the fact that it does not assign case. Unfortunately, this cannot be shown for Dutch, since it lacks morphological case, but we can show it for German. While the German preposition für normally assigns accusative case, it does not assign accusative case to N2 in the was für construction. Instead, the case of N2 depends on the case of the complete was für phrase: if the was für phrase is a subject, N2 is assigned nominative case; if it is a direct object, it is assigned accusative case; and if it is the complement of a preposition like mitwith, it is assigned dative case. This is shown in (212).

Another fact that might be taken to show that voor is not a true preposition is that the string voor + noun phrase cannot undergo R-pronominalization, which is normally possible with voor-PPs.

The conclusions in Subsections A and B that een is a spurious article and that voor is neither a true preposition could be seen as supporting the assumption that wat voor een is a complex modifier that is part of the lexicon as such: the availability of the string wat voor could then be accounted for by assuming that it is a reduced form of wat voor een. However, analyses that adopt this assumption run into problems with the wat-voor split. If wat voor (een) is a complex modifier, the examples in (214) would violate the lexical integrity constraint, according to which parts of lexical items cannot undergo syntactic processes: in these examples, wat is extracted from the lexical modifier wat voor (een). The assumption that wat voor (een) is a complex modifier therefore forces us to introduce additional mechanisms to allow the violation of this constraint; cf. Corver (1990/1991) for a good overview of several proposals from the literature.

Alternatively, it has been argued that wat should be considered a nominal predicate, i.e. the wat-voor construction should be analyzed like the N-van-een-N construction in Section 18.2.1. Since arguing for this would lead us into a thicket of theory-internal issues of generative grammar, we cannot go into this matter here; for a discussion of this analysis, see Den Dikken (1995b) and Bennis et al. (1998), who provide more or less similar analyses for the two constructions in question.