Section 11.3.1.2, sub IV, has shown that long wh-movement is normally excluded from factive clauses. This is illustrated again in example (244b): while long wh-movement is perfectly acceptable with the non-factive matrix verb denkento think, it leads to a degraded result with the factive matrix verb wetento know. Note, however, that some speakers do allow long wh-movement when the wh-phrase is D-linked, such as welk boekwhich book in (244b'). Recall that we do not include the intermediate trace in the initial position of the embedded clause if it is not immediately relevant to our discussion.

The percentage sign in (244b') indicates that judgments vary from speaker to speaker and from case to case; the latter can be seen from the fact that the degraded (b)-examples in (244) improve for many speakers when we substitute the factive verb betreurento regret for wetento know, as in (245).

That there is a great deal of speaker variation is evident from the fact that the judgments on examples such as (245a) found in the linguistic literature also vary considerably: some researchers reject such examples as completely ungrammatical (e.g. Hoeksema 2006:147), while others accept them as perfectly acceptable (e.g. Bennis 1986:104) or suggest an intermediate status (Barbiers 1998). The diacritics here should not be seen as the expression of absolute but of relative judgments: the use of a double question mark instead of an asterisk in (245a) at least does justice to the fact that this example deteriorates when the anticipatory pronoun hetit is added (cf. *Wati betreurde Jan het [dat Marie ti gekocht had]?) and that it is less felicitous than examples such as (245b), which involve the extraction of a D-linked wh-phrase. Note also that examples such as (245b) are sometimes given as perfectly acceptable (e.g. in Zwart 2011:209), but since at least some speakers feel uncomfortable with them, we have added a question mark.

The crucial point for our present discussion is that the acceptability contrast between long wh-movement of non-D-linked and D-linked wh-phrases from factive complements is beyond doubt. This contrast shows that certain islands are not strong (absolute), but weak (selective) in the sense that they block wh-extraction of certain elements but not others. It is often claimed that weak-island violations are sensitive to the referential properties of the wh-phrase, in that extraction is only possible if the descriptive part of the wh-phrase denotes a certain pre-established set of entities in the domain of discourse; cf. Szabolsci (2006: §5) and the references cited there. D-linked wh-phrases such as welk boekwhich book satisfy this criterion, while the non-D-linked pronouns wiewho and watwhat usually do not, and at best presuppose the existence of some entity that satisfies the description of the predicative part of the question. Example (246b) shows that weak islands normally also block long wh-movement of non-arguments like adverbial adjuncts (but see Szabolsci 2006 for some exceptional cases).

Wh-extraction is not possible from embedded interrogative clauses: this holds for polar yes/no questions as well as for wh-questions. That yes/no questions are islands for question formation is illustrated in (247b); the fact that the wh-phrase welk boekwhich book is D-linked shows that such islands are strong.

Although examples such as (247b) are not often discussed explicitly, their degraded status can be easily accounted for by assuming that the clause-initial position of the embedded clause is not accessible to the wh-phrase due to the presence of a phonetically empty polar question operator. The postulation of such an empty operator may be necessary anyway to rule out wh-movement in polar main clauses such as (248a); wh-movement is possible only if the position preceding the finite verb is radically empty, which accounts for the fact that (248b) is a pure wh-question that leaves no room for a polar interpretation. For completeness, we have added example (248c) to show that the wh-element cannot remain in situ either.

If the clause-initial position of embedded polar questions is indeed occupied by a phonetically empty question operator, the unacceptability of the long wh-movement in (247b) follows from the standard analysis in generative grammar that wh-extraction cannot occur in a single movement step, but must proceed via the clause-initial position of the object clause. This analysis can be straightforwardly extended to account for the unacceptability of cases like (249), where long wh-movement takes place from embedded wh-questions. Note that (249c) is perfectly acceptable if the adverbial phrase modifies the matrix clause, but this is of course not the reading intended here (as indicated by the trace tj).

Wh-islands have been reported to be weak in many languages, including English. This does not seem to be the case in Dutch, as most speakers seem to consider all examples in (249) to be (equally) bad; cf. Koster (1987:192ff) and Zwart (2011:208). However, Koster (1987:22) claimed that long movement is more acceptable when the wh-phrase in the clause-initial position of the embedded clause is not a subject, as in the examples in (250), to which Koster assigns a mere question mark. It should also be noted that Koopman & Sportiche (1985) has claimed that long wh-movement of PPs in examples such as (250a') is more acceptable than long wh-movement of objects in examples such as (250b'), although Koster (1987) does not seem to agree with this. To our knowledge, such wh-island violations have not been discussed elsewhere, and since their exact status is not clear to us, we simply mark them with a percentage sign.

Long wh-movement is typically associated with extraction from direct object clauses. It is sometimes claimed that long wh-movement is excluded from subject clauses; cf. Huang (1982). Examples given to illustrate this usually concern subject clauses in non-extraposed position or subject clauses introduced by the anticipatory pronoun hetit; cf. Zwart (2011:202ff).

However, Section 11.3.1.2, sub III, has already shown that there are subject clauses in extraposed position that allow long wh-movement, provided that the anticipatory pronoun het is not present. We illustrate this again in (252b) with the passive counterpart of the construction in (252a) with an object clause. The fact that the extracted phrase is the non-D-linked pronoun watwhat shows that subject clauses are not even weak islands.

The fact that long wh-movement from subject clauses is nevertheless rare is due to the fact that such clauses are usually preceded by the anticipatory pronoun het when they occur in extraposed position; cf. Section 11.3.1.2, sub III, for more details.

Adverbial clauses differ from argument clauses in that they are always islands for wh-formation; cf. Huang (1982). This is illustrated in (253) for adverbial clauses indicating time and reason. The fact that the primed examples involve the D-linked phrase Welke foto’swhich pictures shows that adjunct clauses are strong islands for wh-movement.

Section 11.3.1.1, sub VB has shown that, contrary to what is commonly assumed, there are reasons to assume that noun phrases are islands for postnominal wh-phrases. This was argued on the basis of examples such as (254), which show that both the stranding option and the pied piping option are ruled out.

The islandhood of noun phrases for wh-phrases embedded in postnominal clauses such as those in (255) is uncontroversial. This holds regardless of the syntactic status of the postnominal clause: the (a)-examples show this for a clausal complement and the (b)-examples for a relative clause. The fact that the primed examples involve D-linked noun phrases shows that complex noun phrases are strong islands for wh-movement. Note that extraposition of the clauses would not improve the result.

Islands for question formation are usually clausal in nature, the reason being that non-sentential clausal constituents regularly allow stranding or pied piping; cf. Section 11.3.1.1, sub V and VI. We will see, however, that coordinate structures are notable exceptions. First, the examples in (256) show that whole coordinate structures can easily be questioned.

However, it is impossible to question one of the conjuncts: the (a)-examples in (257) show that wh-movement of one of the conjuncts while stranding the rest of the coordinate structure is excluded; the two (b)-examples show that pied piping of the entire coordinate structure is also excluded.

Although it is not entirely clear what the correct representation of “split” coordinate structures like (258a) is, it might be interesting to note that such cases do not allow question formation either.

The examples above have shown that wh-extraction from coordinated structures is not possible. A possible exception is across-the-board movement, which can extract wh-phrases from coordinated structures, provided that all conjuncts are affected in a parallel way. Note that the strikethrough in (259b) is the result of backward conjunction reduction, which need not concern us here.

Across-the-board movement always involves subextraction from a conjunct; it must leave a remnant. This is shown by the unacceptability of examples like (260a). However, it is not clear whether this is due to a syntactic constraint, because example (260b) shows that wh-movement of the full coordinate structure is also impossible. The use of the dollar sign indicates that this may be a simple economy effect, because the answer to Wat heeft Jan gekocht?What has Jan bought? can take the form of a list: Een boek, een plaat, ...A book, a record, ....

Given that the wh-phrase in across-the-board movement constructions is associated with two independent gaps, it is controversial whether the examples in (259) are derived by wh-movement in a standard way. We will not elaborate on this theoretical issue here, but refer the reader to De Vries (2014) for a detailed discussion.

Standard German differs from standard Dutch in that many German speakers do not allow long wh-movement constructions such as (261a). Such speakers may employ various alternative strategies to overcome this problem, one of which is the use of the resumptive prolepsis construction illustrated in (261b), in which a proleptic phrase (here: von welchem Maler) obligatorily binds a resumptive pronoun within the embedded clause; cf. Salzmann (2006) for a detailed discussion.

The resumptive prolepsis construction is not unique to speakers who do not allow long wh-movement, as shown by the fact that in standard Dutch, the two constructions in (262) are possible side by side.

Long wh-movement and resumptive prolepsis constructions exhibit a number of similarities, which we will return to in Section 11.3.7. These might lead one to think that they are both derived by wh-movement (in which case something special should be said about the use of the preposition von/van and the insertion of the resumptive pronoun). However, Salzmann (2006) argues that there are several reasons not to follow this line of thinking. One of the main reasons is that the resumptive prolepsis construction is not sensitive to islands. This is illustrated in (263) for factive islands: while (263a) shows that long wh-movement leads to a degraded result for many speakers, (263b) shows that the corresponding resumptive prolepsis construction is perfectly acceptable, although Schippers (2012:§5.2.3) has shown that it is quite rare in the Dutch part of her corpus.

The assumption that the resumptive prolepsis construction is derived by wh-movement becomes even less plausible when we consider strong islands like the embedded questions in (264). The contrast between the primeless and primed examples shows that while long wh-movement is impossible, the corresponding resumptive prolepsis constructions are again perfectly acceptable.

If wh-movement is not involved in the derivation of the resumptive prolepsis construction, the proleptic phrase must find its origin in the matrix clause. Consequently, the (obligatory) coindexing in the above examples must be due to the usual constraints on binding of referential pronouns, which does not seem to pose any particular problems, since the pronoun is free in its local domain; cf. Section N22.1. An appeal to the normal mechanisms involved in binding would also explain the fact, illustrated in example (265), that the proleptic phrase can serve as an antecedent of two (or more) resumptive pronouns.

A wh-movement approach, on the other hand, would certainly need additional provisos to account for this possibility, because wh-phrases in clause-initial position are usually associated with only a single argument position: e.g. the interrogative pronoun who in (266a) functions as a subject, as shown by the fact that (266b) is a felicitous response to (266a), but it cannot simultaneously function as a subject and an object, as shown by the fact that (266b') is not a felicitous response.

That the proleptic phrase must be independently licensed within the matrix clause may also explain why resumptive prolepsis is especially common with a limited number of predicates, including denkento think, gelovento believe, hopento hope, vermoedento suspect, vertellento tell, vrezento fear, (niet) wetento know (not) zeggento say, and zich afvragento wonder. The unacceptability of example (267b) follows immediately if the predicate vertrekkento leave is not able to license a proleptic van-PP. By contrast, the wh-movement approach to resumptive prolepsis would have to explain why adjuncts differ from embedded questions in this respect, which will be difficult in view of the fact that they both behave as strong islands in other contexts.

For completeness, we conclude by noting that resumptive prolepsis is also possible in constructions such as (268b'), in which the proleptic phrase is associated with the adverbial proform erthere.