Inﬁnitive Wh -Relatives in Romance: Consequences for the Truncation-versus-Intervention Debate

. Romance clitic left dislocation is widespread across all kinds of nonroot contexts, but it is forbidden in inﬁnitive wh -relatives. This article investigates the extent and nature of this restriction and the consequences it raises for the truncation and intervention analyses of the left periphery of embedded sentences. We will show that current proposals cannot account for the whole gamut of data. In consequence, we will propose that inﬁnitive wh -relatives display a maximally syncretic left periphery, whereas inﬁnitive wh -interrogatives have a full-ﬂedged left periphery, crucially involving ForceP, because they are selected by a higher predicate. This crucial difference between inﬁnitive relatives and interrogatives will also be shown to be consistent with the existence of specialized complementizers for the former but not the latter.


Introduction
After the pioneering studies in the seventies (??), research on main-clause phenomena, also known as root phenomena, has seen a revival in recent years (???).A focus has been on the place of English topicalization and Romance clitic left dislocation (CLLD) within the realm of main-clause phenomena.In particular, Haegeman has defended the view that, whereas English topicalization is a true main-clause phenomenon, Romance CLLD is widespread across all kinds of nonroot contexts.However, beyond this widely accepted empirical generalization, theoretical explanations of the difference have divided into two camps.On the one hand, some approaches (??) have argued for a structural difference, namely that some functional projections are not available in certain nonroot contexts in certain languages (the truncation hypothesis).On the other hand, other approaches (?????) have argued for an explanation in terms of the features of the elements in the left periphery of the sentence and their interactions (the intervention hypothesis).
This article brings new evidence to the fore that may help place the discussion on a different footing.We examine the behavior of Romance CLLD in infinitive wh-relatives (IWRs), like the following. 1p4in@p0.25in@> p2.25in@@p4in@p0.25in@> p2.25in@124578As we are going to discuss at length, these sentences pose a problem for current views of the left periphery that are couched in the cartographic program (??), for they don't allow CLLD: I would like to express my thanks for the comments by the audience at the first International Workshop on the interface of Information Structure and Argument Structure (Seville, 25-27 October, 2017), where a preliminary version of this work was presented.I am also indebted to Lisa Brunetti, M.Teresa Espinal, Sonia Cyrino, and Olga Borik, for their insights at project meetings, and to Ángel L. Jiménez-Fernández and two anonymous reviewers, who encouraged me to sharpen my ideas and express them more clearly.This article has been possible due to research projects FFI2017-82547-P and FFI2016-81750-REDT of Spain's Ministry of Economy and Enterprise (MINECO) and research project 2017SGR-634 of Catalonia's Agency for Management of University and Research Grants (AGAUR). 1 Throughout the text I follow the Leipzig Glossing Rules.I thus use the following abbreviations: 1, 2, 3 = first, second, third person, ACC = accusative, COND = conditional, DAT = dative, F = feminine, FUT = future, INF = infinitive, LOC = locative, PL = plural, PST = past, REFL = reflexive clitic, SG = singular, SBJV = subjunctive, and SUP = supine.p5in@p0.25in@> p1.25in@@p5in@p0.25in@> p1.25in@124578On the one hand, the ill-formedness of (2) seems to follow from the truncation hypothesis straightforwardly: infinitive clauses lack (part of) the space in the left periphery that hosts topics and foci.However, infinitive wh-interrogatives are compatible with CLLD: p5in@p0.25in@> p1.25in@@p5in@p0.25in@> p1.25in@124578This is unexpected if one assumes that infinitive sentences have a defective left periphery (note that in the case of wh-interrogatives, the wh-word must follow the dislocate, as we will discuss in sections ?? and ?? below).Obviously, if we follow ?:484 in assuming that infinitive clauses do not allow main-clause phenomena at all, the possibility of (3) remains mysterious.
On the other hand, if one considers the ill-formedness of (2) to be the result of the intervening role of the dislocate in the path of the wh-relative pronoun or adverb, the grammaticality of parallel finite relatives, as in (4), remains unexplained.p5in@p0.25in@> p1.25in@@p5in@p0.25in@> p1.25in@124578 In this article I will investigate the extent and nature of the ban against CLLD in IWRs and explore its consequences for the structure of the left periphery commonly assumed since ?: (5) ForceP TopicP FocusP TopicP FinitenessP . . .

Foc
Top Force I will show that neither the truncation nor the intervention analysis can account for the whole gamut of facts, so that we must rethink our current assumptions about the left periphery of infinitive sentences and assume crucial structural differences between IWRs and infinitive wh-interrogatives.
The structure of the article is as follows.In section 2 I will present some basic properties of IWRs, which clearly support treating them as wh-relatives.In section 3, I will briefly review the general assumptions about the structure of infinitive sentences.Then, in section 4, I will consider the possibility of extending the truncation analysis of infinitive clauses to IWRs.Section 5 will present the limits of the intervention analysis, while offering clear support for analyzing CLLD as movement.Then, in section 6, I will present a new proposal, which argues for a maximally syncretic left periphery for IWRs but a full-fledged one for selected wh-interrogatives; this contrast will be linked to the nonselected nature of IWRs.Finally, I will close the article with the main conclusions and theoretical consequences.

Description of IWRs
IWRs are widely attested across the Romance landscape (see, among others, ????): p4.25in@p0.25in@> p2in@@p4.25in@p0.25in@> p2in@124578 These relatives involve the same relative-gap configuration as their finite counterparts and are formed with the same relative pronouns and adverbs.Moreover, just as happens with their finite counterparts, extraction from IWRs is impossible, as the following Catalan and Spanish examples illustrate.
Since we can establish beyond doubt that IWRs are bona fide wh-relatives, in the next section I will briefly consider the particular properties of infinitive sentences in general, as a necessary introduction to the discussion of truncation in section ??.

The Structure of Infinitive Clauses
The internal structure of nonfinite clauses in general and of infinitive clauses in particular hasn't been pursued in much detail; it has been assumed to be similar to that of finite clauses.For example, even though ?showed that root transformations didn't apply in infinitive clauses, ?firmly defended the view of infinitive clauses in English as nondefective, involving, they argued, a full CP in which the particle to was the morphological realization of nonfinite inflection.However, this work came prior to the explosion of functional projections following the work of ? and particularly ?, who developed the now classical split CP already depicted in ??.
Rizzi is not particularly explicit about infinitives, but he distinguishes the cases of control and raising complements: p5in@p0.25in@> p1.25in@@p5in@p0.25in@> p1.25in@124578Even though Rizzi acknowledges that ??a is "slightly marked" in Italian in comparison with its finite counterpart and is impossible in French (fn.24: "Speakers of French are reluctant to accept CLLD with infinitives"), he interprets the contrast in ?? in structural terms: in the control case, we have a full-fledged CP, hence there is a TopP to host the dislocate, whereas in the raising case, we have a TP without any trace of the CP area, hence no room for topics.Since Rizzi considers the prepositional complementizer to be realized in Fin, we derive the following structure, in which it is a priori possible that the CLLD dislocate occurs in either TopP.
In contrast, the raising structure lacks any projection above TP, which leaves no room for CLLD.(On the categorial status of infinitive complements, the reader can consult ??????, among others.)Obviously, if we take CLLD as a hallmark of a rich CP periphery, we can conclude that IWRs are TPs or, at least, defective CPs without TopP, whereas infinitive interrogatives must be full-fledged CPs.If we maintain the structure in ?, the defective infinitive structure would involve at most FinP, which would syncretically realize both [finiteness] and [force]: In the following section, we discuss the details of this proposal.

Truncation
The truncation hypothesis (??) was originally formulated to account for the possibility of using infinitives as main-clause predicates in initial stages of child language, but it soon found its way into adult-grammar studies (????), as a means to account for the limited variety of discourseoriented material (i.e.topic, focus) in certain infinitive clauses.Let us consider the main facts.In answer to this theoretical problem, we can, as originally suggested by ?, conceive of truncation as a case of syncretism: the features associated with Fin and Force are realized under the same functional projection, FinP in the case at hand (see section ?? and ?for a similar proposal): The second problem we must face is empirical, and a harder bullet to bite: generalized truncation of infinitive clauses doesn't extend to infinitive wh-interrogatives, which do allow CLLD in some Romance languages, as we have shown for Catalan in ??, repeated here along with a parallel Spanish example.

Empirical Support
p4.5in@p0.25in@> p1.75in@@p4.5in@p0.25in@> p1.75in@124578 In contrast to Catalan and Spanish, the case of Italian is more restricted.This contrast would support a radical-truncation approach to infinitive wh-interrogatives in Italian. 2 Hence, whereas the left periphery of interrogatives allows CLLD with much variation across Romance, relative clauses are strikingly homogeneous: CLLD must follow the relative word in finite relatives but is forbidden altogether in IWRs.

An Even Finer Left Periphery
We have shown how relatives and interrogatives fit into the ?proposal for the left periphery; however, we must consider a further development with consequences for our study.
Along with a lower position for wh-elements in FocP, ? and ?argue for a higher dedicated Int(errogative)P that hosts Italian se 'whether' and perché 'why', which is crucially surrounded by TopPs: ( Rizzi argues that wh-words in Italian have the complex distribution given in table ??.Note that CLLD can in principle target any of the topic projections, but particular restrictions apply, as in the case of dove/cosa 'where/what', which doesn't allow a CLLD to its right, in contrast to se/perché 'whether/why'.Moreover, in the latter case, we have no empirical evidence for deciding the exact position of the CLLD to the right of the wh-word, but one would expect that the same restrictions would hold which apply in the case of dove/cosa.On the interaction between topics and whelements see section 5.2.The special behavior of se 'whether' and perché 'why' in Italian is widespread across Romance and beyond (????).For instance, Catalan shows a rigid CLLD-wh-interrogative order but allows the reverse wh-interrogative-CLLD order with si 'whether' and per què 'why' (?:224): ( 22 Now, let us consider infinitive interrogatives.With most wh-elements, we only find the CLLDwh-interrogative order, and with 'whether' the reverse order seems much more restricted than with finite interrogatives (if possible at all): compare ( 22) with ( 24) and ( 23) with (25).p5in@p0.25in@> p1.25in@@p5in@p0.25in@> p1.25in@124578 p4.5in@p0.25in@> p1.75in@@p4.5in@p0.25in@> p1.75in@124578Let us see how Rizzi's expanded schema in table ?? applies to these cases.In the case of lower wh-interrogatives, which are hosted in Spec,FocP, we can assume a structure like the following, where after the movement of wh-word and the infinitive to the specifier and head of FocP, respectively, and the restriction on the lower TopP already commented, we correctly expect the CLLD-wh-interrogative order only.However, we have seen in ?? and ?? that the wh-interrogative-CLLD order is far from perfect and, in any event, worse than the CLLD>wh-interrogative order, against expectations.Note that since the higher TopP is activated, we cannot expect truncation to account for the impossibility of a lower dislocate.This pattern remains unexplained in cartographic studies, particularly under proposals like ??, which explicitly "assume that topicalisation and focalisation depend on the presence of Force" (?:335).If so, Romance IWRs would be special in lacking Force, unlike their finite counterparts and infinitive wh-interrogatives.

ForceP
To end this section, we can conclude that the truncation analysis offers a simple solution for IWRs, since it forbids focus or topic material in the left periphery, but it cannot be extended to infinitive interrogatives, which do allow CLLD.In the next section, we consider an alternative that tries to derive these structural restrictions from general intervention effects between different kinds of constituents.

Intervention
An alternative to the cartographic truncation analysis is to derive the restrictions on CLLD from intervention effects (?????), in fact from a version of Rizzi's Relativized Minimality (??).?discusses in some detail a particular intervention effect involving discourse linking (D-linking; see ? for the original idea and ???? for several approaches).She considers the following standard wh-island effects (I mark the offending wh-element with boldface).
Even though conceptually appealing, Haegeman's approach is problematic in several respects, particularly when applied to relatives.First of all, she extends the original notion of D-linking in an unconstrained and disputable way.?originally applied this concept to wh-interrogatives, for D-linking is crucially grounded on a closed salient set of possible answers.To my knowledge, ?:18 was the first to connect relatives with D-linking.He suggests that the asymmetry noted by ?:50-51 between relatives and interrogatives with regard to extraction from wh-islands is due to referentiality.Rizzi One might expect a parallel between interrogatives and relatives with respect to the complexity of the wh-phrase: D-linking should be more easily available with nominally restricted wh-phrases.However, ?:164 even maintains his claim for (32), with a clearly nonreferential antecedent and a null operator.
(32) I know of nobody that I really know how to talk to.
This extension of D-linking to relative pronouns and adverbs has been particularly defended for those languages where an overt determiner is necessary to form complex wh-relatives, such as Bulgarian and most Romance languages (see, e.g., ?:277-278).Whether such a move is theoretically motivated is unclear, but different predictions follow regarding the particular choice of features (see, for instance, ???).In any event, what defenders of the D-linking analysis typically assume is a connection between the D-linked phrase and some TopP in the left periphery.?entertains two possibilities for a D-linked interrogative like the following.The first possibility is moving the whole wh-phrase to FocP, then moving the presupposed nominal part to a higher TopP: The alternative is moving the D-linked wh-phrase to TopP and then moving the wh-word to FocP (note that ?assumes a lower TopP between FocP and FinP): In (36), the D-linked wh-phrase would cross over the dislocate in the lower TopP, and since both share the [top] feature, we predict a bad result, which is correct.In (37), the D-linked wh-phrase moves to the lower TopP, where it should provoke the same minimality effect for the movement of the dislocate to the higher TopP, contrary to fact.
In the case of wh-relatives, ?argues that the relative pronoun/adverb is hosted in the specifier of ForceP, so we would expect two options as well: In (38), the relative wh-phrase would cross over the dislocate in the highest TopP, and since both share the [top] feature, we predict a bad result, contrary to fact.In (39), regardless of whether it moves to FocP or to TopP, the D-linked relative wh-phrase must cross over the dislocate in the lower TopP, where it should provoke the same minimality effect for the movement of the dislocate to the higher TopP, again contrary to fact.As ?discusses, in Italian, a relative pronoun may cross over a dislocate, but not conversely (to help the reader, I will mark dislocates with boldface): p5in@p0.25in@ > p1.25in@@p5in@p0.25in@> p1.25in@124578 Abels's technical solution to this quandary is to develop Rizzi's idea that relative pronouns/adverbs are akin to topics and to specify them as both operators and topics.Crucially, since relative pronouns are a subclass of topics, by the Elsewhere Condition they will be allowed to cross over the less specific element (the dislocate), but not conversely.However, even though ingenious, Abels's solution is problematic on empirical and theoretical grounds.On the one hand, while this prediction might be correct for Italian finite relative clauses (40), it is incorrect for Spanish and Catalan relatives regardless of tense and for IWRs across the board, Italian included, as we have demonstrated at length in section ?? (see also section ??).On the other hand, assuming a topic-like nature for wh-relatives in IWRs is counterintuitive, for they are typically associated with nonspecific indefinite antecedents, and we know that topics are preferably definite and specific NPs (?).
However, Haegeman's analysis of CLLD in Romance is untenable.I will briefly show that CLLD not only does involve movement (section ??) but also shows intervention effects (section ??).
Then I will come back to IWRs and show that intervention analyses, like truncation analyses, cannot account for the data (section ??).

Romance CLLD as Movement
The most compelling reason for analyzing Romance CLLD as movement is its sensitivity to strong islands (see ?:6.2 and ?:ch. 2 for additional arguments).Leaving aside French (?), where the Similar effects are well documented for Spanish by ? and for Catalan by ?:ch. 3 and ?. Indeed, this behavior extends beyond Romance, for example to CLLD in Greek (?:18), to contrastive left dislocation in German (?:143) and Dutch (?:156-157), and to English topicalization (?:91, ?:175).

CLLD and Intervention Effects
We have seen that island effects clearly point towards the movement nature of CLLD.Another set of evidence concerns the interaction of CLLD with wh-movement, which forms a particularly complex pattern.On the one hand, as ?originally pointed out, CLLD does not block the extraction of another dislocate: p4.75in@p0.25in@> p1.5in@@p4.75in@p0.25in@> p1.5in@124578Moreover, it is also clear that relative pronouns and adverbs may surface above a dislocate, as discussed by many authors (???) and in section ??.

. . ]]]]]]]]]]]
p5in@p0.25in@ > p1.25in@@p5in@p0.25in@> p1.25in@124578Secondly, most of the evidence against the intervention effects of CLLD suffers from a basic shortcoming: it involves local CLLD, which, as discussed in ?and ?:ch. 2, displays a mixture of A and A properties, which renders standard tests inadequate.When CLLD is nonlocal and thus clearly A movement, intervention effects become apparent. 5 Finally, note the effect that CLLD adds to wh-islands.Whereas wh-movement is quite insensitive to wh-islands in Romance, the addition of a dislocate results in a stronger effect, as the following Spanish examples show.give.PST.3PL'*Whom don't you know when they gave the book to?' p5.25in@p0.25in@> p1in@@p5.25in@p0.25in@> p1in@124578 She suggests that the contrast may follow from local CLLD being A movement, which typically lacks weak-crossover effects, while nonlocal CLLD is A movement. 6The intervention effects discussed in this section are unexpected for ?, who crucially build on the availability of the specifier of TP as a landing site for CLLD.For them, null-subject discourse-configurational languages like Spanish and Japanese allow the lowering of the feature [top] from the C domain to T, hence making the preverbal subject position available for CLLD/scrambling (an idea pursued before: ???).As a consequence, it is predicted that operator movement to the C domain will not be affected by the dislocated or scrambled constituent To sum up, once we control for the disturbing factors associated with local CLLD, we can safely conclude that CLLD displays intervention effects in Spanish and Catalan, in accordance with the movement approach to this construction defended in ??.

IWRs and CLLD
Now that we have offered compelling evidence that CLLD is formed by means of movement (sec.??) and that it displays intervention effects (sec.??), we can return to IWRs and consider what predictions the intervention account makes for this construction in Romance.We have seen that finite relatives admit CLLD in all Romance languages: p4.75in@p0.25in@> p1.5in@@p4.75in@p0.25in@> p1.5in@124578However, their infinitive versions are impossible altogether, as we have already shown in ??: p4.75in@p0.25in@> p1.5in@@p4.75in@p0.25in@> p1.5in@124578This is a serious problem for Haegeman's (?) claim that CLLD doesn't yield intervention effects.If the problem is not the presence of an intervening dislocate, we are forced to conclude that the ungrammaticality of these examples must follow from structural deficiencies linked to the nonfinite character of the construction, namely from truncation, which is precisely what the intervention approach wants to avoid.
It must be remarked that these new data involving relative sentences and CLLD are not a problem for Haegeman's intervention approach alone but for Rizzi's (??) and Abels's (?) analyses as well, as we have shown in section ??.
To sum up, the resultant picture is certainly discouraging for intervention approaches, which must assume ad hoc assumptions or empirically dubious generalizations, at least for Spanish and Catalan.Such a negative conclusion does not mean that intervention effects between wh-elements and CLLD do not exist-we have shown throughout this section that they do-nor that they don't play a role in syntax, but rather that they cannot explain the full distribution of CLLD in relative and interrogative clauses, since the difference boils down to the [±finite] nature of the clause, not to different intervention patterns.

A New Proposal: Syncretism in the Left Periphery
We have shown beyond doubt that infinitive sentences allow less material in the left periphery than their finite correlates.Moreover, in the case of IWRs very little room exists for moving anything to the left periphery.Hence, the truncation proposal seems a simple solution, which becomes even more attractive when we confirm that the intervention approach cannot offer an explanation for the differences based on the [±finite] nature of the clause.However, we cannot assume a unified pruning analysis for both relatives and interrogatives.Even though this is certainly disappointing at first sight, I think it is a sensible conclusion.As ?originally remarked, the role of Force is crucial for linking the [+interrogative] feature of the sentence with the corresponding selectional requirements of the higher governing predicate (see also ??? for similar ideas, and ??? for an initial discussion of the issue of complementizer selection in the context of English infinitive relatives).Rizzi suggests that we have a double relation crucially mediated by Force: the [+interrogative] Force head is locally selected by the higher verb (see ? for discussion), whereas the interrogative wh-element landing lower in the CP area (here, in IntP) gets licensed by Agree through the standard probe-goal procedure (??).Schematically: Crucially, as originally remarked in ?:301 and repeated elsewhere, "the higher verb selects the specification of Force, not the TopP: verbs select for declaratives or questions, not for clauses with or without topic (or focus)."Hence, we can expect Force to remain active in infinitive interrogatives, with the consequences we have been reporting for TopP.
However, this is not necessarily the case for IWRs.Relative clauses are not selected, so the crucial mediating role of Force is less clear at this point, regardless of your favorite analysis of relatives (see ???).Moreover, Force has been linked to the typically selected sentence types, declarative, interrogative, and exclamative, not to the relative type, which is not associated with any particular illocutionary force.Hence, if truncation in IWRs is not constrained by selectional issues and TopP is linked to the presence of Force, a radical-truncation analysis seems natural, as represented in table ??.
ForceP TopP FocP TopP FinP TP Infinitive interrogative CLLD wh IWR wh Table 3: IWRs and infinitive interrogatives in Catalan and Spanish under the radicaltruncation analysis This can be implemented by means of features first and, only then, by functional categories, as suggested in ?. Specifically, the projection of the features involved in the CP domain allows several degrees of syncretism (see ?:100 for an original proposal involving the syncretic categories T/topic, T/focus, and T/emphasis; see also ? for a general view).For instance, we can have the maximally transparent distribution of features with interrogatives, where each feature associates with a functional head univocally: / 0 wh Obviously, this doesn't exhaust the possibilities, and other combinations are expected.For example, ?(see also (??)) has proposed a dedicated Q embedded P immediately dominating FinP for hosting (certain) wh-interrogatives in Italian that co-occur with a fronted focus (I mark the focus with italics and the wh-word with boldface): p5in@p0.25in@> p1.25in@@p5in@p0.25in@> p1.25in@124578 Accordingly, he proposes the following representation (where Mod is a projection hosting modal adverbs): The dedicated projection Q embedded P is unavailable in languages like Catalan and Spanish (I discard the modal adverb since it adds nothing to the discussion): p5.25in@p0.25in@> p1in@@p5.25in@p0.25in@> p1in@124578 In these languages, we can think of a syncretic Wh/Foc projection, in the spirit of ?'s (?) proposal, for which both the interrogative wh-word and the preposed focus compete.Moreover, the asymmetry between infinitive wh-interrogatives and IWRs just presented is consistent with the existence of infinitive prepositional relatives, which involve different specific complementizers (in boldface): p3.5in@p0.25in@> p2.75in@@p3.5in@p0.25in@> p2.75in@124578In Catalan and Spanish, they also select a specific complementizer, per 'for' and por 'for', respectively: 7 p4in@p0.25in@> p2.25in@@p4in@p0.25in@> p2.25in@124578 Crucially, this variability is not found with infinitive interrogatives, which involve the same yes/no interrogative complementizer 'whether' that we found in finite interrogatives: 8 p4.75in@p0.25in@> p1.5in@@p4.75in@p0.25in@> p1.5in@124578Hence, infinitive prepositional relatives can easily be integrated into our analysis as follows.

P RelOp
The preposition is the realization of the nonfinite complementizer, just as in other infinitive constructions across Romance (see ?:ch.14): p4.75in@p0.25in@> p1.5in@@p4.75in@p0.25in@> p1.5in@124578To sum up, the bulk of evidence clearly points toward a truncation analysis of IWRs, which only project FinP, thus lacking room for CLLD.In contrast, infinitive wh-interrogatives do project the whole left periphery, just like their finite counterparts, because the presence of ForceP is crucial for satisfying the selectional requirements of the higher predicate.The resultant picture is thus less homogeneous but more consistent on theoretical and empirical grounds. 9

Conclusions
This article has shown that the ban against CLLD in IWRs cannot be explained in terms of intervention effects but rather must be analyzed as a case of truncation in infinitive contexts.We have defended the view that, insofar as CLLD is a case of movement, the predictions of standard 7 Spanish is unique in allowing the complementizer que 'that' with IWRs: p3.5in@p0.25in@> p2.75in@@p3.5in@p0.25in@> p2.75in@124578 8 As ?:ch. 5 remarks, neither French, Occitan, nor Sardinian allows these constructions.Sardinian allows instead a prepositional complementizer de 'of', in contrast with the general tendency in Romance: p4in@p0.25in@> p2.25in@@p4in@p0.25in@> p2.25in@124578 9 As one anonymous reviewer points out to me, ?:292-293 also assume a combination of intervention and truncation for explaining some of the restrictions in the Spanish left periphery (see also ? for the interaction of truncation and intervention in Basque).Indeed, they explicitly assume truncation in raising constructions, whereas they must assume ad hoc defectiveness of discourse features for control structures.In the latter construction, C cannot transfer the relevant δ feature to T, which makes CLLD impossible in the specifier of TP.Unfortunately, it is unclear how their proposal would account for our crucial cases, namely the compatibility of CLLD with infinitive wh-interrogatives and the impossibility of CLLD with IWRs.intervention approaches cannot explain why IWRs contrast with their finite counterparts.Moreover, we have seen that truncation cannot be extended to infinitive wh-interrogatives, which do allow CLLD.The contrast between IWRs and interrogatives has been derived from the necessity of projecting ForceP in the latter, which are selected by a higher predicate.Finally, this crucial difference between interrogatives and relatives has been shown to be consistent with the existence of specialized prepositional complementizers for the latter but not for the former.
The resultant picture doesn't negate the role of intervention effects, which are real and evident, in the left periphery of relatives and interrogatives, but it places the burden of explanation primarily on the available structure.
Force Fin/Force [ TP PRO descansar]]] rest.INF 'a place to rest' If we adopt this solution, the theoretical problem simply fades away.

Table 1 :
Romance CLLD is found in many subordinate contexts, as in the following Spanish examples from ? (see ?? and ?for similar examples in Italian and Catalan, respectively): Radical-truncation analysis However, this analysis must face two problems.First, from a theoretical point of view, since truncation is the absence of structure above FinP, we are forced to assume that the wh-relative pronoun/adverb lands in ForceP in finite clauses but in FinP in infinitive clauses: The following sentences, kindly provided by Bettina Capello, are judged as marginal or plainly bad by several informants.
's Italian examples are reproduced in (30) and (31): the relatives in (30) are fine but the interrogatives in (31) are not.Cinque writes that "[t]he relative wh-phrase can plausibly be referential more easily than the interrogative wh-phrase, which behaves more typically like a nonreferential operator." boundary between CLLD and hanging-topic left dislocation is less clear than elsewhere, we have undisputed examples of CLLD affected by strong islands in all Romance languages, as originally pointed out by ?:408 for Italian: That you wrote to Giorgio means that you're still in love.' table, you are going to put.c. *Robin knows where, the birdseed, you are going to put.Force [ Top* [ Int [ Top* [ Foc a Gianni [ Top* [ Mod ieri [ Top* [ Q emb che cosa [ Fin [ IP .
The person whom they ordered John to kill was a Russian spy.' . Our examples clearly call this prediction into question.The person who (s)he didn't know who gave the book to was a Russian spy.' Certainly, this cumulative effect would be mysterious if CLLD didn't create intervention effects.