The examples in (220a&b) show that direct objects can undergo the wat-voor split, provided that the remnant is left-adjacent to the verb(s) in clause-final position. Scrambling of the remnant blocks the split, as the resulting structure in (220b) is an instantiation of the freezing effect. PP-over-V also leads to a degraded result: Dutch speakers may differ somewhat in their judgments of (220c), but all agree that it is marked compared to (220a).

The only elements that may intervene between the remnant and the clause-final verb(s) are phrases competing for the same position: (221a&b) provide examples with a verbal particle, aan, and a PP-predicate, in de kast, respectively. The latter example is perhaps slightly marked, but certainly not unacceptable. Note that when R-extraction has taken place from the PP-predicate, as in (221b'), the result is perfectly acceptable.

The examples in (222) show that reversing the order of the wat-voor remnant and the particle or (stranded preposition of) the PP-predicate yields an unacceptable result.

The data concerning the nominative subject of the clause is more complex than the data concerning the direct object. The following subsections will show that the wat-voor split is allowed in passive constructions and in clauses with an unaccusative verb, provided that the subject occupies its base position and not the (derived) subject position. If the construction contains a transitive or intransitive verb, the split seems to be possible only in expletive constructions.

The primed examples in (235) show that wat-voor split of nominal indirect objects always leads to a degraded result; note that some speakers consider the primeless examples somewhat degraded (while a prepositional indirect object seems to be preferred by most speakers).

As shown earlier in (203), repeated here as (236), wat-voor split of the complement of a preposition is also excluded due to the fact that subextraction from a nominal complement of a preposition is generally excluded.

Note that the wat-voor split differs in this respect from the exclamative wat-construction discussed in Section 15.2.2.1, sub IV. The two (a)-examples in (237) suggest that this construction is similar to the wat-voor construction: the fact illustrated in (237a) that wat and its associated noun phrase can precede the finite verb in clause-initial position suggests that the two form a constituent, and the availability of the split pattern in (237a') suggests that wat can be extracted from this constituent by wh-movement. However, such a movement analysis of (237a') runs into problems with (237b): since subextraction from a nominal complement of a preposition is normally excluded, the movement analysis wrongly predicts that this example is unacceptable.

We must therefore conclude that the wat-voor split and the split found in exclamative wat-constructions are not of the same kind.

Wat-voor split of a nominal predicate is perfectly acceptable, as shown in (238).

If wat is an argument in its own right, it can license a so-called parasitic gap in the infinitival adverbial phrase [zonder ... te lezen] in (239a). The complement of lezen need not be expressed overtly, but can be expressed by a phonetically empty parasitic gap PG, the content of which is identified by the moved wh-phrase (which is indicated by the subscript “i”). In other words, the interpretation of this example is something like “for which x, Jan threw x away without reading x”. As shown in (239b), a parasitic gap can also be licensed by moving a wat-voor phrase as a whole into the clause-initial position.

The N1 wat of the wat-voor phrase, on the other hand, cannot license such a parasitic gap: it cannot license a parasitic gap functioning as the direct object of the infinitival verb lezen (cf. (240a)), nor a parasitic gap functioning as the N1 in a wat-voor phrase functioning as the direct object of lezen (cf. (240b)). It has been suggested that this is due to the fact that parasitic gaps can only be licensed by arguments.

For completeness’ sake, note that some speakers consider example (241b) to be acceptable as well. If so, this example poses a problem for the earlier claim that scrambling induces a freezing effect. Since it is generally assumed that Dutch parasitic gaps must be licensed by a wh-moved or scrambled phrase (cf. Bennis & Hoekstra 1984), it would follow that the wat-voor phrase in (241b) is scrambled, and consequently a freezing effect is incorrectly predicted to arise.

In this context it should also be mentioned that Beermann (1997) claims that in German one occurrence of wat can license the gaps in two or more wat-voor phrases. Example (242a) shows that this is not possible in Dutch. In fact, the examples in (242b&c) show that wat-voor split yields degraded results in such examples anyway; the only fully acceptable option is to move the whole subject into the clause-initial position, as in (242d).

The previous subsection has shown that the interrogative N1 wat does not function as an argument; only the full wat-voor phrase functions in this way. This subsection shows that this conclusion is supported by considerations concerning the so-called intervention effect. Arguments and non-arguments differ in that the latter are more sensitive to such effects than the former. For example, as shown in (243), an interrogative direct object can be moved across the negative adverb niet, whereas an interrogative adverbial phrase of manner cannot. Below we will see that N1 wat behaves like a non-argument in that it cannot cross certain adverbials, i.e. wat-voor split is sensitive to the presence of such adverbials.

The examples in (244) show that the presence of time and place adverbs like gisterenyesterday and daarthere have no effect on the wat-voor split. The split is possible as long as the remnant follows the adverb.

The situation is different with manner adverbs like zorgvuldigcarefully, modal adverbs like zekercertainly, frequency adverbs like vaakoften, or the negative adverb nietnot; the (a) and (b)-examples in (245) to (247) show that these adverbial phrases allow movement of the whole wat-voor phrase, but block the wat-voor split. Perhaps the (b)-examples are slightly better when the wat-voor remnant precedes the adverbial phrase, as in the (c)-examples, but they still seem to be severely degraded. Note in passing that if we were to claim that the (c)-examples are grammatical, they would create a problem for the freezing principle similar to that discussed in relation to example (241b).

The examples in (248) show that in the context of long wh-extraction, wat-voor split can also be blocked by negation in the matrix clause. The (a)-examples first show that long wh-movement is possible with both the unsplit and the split pattern. The (b)-examples show that while long wh-movement of a wat-voor phrase across negation is reasonably acceptable, long wh-movement of wat in isolation leads to a severely degraded result. This again shows that N1 wat is similar to adverbial phrases, which cannot be extracted from embedded clauses either when the matrix clause contains negation.

Wat-voor split can lead to differences in meaning if the sentence contains a universal quantifier like iedereeneverybody. Consider the examples in (249). Although the judgments are apparently not as sharp for all speakers, it seems that the preferred answer to (249a) involves the characterization of a kind of book, e.g. a textbook on linguistics: it is a textbook on linguistics that everyone has read. The preferred answer to (249b), on the other hand, involves what is called a pair-list reading: Jan read a textbook on linguistics, Peter read a novel, and Marie read a study on biochemistry. This difference in meaning is sometimes expressed by assuming that the scope of the universal operator with respect to the question operator is different in the two examples: in (249a) the question operator has wide scope, while in (249b) it has narrow scope.

The difference between the two examples can be highlighted by modifying the universal quantifier iedereen with the adverb vrijwelnearly, as in (250). This modifier blocks the pair-list reading (because it leaves unspecified which entities must be excluded from the answer list), and as we can see in (250b), the wat-voor split now leads to a severely degraded result. Assuming that this is due to a semantic anomaly, it clearly shows that only the pair-list reading is available for constructions like (249b) and (250b). It may be, however, that (249a) is truly ambiguous and also allows the pair-list reading, but there seem to be no syntactic arguments to justify such a view.

The examples in (251) show that the presence of an indefinite argument with an existential interpretation can also severely hinder the realization of a wat-voor phrase, either split or unsplit. On the other hand, if the indefinite noun phrase is generic, the result is perfectly acceptable, as shown in (252).

Note that the contrast between the examples in (251) and in (252) does not only apply to wat-voor phrases; for example, if we replace the wat-voor phrase in (251a) with the wh-phrase welke jurkwhich dress the result is still unacceptable. The unacceptability of non-specific indefinite subject DPs in wh-questions is due to the fact that they simply do not provide the hearer with enough information to answer the question adequately; in order to answer a question like (251a) properly, the hearer must at least be able to establish the identity of the woman involved in the relevant event of wearing a dress.