2.1 The analysis of wat

The neuter wh-pronoun wat can be many things in Dutch, which suggests that it is not much in itself. It can be an indefinite (16a), a nominal modifier (16b), a relative pronoun (16c), a wh-pronoun (16d) or an exclamative marker (16e).

1. (16)

   1. (a) Jan heeft wat gegeten. Jan has wat eaten ‘Jan has eaten something.’

   2. (b) Jan heeft wat boterhammen gegeten. Jan has wat sandwiches eaten ‘Jan has eaten some sandwiches.’

   3. (c) Alles wat ik ooit dacht te weten. everything wat I ever thought to know ‘Everything I thought I knew’

   4. (d) Wat zal ik vanavond eten?wat shall I tonight eat ‘What shall I eat tonight?’

   5. (e) Wat een ellende is dit!wat a disaster is this ‘What a disaster this is!’

Instead of listing wat in the lexicon five times, we follow Postma (Reference Postma, Bok-Bennema and Cremers1994) in assuming that there is only one wat, whose interpretation is determined by the syntactic context in which it appears (cf. Cheng Reference Cheng1991 for a similar claim for Chinese indefinites). This is illustrated in (17). In its base position, wat cannot be interpreted as a wh-pronoun and is instead an indefinite pronoun. (We ignore echo questions, and cases of multiple wh-questions, where wat can be licensed in situ by a moved wh-operator.) In clause-initial position, by contrast, the indefinite interpretation is not available, and wat is obligatorily construed as a wh-pronoun.Footnote ⁶

1. (17)

   1. (a) Jan heeft wat gegeten. Jan has wat eaten ‘Jan has eaten something.’ not: ‘What has Jan eaten?’

   2. (b) Wat heeft Jan gegeten?wat has Jan eaten ‘What has Jan eaten?’ not: ‘Jan has eaten something.’

For wat to appear in such a variety of contexts, it must be rather unspecified. This indeed seems to be the case. Wat can modify both neuter and non-neuter mass nouns (18), suggesting it is unspecified for gender. It can also modify a plural noun, as in (19), suggesting it is not specified for number. In addition, wat can appear in an expletive construction, which in Dutch triggers a strong definiteness effect. Hence, the grammaticality of (20) suggests that wat is not specified for definiteness either.

1. (18)

   1. (a) het brood / wat brood the_neuter bread wat bread

   2. (b) de kaas / wat kaas the_non-neuter cheese wat cheese

1. (19) wat boek-enwat book-plur

1. (20) Er is wat gekomen. there is wat come ‘Something arrived.’

What features, then, does wat express? We assume a privative system with the features [plural], [non-neuter] and [definite]. For an item to bear no specification for these features means that this item is not endowed with the relevant features. This explains the flexible combinatorial possibilities of wat, i.e. the fact that it can combine with a plural or singular noun. If wat is not combined with anything and appears on its own, it is interpreted by default rules: singular in the absence of [plural], neuter in the absence of [non-neuter], and indefinite in the absence of [definite].

As regards its categorial status, we think of wat as the spell-out of a QP, cf. (15). In short, wat is an indefinite numeral, in close kinship with Dutch veel ‘many’ and weinig ‘little’. Veel and weinig have been analysed as Q-elements in the literature (cf. Corver Reference Corver1997). Informally, the denotations we assign to them are those in (21):

1. (21) veel ‘much’, ‘many’ = high quantity weinig ‘few’, ‘little’ = low quantity wat ‘some’ = quantity

There are three arguments in favour of the suggested parallel. First, all three elements can be modified by adverbs like heel ‘very’ and nogal/best ‘quite’:

1. (22)

   

Secondly, they appear in complementary distribution with elements such as determiners and numerals.

1. (23) *twee veel/weinig/wat boeken two many/few/some books

Thirdly, in noun ellipsis contexts, they obligatorily occur with quantitative er when modifying a count noun (a characteristic property of numerals), but they occur without quantitative er when modifying a mass noun:

1. (24)

   1. (a) (Over boeken): Ik heb *(er) veel/weinig/wat gelezen. about books I have there many/few/some read ‘(As for books) I read many/few/some.’

   2. (b) (Over kaas): Er ligt (*er) veel/weinig/wat in de koelkast. about cheese there is there much/little/some in the refrigerator ‘(As for cheese) there is much/little/some in the refrigerator.’

Interestingly, wat is not always in complementary distribution with veel and weinig, as is evident on the basis of (25a). In such cases it has the same position as the phrasal een beetje ‘a bit’ (cf. (25b)):

1. (25)

   1. (a) Ik heb wat veel/weinig boeken gelezen. I have wat many/little books read ‘I have read quite a bit too many/too few books.’

   2. (b) Ik heb een beetje veel/weinig boeken gelezen. I have a bit many/little books read ‘I have read a little too many/too few books.’

This means that, if veel and weinig are heads of QP (as argued, for instance, in Corver Reference Corver1997), then wat cannot be the head of QP. But this is exactly what we expect if wat is the spell-out of a phrase rather than a head (we will see that the same can be argued for die and wie). If so, wat can appear in a specifier position. Hence, examples like in (25) are represented as in (26), where QP-wat occupies the specifier position of another QP that dominates NP. The head of the latter projection can either be spelled out, e.g. by veel or weinig in (25), or remain empty, as in (19). These possibilities are represented in (26).

1. (26)

   

Since we claim that die and wie also spell out phrases, we expect these pronouns to be able to occur as specifiers too. This seems correct. Possessor wh-phrases in Dutch of the type given in (27b) provide ancillary support for the idea that wh-pronouns reside in the Spec of a DP. There are even languages (like Greek) which fully lexicalize the structure.

1. (27)

   

2.2 The analysis of wie

Following Déchaine & Wiltschko (Reference Déchaine and Wiltschko2002), we assume that DPs have a PhiP-layer, where among other things gender is expressed. It is this layer that can be spelled out by the non-neuter wh-pronoun wie:

1. (28)

   

2.3 The analysis of die

PhiPs can be embedded under a DP-layer. This is what the non-neuter relative pronoun die spells out.

1. (29)

   

The contribution of this layer is definiteness (cf. Bennis Reference Bennis, Dongelmans, Lalleman and Praamstra2001), as schematically indicated in (30):

1. (30)

   1. (a) dat=wat+definiteness

   2. (b) die=wie+definiteness

Recall that wat cannot appear in a topic position without triggering a question interpretation (cf. (31a)). This is in contrast to dat (cf. (31b)), which suggests that the latter is definite.

1. (31)

   1. (a) Hans heeft wat gelezen. → Wat heeft Hans gelezen? Hans has something read something has Hans read ‘Hans has read something.’ *‘Something, Hans has read.’

   2. (b) Hans heeft dat gelezen. → Dat heeft Hans gelezen. Hans has that read that has Hans read ‘Hans has read that.’ ‘That, Hans has read.’

As we would expect on the basis of their ability to function as topics, dat and die can be topic-dropped, as illustrated in (32). This is impossible with wat and wie, given their non-topicalizability.

1. (32)

   1. (a) Wie het weet, (die) mag het zeggen. who it knows d-pron may it say.inf ‘Whoever knows it may say it.’

   2. (b) Wat je weet, (dat) mag je zeggen. what you know d-pron may you say ‘Whatever you know you can tell.’

An apparent problem is the following. If wie is a sub-constituent of die, wie cannot have more features than die. However, die is a pronoun that refers to non-neuter antecedents, whereas wie is one that refers to non-neuter, human entities. It is reasonable, therefore, to ask why die does not pick out strictly human referents. Our response is that the situation is essentially due to an accidental gap. Both wie and die have [non-neuter] as their specification for gender. It so happens that the Dutch lexicon has no PhiP that expresses only [non-neuter]. What is available is a PhiP, spelled out as wie, which is also endowed with specification for [human].Footnote ⁷

To summarize, die, wie and wat are spell-outs of phrases corresponding to different layers of a nominal structure, namely DP, PhiP and QP, respectively. With this in place, we can now go through the derivations of doubling constructions, which is the topic of the next section.

3.1 Partial copying with wh-pronouns

The different patterns arise because of (i) the optionality of generating either a DP or PhiP in the base position and (ii) the optionality of pied-piping DP and PhiP. In this section we will go over all the cases of non-identical doubling and give an account of the ungrammatical ones, i.e. those that violate the generalization in (12).

The most straightforward case is that of full copying. This is unsurprisingly derived by (i) copying of a PhiP, (ii) remerging it higher in the structure and (iii) spelling out two chain links:Footnote ⁸

1. (33)

   

Turning now to partial copying, (34)–(36) schematize the grammatical patterns attested in Dutch dialects. In (34) we copy only the QP from within the PhiP. In (35) partial copying applies to PhiP within a DP, and in (36) partial copying targets the QP inside a DP.

1. (34)

   

1. (35)

   

1. (36)

   

Examples (37)–(39) illustrate what goes wrong in the ungrammatical doubling cases. To derive an example in which the first element is more specified than the second, in violation of our generalization in (12), full copying has to be followed by something that the copy procedure does not itself provide, namely the addition of features and/or structure. We indicate this illegitimate addition by use of a dotted line.

1. (37)

   

1. (38)

   

1. (39)

   

To conclude, by invoking the mechanism of partial copying we can derive the attested examples of non-identical doubling in a straightforward way. The inclusiveness condition suffices to rule out the unattested examples.Footnote ⁹

The analysis proposed here builds on earlier proposals according to which doubling is derived by splitting a big DP, moving up one part and stranding the other. An early example of this is Uriagereka (Reference Uriagereka1995), where clitic doubling in Western Romance is derived from a structure [DP (double) [D clitic [NP pro]]]. The clitic undergoes head movement out of this DP to a clause-initial F position, stranding the double. Another prominent example is Kayne (Reference Kayne, Epstein and Seeley2002), who extends Uriagereka's analysis to account for condition C effects. See also Cecchetto (Reference Cecchetto2000), Belletti (Reference Belletti2005) and Poletto & Pollock (Reference Poletto and Pollock2004) for implementations of the big XP approach. One important difference from such analyses is the spell-out mechanism that we assume here (i.e. a pronoun is the spell-out of the highest functional projection, QP, PhiP or DP), which ensures that in the case of partial copying of wh-pronouns the original is not affected by the partial copying operation but is spelled out as a whole. Another difference is that in our analysis all pronouns are phrasal, i.e. neither the doubler nor the doublee is a head, which is in contrast to the literature on Romance. For an analysis of subject doubling cases like (6) along the lines proposed here, see van Craenenbroeck & van Koppen (Reference Craenenbroeck, van Koppen and Barbiers2008).

A question that this analysis does not address is why wh-pronouns are never spelled out in their thematic base position. Although this question also arises for constructions involving wh-movement but no doubling, it becomes even more prominent with doubling constructions, where we observe that intermediate copies can be spelled out. The most explicit theory dealing with this issue is Nunes (Reference Nunes2004). He proposes that movement as copying in general gives rise to a problem at PF. Consider example (40):

1. (40) X_i … Y … X_i

Here, X has been copied and merged in a higher position. Under the assumption that the two copies are PF-identical, linearization rules, such as Kayne's Linear Correspondence Axiom (LCA), encounter an ordering conflict: X has to be ordered before and after Y at the same time. The general solution is to not spell out one of the copies (see also section 4.1). The copy chosen for deletion is usually the lower one.Footnote ¹⁰ To explain the fact that wh-pronouns can be spelled out in intermediate SpecCP positions, Nunes proposes that these pronouns adjoin to C rather than move to SpecCP. In this position the wh-pronoun undergoes morphological reanalysis with C, i.e. the two become one word for the purposes of morphology. The consequence of this operation is that the wh-pronoun becomes invisible for the LCA, as this linearization procedure cannot look into the structure of words. In short, some varieties allow movement of wh-pronouns to embedded C-positions and the consequence of this syntactic option is doubling. In varieties that move pronouns to embedded SpecCP positions, intermediate copies have to be deleted at PF.

Since in our analysis pronouns are spell-outs of categories that are maximal projections and not heads, movement of the PhiP/DP to a position adjoined to C is not immediately expected. The way to uphold Nunes' insight is to assume morphological reanalysis affecting PhiP/DP in SpecCP and C. For this to be possible, one has to assume that under certain conditions non-terminal nodes can undergo morphological merger. This assumption can be independently motivated. Consider the case of auxiliary contraction, for which there is evidence that it is a morphological process, thus sensitive to syntax and not reducible to phonology (see Goodall Reference Goodall, Everaert and van Riemsdijk2006 for a recent overview). The contrast between (41a) and (41b) shows that contraction is sensitive to the syntactic configuration: the pronoun c-commands the auxiliary in (41a) but not in (41b). Although one could assume that the relevant structural relation is between two terminals, namely D and I, (41c) shows that contraction between the pronoun and the auxiliary is possible even when the pronoun is modified. The terminal D in (41c) does not c-command the auxiliary, suggesting that it is a projection of D that satisfies the c-command requirement on contraction.

1. (41)

   1. (a) You've got a lot of nerve.

   2. (b) *John and you've got a lot in common. (Radford Reference Radford1997: 331)

   3. (c) Not even you've solved this problem.

Since nothing now excludes reanalysis of PhiP/DP with C, the difference between doubling and non-doubling can be characterized in terms of this operation, as in Nunes' proposal.Footnote ¹¹

3.2 Extensions

In the previous sections we have argued that wat is a sub-constituent of wie, and wie a sub-constituent of die, and we have motivated these claims in terms of features. But now consider the examples in (42):Footnote ¹²

1. (42)

   1. (a) Wat denk je hoeveel mensen er op deze afdeling werken? what think you how.many people there in this department work ‘How many people do you think work in this department?’

   2. (b) Wat denk je hoe dat kindje zich dan moet voelen? what think you how that child refl then must feel ‘How do you think this child must feel?’

   3. (c) Wat denk je welke impact het zou hebben? what think you which impact it would have ‘Which impact do you think it would have?’

Can these constructions be analysed as involving partial copying? It is not completely straightforward that wat is a partial copy of hoe or hoeveel: in what way can hoe be decomposed into wat plus something additional, as we did for wie and die? Moreover, under the standard assumption that welke ‘which’ in (42c) is a determiner head (cf. Longobardi Reference Longobardi1994, Corver Reference Corver1997), wat would have to be a partial copy of a head, thereby violating lexical integrity.

Despite initial appearances, we would like to explore the possibility that the examples in (42) involve copying of a sub-constituent of the lower constituent. It proves insightful here to take into consideration degree questions in Scandinavian languages. Besides degree questions in which the whole adverb phrase is fronted (cf. (43a)), Icelandic can split the phrase and leave the adjective in situ (cf. Svenonius & Kennedy Reference Svenonius, Kennedy and Frascarelli2006). The fronted question word is then spelled out as hvaÐ ‘what’ (cf. (43b)).

1. (43)

   1. (a) Hversu gammall ertu? Icelandic how.much old are.you ‘How old are you?’

   2. (b) HvaÐ ertu gammall? what are.you old ‘How old are you?’

Standard Norwegian has the question word kor ‘where’ in constructions parallel to (43a), as given in (44a). Northern Norwegian dialects also have the ‘adjective in situ’ construction, but there is no overt clause-initial question word in this case. Example (44b) is distinguished from a yes/no-question through intonation.

1. (44)

   1. (a) Kor gammel er du? Standard Norwegian where old are you ‘How old are you?’

   2. (b) Er du gammel? Northern Norwegian Are you old ‘How old are you?’

In order to account for the correct semantic interpretation of degree questions, Svenonius & Kennedy argue that the syntactic representation involves a functional head, i.e. degree (or measure), which selects a phrase headed by a gradable adjective. This degree head is interpreted as a variable which is bound by the measure phrase in its specifier:

1. (45)

   

Given this analysis, the variation in degree questions boils down to variation in the spell-out rules.Footnote ¹³ In Standard English, the degree head is spelled out as how and is syntactically bound by an operator, a measure phrase in its specifier which presumably remains empty as a result of a doubly-filled COMP effect (cf. (46a)). Icelandic and Norwegian also have this option (cf. (46b) and (46c)). In addition to fronting the entire DegP to first position, however, Icelandic can also move just the operator, which will then be spelled out as hvaÐ (cf. (46d)). In Northern Norwegian, this operator remains empty (cf. (46e)).

1. (46)

   

Hence, the variation shown in Scandinavian degree questions motivates the presence of an empty question operator which only becomes PF-visible when it is sub-extracted. In Icelandic, PF spells out the operator, and in Northern Norwegian PF signals its presence through intonation.

A similar structure can now be adopted for the Dutch data in (42). The degree phrase occupies an intermediate SpecCP position. From there, the operator is sub-extracted and moved to sentence-initial position, where PF spells it out as wat. The particular spell-out that PF chooses, in Icelandic and Dutch dialects, should not come as a surprise. We argued earlier that wat is a QP denoting unspecified quantity. Hence, it can readily function as an unspecified measure phrase binding the degree variable hoe. For (42b), we assume that the degree head selects an empty adjective phrase. The structures are given in (47):

1. (47)

   1. (a) [[_QP Wat] denk je [[_DegPQP hoe [_AP veel mensen]] er op deze afdeling werken?]]

   2. (b) [[_QP Wat] denk je [[_DegPQP hoe [_AP Ø ]] dat kindje zich dan moet voelen?]]

We would like to suggest that the structure in (48) provides the blueprint for partial copying constructions involving wh-phrases in general. They contain three components: (i) a variable, introduced by a determiner such as ‘how’ or ‘which’ (ii) a range for the variable, introduced by an (empty) NP or AP and (iii) an operator binding the variable (cf. also Wiltschko Reference Wiltschko1998, Rett Reference Rett, Bonami and Cabredo-Hofherr2006Footnote ¹⁴).

1. (48)

   

Hence, the hypothesis that wh-constituents in general contain an operator and a variable provides a source for the category which after sub-extraction gets spelled out as ‘what’. In other words, the internal syntax of wh-constituents accounts for the syntax external to it, namely the partial copying construction.Footnote ¹⁵

If the structure of wh-constituents contains both an operator and a variable, this must be the case not only for wh-phrases but also for wh-pronouns. Indeed, we will assume that PhiP- and DP-internally, the QP moves to the highest specifier position, as indicated for PhiP in (49a) and for DP in (49b).Footnote ¹⁶

1. (49)

   

This move changes the representations of the derivations in section 3.1 slightly. The advantage is that it allows us to draw an interesting parallel with the wat … voor-split, a construction that overtly shows the presence of wat in the higher specifier:

1. (50)

   

In the following section, we discuss parallels between partial copying and the wat … voor-split, suggesting they are basically the same phenomenon.

3.3 The parallel with the wat … voor-split

A crucial ingredient in the analysis is that partial copying should be allowed to target a specifier which is part of a constituent in a derived position. Although this may not seem the most straightforward position from which to allow extraction (but see Torrego Reference Torrego1985, Chomsky Reference Chomsky1986: 25ff.), this is exactly what appears to happen in the wat … voor-split construction:

1. (51)

   

Also, some speakers allow stranding in the matrix VP (Barbiers Reference Barbiers, Alexiadou, Anagnostopoulou, Barbiers and Gärtner2001), as in (52). This cannot be accounted for unless we allow wat to sub-extract from a derived position:

1. (52) [_CPWat had jij dan watvoorboeken gedacht [_CP dat Jan zou lezen?]] what had you then what for books thought that Jan would read ‘What kind of books would you have thought that Jan would read?’

There are in fact four additional similarities between partial wh-copying and the wat … voor-split, which we will now briefly review. First, the higher copy cannot be more specified than the lower one, in line with what we have seen to be the case for wh-copying constructions and in accordance to our generalization in (12).

1. (53)

   1. (a) *Wie denk je wat ik gezien heb? who think you what I seen have

   2. (b) *Watvoorboeken denk je wat Jan gelezen heeft? what for books think you what Jan read has

Secondly, in both wh-copying constructions and the wat … voor-split PF apparently cannot spell out wh-copies in their base position:

1. (54)

   1. (a) * Ik vraag me af [_CPwat_i Jan wie_i gezien heeft]. I ask me prt what Jan who seen has ‘I wonder who Jan has seen.’

   2. (b) * Ik vraag me af [_CPwat Jan watvoorboeken gelezen heeft]. I ask me prt what Jan what for books read has ‘I wonder what kind of books Jan has read.’

Note that not spelling out the lower wat in (54b) gives a grammatical result (cf. (55b)). The strategy of not spelling out the lower copy cannot apply to (54a) on the intended reading, as this would lead to deletion of unrecoverable Phi-features. (In other words, even though (55a) is grammatical, it does not get a ‘who’-reading, but a ‘what’-reading.)

1. (55)

   1. (a) Ik vraag me af [_CPwat_i Jan wie gezien heeft]. I wonder me prt what Jan who seen has ‘I wonder what/*who Jan has seen.’

   2. (b) Ik vraag me af [_CPwat Jan watvoorboeken gelezen heeft]. I ask me prt what Jan what for books read has ‘I wonder what kind of books Jan has read.’

Thirdly, in both constructions any wh-element occuring in SpecCP of an embedded clause must be spelled out:

1. (56)

   1. (a) Wat denk je * (wie) Jan gezien heeft? what think you who Jan seen has ‘Who do you think Jan saw?’

   2. (b) Wat denk je * (wat) voorboeken Jan gelezen heeft? what think you what for books Jan read has ‘What kind of books do you think that Jan has read?’

Finally, both partial wh-copying and the wat … voor-split are blocked by intervening negation (cf. Felser Reference Felser2004 and references cited therein).

1. (57)

   1. (a) *Wat denk je nietwie Jan ontmoet heeft? what think you not who Jan met has Intended: ‘Who don't you think that Jan has met?’

   2. (b) *Wat denk je niet dat Jan voormensen ontmoet heeft? what think you not that Jan for people met has Intended: ‘What kind of people don't you think that Jan has met?’

These similarities strengthen the idea that partial wh-doubling and the wat … voor-split are derived by the same process which only partially copies a constituent.

There is, however, one major difference. In wat … voor-constructions wat is always visible. In wh-doubling constructions, on the other hand, wat shows up in a subset of cases, namely whenever partial copying takes place. When full copying applies, or no copying at all (i.e. there is no movement), wat is not spelled out. This is what we expect, since spell-out of wie subsumes spell-out of wat.

1. (58)

   

In wat … voor-constructions, voor as well as the noun are spell-outs of terminal nodes, neither of which dominates wat. Hence, wat can and will always be spelled out, and we therefore expect that wat voor boeken can occur as a constituent, depending on the position it is in.

1. (59)

   

4.1 Scattered deletion at PF and the ‘big XP’ approach

Nunes argues that spelling out more than one chain member is in general illicit, because it violates Kayne's LCA: if syntactic structure is linearized on the basis of c-command relations, movement of α across β leads to the situation in which β both c-commands and is c-commanded by α. Hence, α and β cannot be linearized, as PF is provided with conflicting linearization instructions. One strategy to make linearization possible is to delete all but one copy. Another strategy, termed ‘scattered deletion’, is to delete complementary material in two copies. As an illustration of this mechanism, consider the following Croatian example from Nunes (Reference Nunes2004: 29), taken from Cavar & Fanselow (Reference Cavar and Fanselow1997). Strikethrough indicates deletion at PF. (60b) indicates the syntactic structure that feeds PF; (60c), the result of scattered deletion at PF.

1. (60)

   1. (a) Na kakav je Ivan krov bacio loptu? Croatian on what.kind.of be Ivan roof throw ball ‘On what kind of roof did Ivan throw the ball?’

   2. (b) [[_PP na [kakav krov]]ⁱ je Ivan [[_PP na [kakav krov]]ⁱ bacio loptu [[_PP na [kakav krov]]ⁱ]

   3. (c) [[_PP na [kakav krov]]ⁱ je Ivan [[_PPna [kakav krov]]ⁱ bacio loptu [[_PPna [kakavkrov]]ⁱ]

Note that, under this approach, there is no partial copying: in the syntax, the whole wh-constituent is copied each time and PF subsequently decides where to delete what, thereby deriving what looks like sub-extraction.

Assume that we try to apply Nunes' notion of scattered deletion to the cases of non-identical doubling in the Dutch dialects. First we apply full copying of a wh-pronoun. Subsequently, only part of the higher copy is spelled out, namely as wat, and only part of the lower copy, namely as wie. Although we thus obtain the correct result, scattered deletion is not precise enough as it stands. Nothing yet blocks spelling out the higher copy as wie and the lower one as wat, so that we incorrectly predict partial underspecification of a lower copy to be possible. In other words, additional stipulations are needed to derive the generalization in (12).

Let us now turn to the so-called ‘big-XP’ analysis of pronominal doubling (see Cecchetto Reference Cecchetto2000, Kayne Reference Kayne, Epstein and Seeley2002, Belletti Reference Belletti2005 and Poletto & Pollock Reference Poletto and Pollock2004 in particular for wh-doubling in Romance varieties). In their analysis of wh-clitic doubling in French and Northern Italian dialects, Poletto & Pollock argue that the doubler and doublee are initially contained in one and the same nominal XP. During the derivation, doubler and doublee move to different syntactic positions, creating a doubling effect. Applying the essence of this analysis to the data at hand yields the XPs given in (61), in which wie is (contained in) the sister of wat or vice versa. (62), by contrast, provides the structures assigned by our own analysis. The structures are different in that in (61) both wh-pronouns are dominated by the same category, XP, whereas in (62) one wh-pronoun dominates the other. More precisely, wie is the spell-out of a category that dominates the category that is spelled out as wat.

1. (61)

   1. (a) [_XPwat wie] or [_XPwie wat]

   2. (b) [_XPwie die] or [_XPdie wie]

   3. (c) [_XPwat die] or [_XPdie wat]

1. (62)

   1. (a) [_wie … [_wat …]]

   2. (b) [_die … [_wie … [_wat…]]]

There are two ways in which wie wat can be derived, and these have to be ruled out. We will show that the first derivation does not decide between the two approaches, whereas the second one provides an argument in favour of our approach.

The first way in which wie … wat could be derived involves moving wat to the embedded SpecCP, and subsequent movement of wie into matrix CP. (63) illustrates how this structure would be derived under each of the two approaches.

1. (63)

   

These derivations can be straightforwardly ruled out. In long extraction, the reason why the wh-pronoun lands in an intermediate SpecCP is locality-related: it does not move there for feature checking reasons (cf. Bošković Reference Bošković2007 and section 5.2 below for discussion). Therefore, there is no reason for wat to end up in the embedded SpecCP position and this movement is blocked.Footnote ¹⁷ Moreover, subsequent movement of wie would then be non-local, violating locality (unless one assumes that wat can move to C, à la Nunes (Reference Nunes2004), and wie through the embedded SpecCP). Since both approaches can resort to this reasoning in order to rule out ungrammatical wie … wat, (63) does not favour either.

There is, however, another way of deriving wie … wat which the big XP approach in principle makes available, and one needs extra assumptions to rule it out. Suppose that we move XP to the embedded SpecCP and then sub-extract wie and move it to the matrix SpecCP. The result is given in (64):

1. (64)

   

Note now that it is impossible to carry out a similar derivation in our approach. Since spelling out PhiP subsumes spelling out QP (i.e. PhiP dominates QP), the only way for wat to surface is if it is sub-extracted from PhiP. Movement of PhiP never strands QP. Hence, moving PhiP can at most lead to identical doubling (wie … wie) but never to non-identical doubling (wie … wat). To conclude, our approach rules out the ungrammatical example without further ado.

In summary, while the ‘scattered deletion’ and the ‘big XP’ approach require additional machinery to capture the generalization in (12), the present analysis does not necessitate such measures.

4.2 The direct–indirect dependency controversy

According to the account proposed here, both identical (i.e. full) and non-identical (i.e. partial) doubling involve wh-elements belonging to a single chain that is established in overt syntax. In this respect, our proposal can be thought of as belonging to the family of direct dependency approaches to partial copying (van Riemsdijk Reference Riemsdijk1982, Hiemstra Reference Hiemstra1986, McDaniel Reference McDaniel1989, Beck & Berman Reference Beck, Berman and Lutz2000, Brandner Reference Brandner and Lutz2000, Cheng Reference Cheng and Lutz2000, von Stechow Reference Stechow and Lutz2000, and others). A crucial point of departure concerns the status of the ‘scope marker’ wat. Contrary to most direct dependency accounts, we assume that the creation of a dependency relation between the scope marker and the real wh-element does not rely on a covert movement operation (namely one that replaces the scope marker with the contentful wh-element at LF) but is already established in overt syntax.

In this section, we will compare our direct dependency approach (which takes wat to be a partial copy of wie or die) to the indirect dependency approach and show the advantages of the former. Our conclusion will be that our version of the direct approach fares better in accounting for the Dutch data than the most prominent indirect approaches. We do not wish to claim that the direct approach is superior in general. Arguments go both ways and the type of analysis one wants to assume may differ from language to language, suggesting the existence of a spectrum that ranges from indirect to direct dependencies (cf. Fanselow Reference Fanselow, Everaert and van Riemsdijk2006; also Dayal Reference Dayal and Lutz2000, Horvath Reference Horvath and Lutz2000). We restrict ourselves here to a few remarks about the basic tenets of the indirect approaches, so as to bring out the differences (and, as we will see, similarities) with our own approach.

The indirect dependency approach holds that there is no direct syntactic relation between the higher wh-element and the lower one. Matrix wat is co-indexed with the entire embedded clause and the latter functions as the restriction of this operator. This semantic analysis leaves underdetermined what the syntactic function of the CP-clause is, and thus what its relationship is to the wh-operator. We will discuss two main proposals: on the first, the wh-pronoun in the matrix clause (e.g. wat in Dutch, was in German) is the syntactic object of the matrix verb (Dayal Reference Dayal1994, Felser Reference Felser2001); on the second, the wh-pronoun is an expletive, generated in the functional domain related to the matrix object (cf. Fanselow & Mahajan Reference Fanselow, Mahajan and Lutz2000, Horvath Reference Horvath and Lutz2000). Let us look at each in turn.

According to Dayal, the embedded clause is adjoined to CP or IP. This, however, has been shown to be problematic for German. See Beck & Berman (Reference Beck, Berman and Lutz2000), Horvath (Reference Horvath and Lutz2000), Felser (Reference Felser2001), Fanselow (Reference Fanselow, Everaert and van Riemsdijk2006) for arguments suggesting that the CP-clause occupies a low, VP-internal position. But if German was realizes an argument of the matrix verb, what is the syntactic status of the CP-clause? Felser (Reference Felser2004) proposes that it is a secondary predicate, predicating of was. This analysis is problematic, however. If both was and the CP-clause are base-generated VP-internally, as an object and secondary predicate, respectively, we would expect them to be able to show up in their base-generated positions. As pointed out by Bayer (Reference Bayer1996), multiple wh-questions are a good testing ground: if another wh-constituent occupies the sentence-initial position, was should be able to appear in object position (cf. (65a)). This prediction is not borne out (cf. (65b)):

1. (65)

   1. (a) Wer hat was gedacht? who has what thought ‘Who thought what?’

   2. (b) * Wer hat was gedacht wen wir anrufen sollten? who has what thought whom we call.up should Intended: For which person x and for which person y, x thought that we should call up y?

Felser deals with this problem as follows. She points out that whenever was is modified it gets the indefinite interpretation, unless it moves to the left periphery. This is shown in (66):

1. (66)

   1. (a) Er hat mir was Schönes gesagt. he has me something nice said ‘He said something nice to me.’

   2. (b) Was hat er dir denn Schönes gesagt? What has he you prt nice said ‘Which nice things did he say to you?’

   3. (c) Wer hat dir denn was Schönes gesagt? who has you prt something nice said ‘Who said something nice to you?’ *‘Who said which nice things to you?'

In (65b) was has not been fronted and is predicated of (secondarily) by the CP-clause. This, according to Felser, results in a conflict: was has to be interpreted as [–wh] since it is in situ and modified (by the secondary predicate), but its predicate (the CP-clause) is [+wh]. This explanation seems flawed to us. If one takes into consideration prototypical secondary predicates, then was in situ need not be interpreted as an indefinite but can also be a wh-element:

1. (67) Wer hat was roh gegessen? who has what raw eaten ‘Who ate something raw?’ ‘Who ate what raw?’

Hence, under the assumption that the CP-clause is a secondary predicate, there is (contra Felser) no explanation for the ungrammaticality of (65b). Since this example crucially shows that was cannot overtly occur in the position in which it is supposedly inserted, this key feature of the indirect dependency approach is challenged.

In short, the indirect dependency approach offers a plausible semantics for the scope marking construction, but has a harder time providing a syntactic function for the dependent CP-clause. What is more, some constructions from the Dutch dialects make it even harder to apply the indirect dependency approach to Dutch wh-doubling. Consider sentences (4) and (5), repeated below:

1. (68) Wie denk je die ik gezien heb? North-Holland who think you rel.pron I seen have ‘Who do you think I saw?’

1. (69) Wat denk je die ik gezien heb? Overijssel what think you rel.pron I seen have ‘Who do you think I saw?’

While our approach derives (68) and (69) in a uniform way, by relying on partial copying, this is not the case for the indirect dependency analysis. Note first that the latter approach has nothing to say about examples such as (68), which features ‘who’ in the matrix C, a wh-element that does not qualify as a propositional operator. As for (69), the indirect dependency approach is forced to assume that the dependent CP-clause (the clause introduced by die) is a question, functioning as the restriction on the quantifier wat, which realizes the object of the matrix verb. We therefore expect examples like (70) and (71) to be grammatical, contrary to fact.Footnote ¹⁸

1. (70) * Die denk je die ik gezien heb?rel.pron think you rel.pron I seen have Intended: ‘Who do you think I saw?’

1. (71) * Die heb ik gezien?rel.pron have I seen Intended: ‘Who do you think I saw?’

If the embedded clause is not a question, as (70) and (71) strongly suggest, then it cannot be taken to provide the restriction on the quantifier ‘what’. This poses a problem for the semantics proposed for the scope marking construction under the indirect dependency approach. Moreover, it bears on what the syntactic status of the dependent CP is: if the CP-clause were indeed a question, it would be impossible to treat it as the syntactic object of the matrix verb, since denken does not select a question. If the embedded clause is not a question, however, then nothing stands in the way of it being the object of denken. This is the syntactic analysis we assume for all dependent CP-clauses in these doubling constructions.Footnote ¹⁹ The consequence of this is that dependent CP-clauses introduced by a wh-element instead of die are not questions either, which suggests that the wh-element is not interpreted in that position, namely embedded SpecCP.Footnote ²⁰

The second type of indirect dependency approach circumvents some of the key problems with the first approach. It assumes that was in German is not an argument of the verb but an expletive generated in some functional position, e.g. in SpecAgrOP (cf. Fanselow & Mahajan Reference Fanselow, Mahajan and Lutz2000). This approach allows for an analysis in which the CP-clause functions as the object of the verb, thereby resolving the issue of its syntactic status. Such an indirect dependency approach is in fact close to a direct dependency proposal along the lines of van Riemsdijk. The major syntactic difference is the insertion site of was, SpecAgrOP (Fanselow & Mahajan) versus SpecCP (van Riemsdijk).Footnote ²¹ The two approaches also differ in what moves in covert syntax, this now being the basic difference between the direct and the indirect dependency approach. What moves covertly in the former is the ‘real’ wh-constituent, and in the latter the CP-clause. Note, however, that what moves in covert syntax is not fully determined by the analysis of the overt syntactic structure. Whereas one could imagine that an expletive in SpecAgrOP triggers movement of the entire CP, such a covert movement is not blocked by the assumption that ‘what’ is generated in the highest SpecCP. Hence, under this second type of indirect dependency approach, the difference from direct dependency approaches becomes harder to see and is no longer dependent on a different syntactic function of the CP-clause.Footnote ²²

Arguments in favour of the indirect dependency approach have been provided by differences between the scope-marking construction and long wh-movement. Since in our approach both constructions involve syntactic chains created in overt syntax, any difference between these two constructions will be potentially problematic for us. For instance, it has been observed that negation in the matrix clause only creates a problem for the scope-marking construction, and not for long extraction (cf. Dayal Reference Dayal1994, Reference Dayal and Lutz2000; Herburger Reference Herburger1994; Beck & Berman Reference Beck, Berman and Lutz2000; Felser Reference Felser2001, among others). This has been used as an argument in favour of analysing the two constructions differently (though see Beck & Berman Reference Beck, Berman and Lutz2000, von Stechow Reference Stechow and Lutz2000). (72) illustrates the pattern in German:

1. (72)

   1. (a) Wen glaubst du nicht dass sie liebt? who think you not that she loves ‘Who don't you think that she loves?’

   2. (b) * Was glaubst du nicht wen sie liebt? what think you not who she loves

However, it has sometimes been overlooked that full copying across negation is also impossible (cf. Rett Reference Rett, Bonami and Cabredo-Hofherr2006):

1. (73) * Wen glaubst du nicht wen sie liebt? who think you not who she loves

Full copying has generally been treated as a variant of long wh-movement involving multiple spell-out (see Fanselow & Mahajan Reference Fanselow, Mahajan and Lutz2000, Felser Reference Felser2004, Rett Reference Rett, Bonami and Cabredo-Hofherr2006, among others), the indirect dependency approach being inapplicable to this construction. Hence, an explanation for the ungrammaticality of (72b) that hinges on a particular aspect of the indirect dependency approach cannot be used to account for the ungrammaticality of (73), because that would rule out (72a) too. Therefore, such an explanation does not count as an argument in favour of the indirect dependency approach. It is unclear to us which factor causes ungrammaticality of (72b) and (73), but that factor has to distinguish both of the copying constructions from long extraction. As we show in the appendix, no existing theory of the intervention effect in (72b) and (73) is fully satisfactory in the light of data pertaining to intervention effects involving operators other than negation.

The same reasoning applies to the ‘anti-locality’ property of wh-scope marking (also brought up by Felser Reference Felser2001 as an argument against the direct dependency approach), illustrated in (74b). Since this property is common to full copying as well (cf. (74c)), it cannot be taken to argue in favour of the indirect dependency approach.

1. (74)

   1. (a) Wen hat Maria getroffen? who has Maria met ‘Who did Maria meet?’

   2. (b) * Was hat Maria wen getroffen? what has Maria who met

   3. (c) * Wen hat Maria wen getroffen? who has Maria who met

Since in our analysis long extraction, full copying and partial copying (i.e. the scope-marking construction) all involve syntactic chains created in overt syntax, any differences between the latter two are also potentially problematic. Here again, differences have been observed and used to motivate a different syntactic treatment. At least one set of data, however, in fact seems to lend support to our analysis. Consider the following examples:

1. (75) Es ist egal, was er meint [wann sie kommt] und [wen sie mitbringt]. it is equal what he thinks when she comes and who she with.brings ‘It doesn't matter what he thinks as to when she will come and who she will bring along.’

1. (76) *Es ist egal, wann/wen er meint [wann sie kommt] und [wen sie mitbringt]. it is equal when/who he thinks when she comes and who she with.brings

Although Felser (Reference Felser2001, Reference Felser2004) uses this contrast as an argument against analysing full copying on a par with partial copying (and thus also as an argument against the direct dependency approach), it in fact only argues against an approach in which the ‘contentful’ wh-element replaces the scope marker was at LF. On such an approach, (76) would involve both wann and wen replacing was at LF, which would predict (75) to be as bad as (76). In our account, the contrast follows straightforwardly. If one assumes, as does Felser, that across-the-board extraction targets identical material, this suffices to rule out (76): either wen or wann moves, but neither one is shared by both conjuncts. In (75), on the other hand, across-the-board movement targets the QP inside each of the two wh-phrases, as schematized in (77).

1. (77)

   

Felser points out further that there exists in German a contrast between full and partial copying concerning island sensitivity. Her observation is that partial copying is impossible with matrix predicates that cannot select a DP object, whereas full copying is better.

1. (78)

   1. (a) * Was hat Peter das Gefühl, wen man fragen könnte? what has Peter the feeling who one ask could Intended: ‘Who does Peter have the feeling that one should ask?’

   2. (b) ?Wen hat Peter das Gefühl, wen man fragen könnte? ‘Who does Peter have the feeling that one should ask?’

1. (79)

   1. (a) * Was scheint es, wen Hans geschlagen hat? what seems it who Hans hit has Intended: ‘Who does it seem that Hans has hit?’

   2. (b) ? Wen scheint es, wen Hans geschlagen hat? who seems it who Hans hit has ‘Who does it seem that Hans has hit?’

This follows if was ‘what’ is an argument of the matrix predicate: it competes for an argument slot with das Gefühl in (78a) and with the embedded CP in (79a). Full copying constructions fare better because wen is not an object of the matrix clause but part of a syntactic chain originating in the embedded clause.

This contrast, however, does not seem to exist in Dutch.Footnote ²³ Consider the data in (80) and (81):

1. (80)

   1. (a) * Wat hoorde je het nieuws wie hij had ontmoet? what heard you the news who he had met Intended: ‘Who did you hear the news that he had met?’

   2. (b) * Wie hoorde je het nieuws wie hij had ontmoet? who heard you the news who he had met Intended: ‘Who did you hear the news that he had met?’

   3. (c) * Wie hoorde je het nieuws dat hij had ontmoet? who heard you the news that he had met Intended: ‘Who did you hear the news that he had met?’

1. (81)

   1. (a) * Wat schijnt het wie Hans geslagen heeft? what seems it who Hans hit has Intended: ‘Who does it seem that Hans has hit?’

   2. (b) * Wie schijnt het wie Hans geslagen heeft? who seems it who Hans hit has Intended: ‘Who does it seem that Hans has hit?’

   3. (c) % Wie schijnt het dat Hans geslagen heeft? who seems it that Hans hit has Intended: ‘Who does it seem that Hans has hit?’

In (80), full and partial copying are equally bad. Note that long extraction also gives a bad result (cf. (80c)). The explanation of this is most likely that wie has been extracted out of a complex NP. In our account, this explanation carries over to (80a) and (80b). In (81), long extraction is much better but, crucially, there again does not seem to be a contrast between full and partial copying. Whatever the explanation for the ungrammaticality, the Dutch data do not warrant a syntactic differentiation between full and partial copying.

5.1 Wh-doubling in child language

In this section, we will discuss a construction that seems to run counter to the generalization we have claimed to hold of syntactic chains, namely that a higher chain link is not more specified than a lower chain link (recall (12)). Apart from being a feature of Dutch dialects, doubling constructions also occur prominently in child language in a variety of languages. According to van Kampen (Reference Kampen1997, Reference Kampen, Everaert, Lentz, Nilsen and Zondervan2008), children have wh-doubling constructions before they can do long wh-extraction. See also McDaniel, Chiu & Maxfield (Reference McDaniel, Chiu and Maxfield1995) and Jakubowicz & Strik (Reference Jakubowicz and Strik2008) for the pervasive nature of wh-doubling in child language. In addition to full and partial wh-copying constructions, constructions are attested in which a full wh-phrase precedes a wh-pronoun (cf. van Kampen Reference Kampen1997):

1. (82) Welke jongen denk je wie daar loopt? which boy think you who there walks ‘Which boy do you think walks there?’

Van Kampen (Reference Kampen, Everaert, Lentz, Nilsen and Zondervan2008) argues that the full set of data from child language strongly suggests that children make use of the paradigm of relative pronouns when constructing the wh-constructions below.Footnote ²⁴ The corresponding antecedent–relative pronoun pairs are given in (84):

1. (83)

   1. (a) In welk huis denk je waar jij woont? (child grammar) in which house think you where you live ‘In which house do you think you live?’

   2. (b) Op welke jongen denk je op wie ik verliefd ben? on which boy think you on who I in.love am ‘Which boy do you think I am in love with?’

   3. (c) Welk huis/welk meisje denk je wat ik leuk vind? Which house/which girl think you what I nice find ‘Which house/girl do you think I like?’

1. (84)

   1. (a) het huis waar jij woont (adult grammar) the house where you live

   2. (b) de jongen op wie ik verliefd ben the boy on who I in.love am ‘the boy that I am in love with’

   3. (c) het meisje wat ik leuk vind the girl what I nice find ‘the girl that I like’

An argument in favour of such an analysis, according to van Kampen, is that examples in which the wh-phrase and the pronoun do not yield a legitimate antecedent–relative pronoun pair in Standard Dutch are unattested in the child data:

1. (85)

   1. (a) * In welk huis denk je daar ik woon? in which house think you there I live

   2. (b) * Welke boeken denk je wat ik heb gelezen? which books think you what I have read

   3. (c) * Welke villa denk je wat ik ga kopen? which villa think you what I go buy

1. (86)

   1. (a) * het huis daar ik woon the house there I live

   2. (b) * de boeken wat ik heb gelezen the books what I have read

   3. (c) * de villa wat ik ga kopen the villa what I go buy

Van Kampen argues that children insert these relative pronouns in the C-head of the embedded complement clause, the choice of pronoun being determined by the features of the wh-phrase that has moved through the embedded SpecCP. In other words, the examples in (83) are analysed as a form of agreement between C and a copy of the wh-phrase (cf. Thornton & Crain Reference Thornton, Crain, Hoekstra and Schwartz1994). The analysis in addition captures the fact that doubling of wh-phrases is not attested: a phrase cannot function as a filler of C.

1. (87) *Welke jongen denk je welke jongen daar loopt? which boy think you which boy there walks ‘Which boy do you think is walking there?’

Two questions can be raised: (i) do the child data pose a problem for our generalization, and (ii) can van Kampen's proposal be applied to the dialect data we discuss in this paper, obviating the need for partial copying? We think that the answer to both questions is negative.

Let us start with the latter question. Treating lower copies as fillers of C is not a viable analysis for the basic data that we set out to explain. The reason is the existence of wh-doubling dialects in which the embedded C-head is already filled by a complementizer. See Felser (Reference Felser2004) (citing Bayer Reference Bayer1996) for a similar argument on the basis of comparable German data. Consider the southern Dutch data below (see also the examples in (90)):

1. (88)

   1. (a) Wie peis-de-gij wie da ’k in ’t stad gezien en. Koewacht who think-you-you who that I in the city seen have ‘Who do you think I have seen in the city?’

   2. (b) Wie peis-de-gij nu wie a ’k in ’t stad gezien heb. who think-you-you now who that I in the city seen have St Gillis bij Dendermonde

If in (88a, b) da and a fill the embedded C-position, the wh-pronoun cannot be an agreeing complementizer.

This brings us to the first question. If the lower wh-pronoun is in a specifier position, it could in principle form a syntactic chain with the sentence-initial wh-phrase. To uphold our generalization, we have to assume that welke jongen is base-generated upstairs and is related to a sentence-internal operator, realized as wie (cf. Chomsky Reference Chomsky, Culicover, Wasow and Akmajian1977; McCloskey Reference McCloskey1979, Reference McCloskey and Hendrick1990; van Craenenbroeck Reference Craenenbroeck2004). This pronominal operator is the element undergoing sentence-internal movement and is related to the base-generated phrase through semantic construal (or R-binding, cf. Safir Reference Safir1984). In this respect, the construction in (82) shows a similarity with Germanic contrastive left dislocation (CLD) constructions, for which a similar analysis has been proposed (Koster Reference Koster and Keyser1978, Weerman Reference Weerman1989).Footnote ²⁵Van Craenenbroeck (Reference Craenenbroeck2004) explicitly proposes this analysis for wh-constructions in Dutch dialects, and we refer to this work for several arguments pointing towards the base-generated nature of the wh-phrase.Footnote ²⁶

In the base-generation analysis, it is not unexpected that the data in (83) show feature agreement between the base-generated phrase and the wh-pronoun.Footnote ²⁷ Such agreement is typically also found in CLD-constructions, as the data below illustrate:

1. (89)

   1. (a) Deze jongen, die / * dat ken ik. this boy that_non-neuter that_neuter know I ‘This boy, I know.’

   2. (b) In Groningen, daar / * dat kun je lekker eten. in Groningen, there_PP that_DP can you nice eat ‘In Groningen you can eat well.’

In addition, the base-generation analysis provides a straightforward explanation of the fact that doubling of wh-phrases is ungrammatical. If wh-phrases can never undergo movement, they never occupy an intermediate SpecCP position, which would be necessary if (87) were to be possible. This is required for (87) to surface.Footnote ²⁸

To conclude, if wh-phrases do not move but are base-generated in the position we see them in, they do not form a syntactic chain with a clause- or sentence-internal wh-pronoun. Hence, examples in which a wh-phrase and a wh-pronoun referring to the same syntactic entity co-occur in a sentence, with the wh-phrase preceding the wh-pronoun, do not fall under our generalization and do not constitute a problem.Footnote ²⁹

5.2 On the properties of intermediate movement steps

Given our analysis of (full and partial) wh-doubling, and especially the discussion in section 4.2 of the syntactic and semantic status of the dependent clause in these constructions, the C-head of the dependent clause cannot contain a [+Q] feature. This raises the following two questions. First, what triggers the intermediate movement step? In (90), a wh-pronoun occurs in intermediate position, although the matrix verb denk ‘think’ does not select a [+Q] complement. Secondly, what licenses the appearance of a [+Q] complementizer like of ‘whether’ in these cases?

1. (90)

   1. (a) Wel denkie wel of ik in de stad tegen kommen ben? Drenthe who think.you who if I in the city against come am ‘Who do you think I met in town?’

   2. (b) Wie denk je wie of dat ik in de stad tegenkomme ben? who think you who if that I in the city against.come am Noord-Holland ‘Who do you think I met in town?’

As regards the first question, we follow Bošković (Reference Bošković2007) in assuming that wh-movement to intermediate positions is not triggered by the presence of a [+Q] feature on the intermediate C⁰, but takes place for reasons of locality (cf. Bošković Reference Bošković2007 for details).Footnote ³⁰

As regards the second question, a crucial observation is that complementizer of cannot occur if the lower copy of the wh-chain is not spelled out (i.e. in a non-doubling construction):

1. (91) * Wie denk je of ik gezien heb? who think you if I seen have

What we suggest is the following. We have assumed throughout that wh-pronouns lack a [+Q] feature. The pronoun is interpreted as a question word when it occurs in the relevant syntactic context, i.e. in SpecCP of the matrix clause. At that point the entire wh-chain is marked for the property [+Q]. The contrast between (90) and (91) suggests that the embedded C obtains the [+Q] feature not as a consequence of syntactic agreement but through phonological agreement: spell-out of the lower link in the wh-chain feeds spell-out of of ‘whether’. If this agreement is post-syntactic, the consequence is that the [+Q] feature will not be visible on the category selected by the matrix verb denken ‘think’, namely the projection of the embedded C. This CP is therefore not a question. Hence, there is no clash with the selectional properties of the matrix verb.

This analysis makes a prediction. If spell-out of of is parasitic on an adjacent terminal hosting a [+Q]-feature (i.e. if (91) is ungrammatical), we expect that of cannot appear with a lexical wh-phrase, as the noun will intervene between welke ‘which’ and the complementizer.

1. (92) * Wat denk je welke jongen of (dat) Hans gezien heeft? what think you which boy if that Hans seen has Intended: ‘Which boy do you think that Hans has seen?’

Although Google provides plenty of cases involving co-occurrence of a wh-pronoun and of, it does not provide examples involving co-occurrence of a wh-phrase and of. This is of course only suggestive, and further research is required to verify the prediction.Footnote ³¹