A first argument for the claim that fragment wh-questions are clauses is based on the selection restrictions that the verb imposes on its complements; embedded fragment wh-questions can only occur with predicates that select interrogative clauses. The primeless examples in (257) show that verbs like wetento know and ziento see can take an interrogative clause and the primed examples show that they can also take an embedded fragment wh-question. Examples such as (257a') are particularly revealing because the verb wetento know can only be combined with a very limited set of noun phrases, and noun phrases referring to objects are certainly not part of this set (unlike its English counterpart to know): cf. Ik weet het antwoord/*dat boekI know the answer/that book.

The examples in (258) show that verbs such as bewerento claim, which do not select interrogative clauses, cannot be combined with fragment wh-questions either.

A second argument for assuming that fragment wh-questions are clauses can be based on coordination: since coordination is normally restricted to phrases of the same categorial type, the fact that full clauses can be coordinated with fragment wh-questions suggests that the latter are also clauses.

A third argument is based on case assignment: the wh-phrase constituting the overt part of the fragment wh-question in (260a) is assigned the same case as the corresponding phrase in the antecedent clause, rather than the case normally assigned by the embedding predicate. Note, however, that such cases can be misleading, since they may involve N-ellipsis in addition to sluicing; this is shown by the fact that the possessive pronoun wiens in (260b) has no syntactic correlate in the antecedent clause, whereas the noun phrase wiens auto does.

Since Dutch has overt case marking only on pronominal possessives, we have no better evidence than cases such as (260), but Merchant (2001/2006) provides a number of examples from German (and other languages) that involve nominal arguments. Although the verb wissento know governs accusative case, the wh-phrase that forms the fragment wh-question in (261) has dative case, as does the complement of the verb schmeichelnto flatter in the antecedent clause.

The main argument for claiming that fragment wh-questions are clauses concerns the syntactic distribution of embedded fragment wh-questions like Wie?Who? and Wat?What?. If such fragment wh-questions were noun phrases, we would expect them to have the distribution of nominal phrases and thus to appear before the clause-final verbs. In the other hand, if such fragment wh-questions were clauses, we would expect them to occur in the normal position of clauses, i.e. after the clause-final verbs. The fact that the wh-fragment wat must follow the verb weet in (262) is therefore conclusive for showing that fragment wh-questions are clauses.

The examples in (263) show that, like regular object clauses, fragment wh-questions functioning as direct objects can only occur to the left of the clause-final verbs if they are topicalized or left-dislocated. This is shown in the second conjunct of (263b) with the preposed fragment wh-clause wat.

Another argument concerns the distribution of the anticipatory pronoun het. We would expect this pronoun to be possible when fragment wh-questions are clauses, but not when they are of a nonverbal category. The examples in (264) show that the results are somewhat mixed: the (a)-examples show that fragment wh-questions functioning as objects cannot co-occur with the anticipatory pronoun het, whereas the (b)-examples show that fragment wh-questions functioning as subjects can.

One possible explanation for the contrast between the two primed examples in (264) might be that fragment wh-questions are always part of the focus (new information) of the clause, as evidenced by the fact that they are always assigned contrastive accent. Section 5.1.1, sub III, has shown that the anticipatory object pronoun het tends to trigger a presuppositional reading of the object clause; so it may be that its combination with a fragment wh-question results in an incoherent information structure, which may account for the judgment in (264a'). Although Section 5.1.3, sub III, has shown that the anticipatory subject pronoun het can also trigger a presuppositional reading of the subject clause, we have seen that there are many cases in which this effect does not occur; this means that the information structure of example (264b') may be fully coherent regardless of whether the anticipatory pronoun is present or not. We leave it to future research to examine whether this account of the contrast between the two primed examples in (264) is tenable, but for the moment we conclude that the acceptability of the anticipatory pronoun het in examples such as (264b') supports the claim that fragment wh-questions are clauses.

The argument on the basis of the anticipatory pronoun can be replicated in a somewhat simpler form on the basis of left-dislocation constructions such as (265); the primed examples show that the resumptive pronoun datthat is possible with fragment wh-questions, regardless of the syntactic function of the latter.

The possibility of left dislocation strongly disfavors a nominal analysis of fragment wh-questions. First, example (266) shows that left dislocation is normally excluded with wh-phrases.

Second, the primeless examples in (267) show that resumptive pronouns normally exhibit number agreement with left-dislocated noun phrases, whereas the primed examples show that left dislocation of fragment wh-clauses involves the invariant form datthat, i.e. the form usually found with left-dislocated clauses.

Nominalization also provides evidence for the claim that fragment wh-questions are clauses. The (a)-examples in (268) first show that nominal objects of verbs normally appear as van-PPs in the corresponding nominalizations; cf. Section N15.2.3.2. The (b)-examples show that object clauses are not preceded by van in nominalizations. The fact that the same is true for the fragment wh-questions in the (c)-examples thus shows that fragment clauses are not nominal but clausal.

The final argument again concerns fragment wh-questions acting as subjects. If fragment wh-questions are clauses, we would expect finite verbs to always have (default) singular agreement; if they are not, we would expect finite verbs to agree in number with nominal fragment wh-questions. The examples in (269) show that the former prediction is correct; finite verbs are always singular, even when the fragment wh-question is plural, as here, where welke is interpreted as “which ones”.

A first problem to consider is that sluicing is generally impossible outside the domain of fragment wh-questions. This is illustrated in the (a)-examples in (271): the first conjunct Jan is hier cannot give rise to sluicing in the declarative object clause in the second conjunct, although it can be pronominalized by the pronoun het/dat. The same thing is illustrated in the (b)-examples with an embedded yes/no question. The unacceptability of the primeless examples shows that we need to formulate certain non-trivial conditions on the use of sluicing in order to ensure that it gives rise only to fragment wh-questions.

For completeness’ sake, example (272b) shows that sluicing is not possible in the domain of wh-exclamatives either.

A second problem to be solved is that fragment wh-questions cannot normally contain material that is not part of the wh-phrase. Some speakers of Dutch allow the overt realization of the complementizer of in embedded clauses, but contrary to what would be expected on the basis of the analyses in (270), example (273b) shows that the complementizer does not appear in embedded fragment wh-questions.

The examples in (274) further show that the standard analysis that finite verbs occupy the C-position in interrogative main clauses would incorrectly predict that non-embedded fragment wh-questions like Wat?What? should contain a finite verb.

Of course, the ungrammaticality of the two (b)-examples above can be solved by assuming that sluicing affects the sequence C + TP, but this assumption is less desirable given that deletion and pronominalization usually involve maximal projections. If we want to stick to this standard assumption, the analyses in (270) require additional stipulations; cf. Merchant (2001:281ff).

The third problem is in a sense the opposite of the second: if sluicing involves deletion or pronominalization of the TP projection, we incorrectly predict that TP-internal material will never surface. A first case that proves this wrong has to do with multiple questions. Example (275a) shows that, like in English, Dutch multiple questions allow at most one wh-phrase in the CP projection; we therefore predict that fragment wh-questions also consist of at most one wh-phrase. However, the (b)-examples in (275) show that the presumed TP-internal wh-phrase in multiple wh-questions must be expressed overtly in fragment questions, i.e. omitting it leads to unacceptability; cf. Merchant (2001:285ff).

It may be relevant in this context to note that although multiple questions can easily be main clauses, non-embedded multiple fragment wh-questions are very marked. This is illustrated by the examples in (276).

A second case, not mentioned in Merchant (2001/2006), concerns constructions with floating quantifiers. The examples in (277) show that although the Dutch floating quantifiers nog meerelse and allemaalall must appear TP-internally in embedded interrogative clauses, they still seem to survive sluicing. Note that this problem does not arise in English, where the quantifiers can be adjacent to the wh-phrase in regular wh-questions; cf. Merchant (2006:122).

The primed (a) and (b)-examples in (278) show that we find the similar facts in main clauses. In fact, the primed (c)-examples seem to show that it is even possible to construct fragment wh-questions with adverbial-like material in such cases.

The examples in (279) show that the same is true for the negative adverb niet and its affirmative counterpart wel; such examples are very common.

The facts described above are very similar to what we find in so-called gapping constructions, i.e. elision constructions with two (or more) remnants, such as Ik denk [dat Jan morgen het boek leest] en [dat Marie morgen het artikel leest] I think that Jan will read the book tomorrow and Marie the article; cf. Section C39.2, sub IID, for a detailed discussion and examples. This similarity could be explained by assuming that sluicing are is a special subcase of gapping with a single remnant, i.e. by assuming that gapping leaves one or more remnants rather than two or more; cf. Broekhuis (2018b) and Broekhuis and Bayer (2020) for relevant discussions.

We conclude our list of possible problems with the fact that has received the most attention in the literature, viz. that sluicing is not island-sensitive. In short, the problem is that there are fragment wh-questions for which it is not immediately clear that they can be derived by wh-movement followed by TP deletion, because wh-movement is blocked in the corresponding non-reduced wh-questions. Consider first the examples in (280a&b), which show that relative clauses are islands for wh-extraction. If non-embedded fragment wh-questions are derived by deleting the TP of the matrix clause, we expect that (280c) could not be used to inquire further about the nature of the thing stolen, but this is clearly wrong, since this fragment wh-question would be a perfectly natural response to the assertion in (280a).

The examples in (281a&b) illustrate the so-called coordinate structure constraint, according to which wh-extraction from a coordinate structure is impossible. If non-embedded fragment wh-questions were derived by deleting the TP of the matrix clause, we would expect that (281c) could not be used to ask who the second person involved was; again, this is clearly wrong because this fragment wh-question would be a natural response to the statement in (281a).

The examples in (282) illustrate the so-called wh-island constraint, according to which wh-extraction from an embedded interrogative clause is impossible. We see again that fragment wh-questions are not sensitive to this type of island, as the fragment wh-question in (282c) would again be a perfectly natural response to the sentence in (282a).

The examples in (283) show that while wh-movement from adverbial clauses is impossible, the fragment wh-questions is still a perfectly natural response to what is asserted in (283a).

Examples (280) to (283) make it patently clear that fragment wh-questions are not sensitive to islands for wh-extraction. Although Merchant (2001/2006) mentions many more cases, we will add a slightly more problematic example of potential island insensitivity, involving the extraction of attributive modifiers from noun phrases. Although such extractions are normally not possible in Dutch wh-questions, fragment wh-questions consisting of APs that correlate with attributive modifiers in their antecedent clause are usually judged acceptable by Dutch speakers; they generally prefer Merchant’s example in (284b) to the one in (284b'), in which the full noun phrase is pied-piped (and the noun man may be omitted as a result of N-ellipsis).

However, a possible problem with the wh-movement + TP deletion approach is that the extracted adjective, which is supposed to have an attributive function, does not have the expected attributive -e inflection. Moreover, some of our informants indicate that even the use of the non-inflected adjective in (284b) is marked (just like the German speakers consulted by Merchant). It is therefore not entirely clear whether it is fully justified to use examples such as (284b) as an illustration of the island insensitivity of sluicing.

To conclude our discussion of the island insensitivity of sluicing, note that Merchant found one case in which Dutch fragment wh-questions seem to be island-sensitive: fragment wh-questions obey the language-specific constraint on preposition stranding. First, the examples in (285b) show that wh-movement of wh-phrases from PPs is normally impossible in Dutch. Preposition stranding is only possible (and actually preferred) when we are dealing with a pronominal PP, i.e. a PP consisting of an R-word and a preposition, like waarnaarto what in (285c); we refer the reader to Chapter P36 for a detailed discussion of this.

If the formation of fragment wh-questions were not island-sensitive, we would expect that none of the sluiced counterparts of the questions in (285) would need to include the preposition. However, the examples in (286) show that the preposition is preferably present when the question word is a pronoun and, perhaps more surprisingly, obligatory when the question word is an R-word. We refer the reader to Subsection E below for a discussion of one notable exception to the generalization that the wh-remnant preferably includes the preposition.

Another problematic case for assuming that the formation of fragment wh-questions is not island-sensitive is given in (287) and involves adverbial degree modification. First, consider the (a)-examples, which show that degree modifiers such as hoehow must pied-pipe the adjective kwaad in regular wh-questions. The fact that the adjective kwaad cannot be omitted in the corresponding fragment question in the primed example would be surprising if fragment wh-questions were not island-sensitive. The (b)-examples provide somewhat more complex cases in which the element hoehow is part of the more elaborate degree modifier hoe zwaarhow very, which in turn can be extracted from the adjective hoe zwaar behaard “how very hairy’.

The discussion above has shown that the arguments for and against the claim that the formation of fragment wh-clauses is island-sensitive are contradictory. This suggests that the underlying assumption that the fragment clause must be isomorphic to the antecedent clause needs further scrutiny; Subsection C will show that this cannot be fully maintained.

Dutch is well suited to illustrate that the phonetically empty TP is not syntactically isomorphic to the TP of its antecedent clause because of the verb-second phenomenon found in main clauses: while finite verbs are clause-final (=TP-internal) in embedded clauses, they are in second position (which is standardly assumed to be the C-position) in interrogative main clauses. Consequently, if the phonetically empty TP in a fragment wh-question must have the same syntactic structure as the TP of the antecedent clause, we would expect that embedded fragment wh-questions can only take an embedded clause as their antecedent, while independent fragment wh-questions can only take a main clause as their antecedent. The examples in (288) show that this expectation does not come true: the main clause Er is iemand in de kamerthere is someone in the room in the (a)-examples can be the antecedent of both independent and embedded fragment wh-questions, and the same holds for the embedded clause dat er iemand in de kamer isthat there is someone in the room in the (b)-examples.

The examples in (288) have shown that there is no syntactic isomorphism between the fragment wh-question and the antecedent clause. Example (289) shows that they do not require isomorphic semantic representations either. For example, the phonetically empty TP in the fragment wh-question is not interpreted as containing the modal verb willento want that we find in the antecedent clause; rather, it is interpreted as having a modal meaning normally expressed by kunnencan or moetenmust. A similar example can be found in (290a).

An example of a more complex nature is (291). In response to a pupil’s remark in (291a), a teacher can say (291b), in which it is clear that the omitted part cannot be isomorphic to what the pupil said, as the anaphor mezelf cannot be bound by the interrogative pronoun wie: cf. Wie heeft zichzelf/*mezelf nog niet opgegevenwho has not yet registered?.

The fact that semantic isomorphism need does not have to hold completely does not mean that anything goes, because the propositional content of the fragment wh-question is depends on the propositional content that we find in the antecedent clause. This is illustrated in example (292), which shows that minimally the proposition expressed by the main verb and its arguments must be preserved in the interpretation of the fragment wh-question; cf. Merchant (2006: §1) for an attempt to formally define the term closeness in meaning.

The fact that the fragment clause is not fully isomorphic with its antecedent can be seen as an argument for the proposal in Subsection B3 that it is a subcase of gapping, since the latter also allows for certain differences between gapped clauses and their antecedents. This is illustrated in (293) for the form of the finite verb (i.e. kom-Ø versus komt-t); cf. Section C39.2 for a more detailed discussion of gapping.