Comparative Study: Syntactic Restrictions on Multiple Interrogatives in English and German, Papers of Linguistics

Download Comparative Study: Syntactic Restrictions on Multiple Interrogatives in English and German and more Papers Linguistics in PDF only on Docsity! Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives Kleanthes Kostas Grohmann University of Maryland, College Park grohmann@wam.umd.edu Generals Paper Defense Draft LING 895 April 1998 Advisory Committee: Prof. Norbert Hornstein (main advisor) Prof. David Lightfoot (committee member) Prof. Juan Uriagereka (committee member) Syntactic Inquiries into Discourse Restrictions on Multiple Interrogativesii Table of Contents Abstract iii Acknowledgements iv 1 Introduction 1 2 Superiority Effects and “Operations Prior to Wh-Movement” 3 2.1 Multiple Wh and Superiority Issues 3 2.1.1 The Superiority Condition and Issues Regarding Multiple Wh 3 2.1.2 Past Approaches to Multiple Wh 5 2.1.3 A Minimalist Conception of Multiple Wh 7 2.1.4 Multiple Wh in German, Extended Data and More on Superiority Effects 8 2.2 Overt A-Movement Prior to Wh-Checking 11 2.2.1 Hornstein (1995): Raising and Multiple Wh in English 11 2.2.2 Takahashi (1993): Scrambling and Multiple Wh in Japanese 12 2.2.3 Boskovic (1993): AgrOP and Multiple Wh in Spanish (and Hebrew) 14 2.3 A Possible Application to German 15 2.3.1 The Agr-Complex in German and Scrambling Below the Subject 15 2.3.2 The CP-Domain: Pre-Subject Scrambling as Topicalization 16 2.3.3 Grohmann (1997b): Scrambling and Multiple Wh in German 18 3 DRQ as “Operations Instead of Wh-Movement” 20 3.1 D-Linking I Not Enough 20 3.1.1 Context in English and D-Linking 21 3.1.2 Context in German and DRQ 23 3.1.3 More Data on Contexts in German 26 3.1.4 Crucial Differences 30 3.2 DRQ: Multiple Wh Is Topicalization 33 3.2.1 Outset: The Problem 33 3.2.2 DRQ and Wh-Topics 35 3.2.3 The Rough Syntax 37 3.3 Extensions of Wh-Topics 42 3.3.1 Subjects and Multiple Wh: Some Consequences 42 3.3.2 Scrambling and Multiple Wh: In Response to Wiltschko (1997) 44 3.3.3 Wh-Adjuncts in Multiple Wh-Constructions: Some Speculations 48 3.3.4 Multiple Wh and Quantifier Interaction: A Reaction to Beck (1996) 52 Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives Kleanthes K. Grohmann (University of Maryland, College Park) Language is music. We always try to create that special note. M’hamed Benkreira 1 Introduction Preliminary inquiries into syntactic conditions on multiple interrogatives in the context of discourse restrictions constitute the main body of this study. They are “preliminary” in the sense that the leading idea—if on the right track—warrants extensive treatment in more depth and detail than the scope of this paper allows; and although (or: especially because) this paper is the culmination of work spanning the past year and a half or so (Grohmann 1997b, 1997c, in press), some issues remain to be addressed, while others have been added. The following sections are “inquiries” in the sense that while particular issues are investigated, some in more depth than others, much of it is still speculation (or at least open to discussion). One of the aims of this paper is to suggest a syntactic analysis of the phenomenon in question; however, as will become clear presently, the sensitivity of multiple interrogatives to discourse blurs a straightforward syntactic account to a high degree and thus the second goal of this study is to investigate some of these restrictions. This paper investigates syntactic, semantic and ultimately pragmatic (discourse) properties of Wh-elements in multiple interrogatives.1,2 In particular, I will look at Multiple Wh-constructions and consider an approach that involves movement operations prior to—or rather instead of—overt Wh-movement.3 While English and German data constitute an important part of the approach, the attempt is made to tie in these observations with facts from other languages; an appropriate typological validation of the present proposal, however, is another project altogether. This project will hopefully lead to further fruitful studies along the lines suggested here. It is amazing, though, to see how much mileage we can gain from elaborating on a few basic observations. 1 Some notational remarks: I will often refer to Wh-elements simply as “Wh” throughout this paper for expository purposes. Likewise, I will use the term “Multiple Wh” to denote multiple Wh-constructions. 2 Although I will use the term “Multiple Wh” throughout, it should be noted that I am primarily concerned with Double Questions (“Binary Wh” as termed by Pesetsky (1998)), i.e. interrogative constructions that contain exactly two Wh-elements. It is my belief that in order to understand the phenomenon of n-nary questions, we need to have a good grip on binary Wh-constructions. I hope to contribute towards such an endeavour. 3 As will become clear in the text, “prior to” does not refer to “pre- versus post-Spell Out” but rather to movement that targets a projection below the Wh-projection. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives2 The syntactic framework adopted here is the Minimalist Program in its (more or less) well understood current manifestation.4 As such, knowledge of the basic theoretical background and underlying assumptions will be largely presupposed. As for the semantics, the reader will notice that no particular approach is endorsed here in order to keep the details as plain yet explicit as possible, yet the general direction instigated by Karttunen (1977) with respect to interrogatives is adopted where the interpretation of questions is the set of propositions that jointly constitutes a complete true answer. While the details of the semantics involved do not play an important role here, the role of “sets of propositions” does, especially the referent sets of the two Wh-elements involved in Multiple Wh and their contextual restrictions cross-linguistically. What starts out as the standard paradigm, this study will be interested in how to explain the different grammaticality judgements for English versus German, as illustrated in (1) and (2):5 (1) a. Who kissed whom? b. * Whom did who kiss? (2) a. Wer hat wen geküßt? b. Wen hat wer geküßt? The leading idea is that context restricts the use and interpretation of Multiple Wh. In particular, (some) languages encode this restriction with a condition that these Wh-elements be syntactically marked as highly thematic; this results in overt topicalization of the two Wh-elements instead of movement to SpecCP (the projection containing the Wh-feature). I will call this property “Discourse-Restricted Quantification” (DRQ). It can be thought of as an overt movement-variant of D(iscourse)-Linking (Pesetsky 1987). In addition, this approach can also account for apparent asymmetries among languages with respect to observance of the Superiority Condition (Chomsky 1973). It has other properties, too, however which I try to disentangle somewhat in this study. I will proceed as follows. The next section will provide some background on the present phenomenon: the Superiority Condition will be presented along with the crucial data from English and German before considering the influence A-movement may have on Multiple Wh; this approach will then be extended and applied to German. Section 3 contains the main proposal, after a brief discussion of D-linking: DRQ is a special kind of D-linking (an overt movement-variant) and 4 The Minimalist Program is outlined in some depth in the collection of papers in Chomsky 1995 and much current work which can be found in volumes edite by Abraham et al. (1996), Abraham and Grohmann (1997), Wilder et al. (1997), Epstein and Hornstein (forthcoming), Martin et al. (in press) as well as studies by Collins (1997), Hornstein (1997), Kitahara (1997), Zwart (1997a), Chomsky (1998), Uriagereka (in press), to name but a few. Technical implementations—when relevant—will be introduced and clarified in the text. 5 Note that throughout this paper, I will give the German examples literally to the corresponding English ones; this will be marked either by referring the reader to the specific English example in the text or with the German example immediately following the English one. Only in cases where there are no corresponding English examples will proper glosses be provided. Kleanthes K. Grohmann 3 operates strictly in Multiple Wh in German; syntactically, DRQ can be characterized as topicalization of Wh-elements. In section 4, the proposal will be refined, more data will be presented in support of it and a possible implementation of DRQ will be pursued for other languages (and for child language); the basic conjecture is that D-linking has properties of syntactic, semantic and pragmatic topicalization. It will be concluded that these inquiries are preliminary indeed; yet, they offer plenty of research material for the future. 2 Superiority Effects and “Operations Prior to Wh-Movement” This section will lay out the basic empirical coverage of the phenomenon of Multiple Wh. First, the Superiority Condition will be introduced to account for some minimal pairs in English, before showing some apparent Superiority violations from German. Some analyses of Superiority will be discussed briefly, both from a GB-perspective and in a minimalist framework. After presenting a possible minimalist adoption of the phenomenon, recent case studies will be presented which will be extended as a first step towards accounting for the German facts. As will become clear, a possible connection between (multiple) Wh-movement and scrambling will be drawn which will be the starting point for further considerations. 2.1 Multiple Wh and Superiority Issues While the Superiority Condition holds well on a descriptive level for English, some data seem to suggest its absence in German. In order to investigate a possible asymmetry any further, let us establish a theoretical approach to the Superiority Condition and a good understanding of the data. 2.1.1 The Superiority Condition and Issues Regarding Multiple Wh Multiple Wh-constructions of the kind we are interested here involve two Wh-elements. One of the two elements is displaced (namely, under standard assumptions, fronted to sentence-initial position—COMP or SpecCP), while the other Wh remains in situ, in its base-generated position. The standard case for an English matrix question—and the same goes for embedded questions—involving two bare Wh-phrases is illustrated in (3):6,7 6 Please keep in mind throughout the paper that the desired interpretation of Multiple Wh is distributive (or pair- list), as will be explained in more detail presently. 7 The relevance of “bare” versus “complex” Wh-phrases will be illustrated shortly. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives6 Assuming Wh-movement at LF, the trace of whom is lexically governed by the verb kiss in (25a), while the trace of who is antecedent governed by who (made possible through COMP-indexing). On the other hand, the trace of who in the LF-representation (25b) is not properly governed: it adjoins to whom which percolates its index to COMP (not who); thus the ECP is violated and the construction (1b) is ruled out. This approach cannot, however, account for “Pure Superiority”, as pointed out by Hendrick and Rochemont (1982):10 (10) a. Who did you persuade to buy what? b. * What did you persuade who to buy? In the grammatical construction, both Wh-traces are properly governed at LF: persuade lexically governs the trace of who and buy the trace of what. But as can be trivially observed in (10b), the same government relations obtain. The generalization thus seems to be that only subjects (and adjuncts) fall somehow under the ECP, namely when they need to be antecedent governed. Based on this and other empirical shortcomings, Pesetsky (1987) proposes to evaluate well- formedness of Wh-movement on the basis of the Nested Dependency Condition (NDC).11 In cases of overlapping, the NDC forces one Wh-trace dependency to contain the other. This works well especially for cases in which two Wh-elements are generated in the same clause and either one could move to one of to SpecCP-positions:12 (11) a. What did you wonder who (PRO) to see t in t? b. * Who did you wonder what (PRO) to see t in t? Here, the path of the grammatical Wh-trace dependency is contained in the other, while in the ungrammatical case they overlap (“cross”). 10 Hornstein and Weiberg (1987) set out to account for these facts by invoking a number of further stipulations, making the approach look rather “artificial” (Norbert Hornstein, personal communication). 11 Or, as in Pesetsky 1982, the Path Containment Condition. 12 For reasons that will become clear in section 3, I want to use bare Wh-phrases in (11) in order to not confuse matters with D-linking (see below). Maybe the supposedly grammatical sentence does not sound as good as a corresponding sentence with two which-phrases, which shall be not our concern here, though: (i) a. Which moviei did you wonder which actorj (PRO) to see tj in ti? b. * Which actorj did you wonder which moviei (PRO) to see tj in ti? Kleanthes K. Grohmann 7 Furthermore, Pesetsky introduces the notion of D(iscourse)-linking. Some elements receive their interpretation not by moving to a specific position (to check a feature, in minimalist terminology), but by some way of linking to the discourse. Although the technical details of this approach may not be very clear at this point, the idea is straightforward: D-linked Wh-elements do not have to move to SpecCP (in the overt syntax). This accounts, for example, for the grammaticality of which-phrases (see the next sub-section). We will return to this in section 3. 2.1.3 A Minimalist Conception of Multiple Wh Let me briefly lay out how Superiority could be viewed from a minimalist perspective. The main motivation to reconsider past approaches is that the notion of “government” (which played a crucial role for the ECP) is eliminated. Thus the range of phenomena that the ECP could account for need to be re-evaluated to fit current assumptions. Recall that the main premise in the Minimalist Program for movement is that it takes place in order to check some morphological feature. If the feature is “weak,” movement must be delayed until LF (according to the economy principle Procrastinate) while a “strong” feature forces the feature to move overtly and pied-pipe lexical material for PF-reasons. As for locality and economy, the Minimal Link Condition (MLC) ensures that movement is confined to a well-defined range of possibilities. The MLC is defined as follows (from Chomsky 1995:311), interpreting “closer” as in (13), pace Chomsky (1995:299) and Kitahara (1997:15):13 (12) Minimal Link Condition K attracts α only if there is no β, β closer to K than α, such that K attracts β. (13) Closeness β is closer to H(K) than α iff β c-commands α, and β is not in the minimal domain of CH, where CH is the chain headed by γ, and γ is adjoined to H(K). The MLC rules out directly the standard cases of Superiority violations in English: K (in this case, following (12), C0 being marked [+Wh]) cannot attract α containing a Wh-feature (what) because β (who) also contains the relevant Wh-feature, is closer and thus intervenes.14 This is exemplified in (14) where the intervening, closer element (“β” from above) is marked in boldface: 13 Note that Chomsky (1998) dispenses with the notion of “(checking) domain” and seriously revises other minimalist concepts. As the current version of this paper is a draft, however, I will dispense with discussing many of its interesting implications. 14 Note that while generally accepted, the MLC is not uncontroversial, neither in its conception nor its definition. I follow one possible approach here. For alternative instantiations of the MLC with respect to Multiple Wh, see Fanselow 1997, Grewendorf and Sabel 1997, Müller 1997 among others; for a discussion that casts doubt over the MLC as a principle (as opposed to “Fewest Steps”), see Zwart 1996, 1997b. For reasons of space, I cannot treat either here. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives8 (14) a. [CP whoi C 0 [AgrSP ti bought what]] b. * [CP whati C 0 [AgrSP who bought ti]] Kitahara also adopts Huang’s (1982) “null pronominal hypothesis” for when/where-type of Wh-expressions. Interrogative adverbs such as when and where (among others) thus differ from why and how in that they could be analysed as being part of a prepositional phrase with an empty preposition. Huang gives as evidence the grammaticality contrast between from where versus *P why/how. He also considers the existence of pronominal non-interrogative counterparts. I will return to this briefly. Again, not even an MLC-type of approach rules out immediately the behaviour of why and how. In general terms, the MLC straightforwardly accounts for Superiority effects and suits well reformulating the Superiority Condition in current terminology. I will adopt the MLC as a general condition on movement and Superiority in particular, and extend the approach to capture the German data presented in the next sub-section. 2.1.4 Multiple Wh in German, Extended Data and More on Superiority Effects The same contrast as shown in (3) for English does not arise in German (cf. (1-2) above). Thus this subject-object asymmetry does not seem to hold for all languages as the Superiority Condition would suggest. Take (15), for example, which is the German equivalent of (3), repeated as (16). (15) a. Wer hat was gekauft? b. Was hat wer gekauft? (16) a. Who bought what? b. * What did who buy? The situation is the same as in English, i.e. in one case the Wh-subject moves, in the other the object; as in English, the Wh-subject can be seen as the “higher” element (with respect to the Superiority Condition) as it is base-generated “above” the object. Unlike in English, however, both operations yield grammatical sentences. This seems to suggest that the Superiority Condition does not hold for German, at least not in the same way that it does for English. But the situation is more complicated than shown so far. In particular, many of the ungrammatical English constructions are fine in German;15 also, exceptional conditions do not render some constructions ungrammatical in English. 15 In this paper, I will employ data from a northern dialect of German. To my understanding, grammaticality judgements about Multiple Wh-constructions in German vary between northern and southern dialects. I will rely on judgements from my own dialect which are confirmed by speakers from the region of East Westphalia. Kleanthes K. Grohmann 11 (29) a. Who did you persuade to buy what? b. * What did you persuade who to buy? (30) a. Wen hast du überredet was zu kaufen? b. Was hast du wen überredet zu kaufen? In these cases, there is no Wh-subject present; both Wh-elements are objects. From the observational viewpoint of the Superiority Condition, (29) can be accounted for straightforwardly: who is superior to what. And again, the Superiority Condition seems to be violated in German.18 I have presented data from English and German Multiple Wh-constructions. The set of data so far suggests certain asymmetries between the two languages as well as asymmetries arising among several constructions in English. The remainder of the paper will deal with the interpretation of Multiple Wh and the syntactic processes involved culminating in an alternative proposal in the narrow sense and to some degree its application from a more broader, cross-linguistic perspective. 2.2 Overt A-Movement Prior to Wh-Checking The intuition behind the following is very much in the spirit of some recent proposals: NP- movement (A-movement) and Wh-movement (A’-movement) may interact. In particular, I will suggest that scrambling may obviate Superiority effects. I will first show what A-movement can do to Multiple Wh in English, before looking at scrambling and Multiple Wh in Japanese and Spanish—and possibly Hebrew—, before looking at a possible implementation for German. 2.2.1 Hornstein (1995): Raising and Multiple Wh in English The standard instance of A-movement—alongside passivization—is raising. Moreover, unlike scrambling, (obligatory) raising can be observed in English, such as in (31): (31) a. * Seems (to his friends) Paul to enjoy commercials b. Paul seems (to his friends) to enjoy commercials The numerations of (31a) and (31b) are identical; this rules out the other option that English allows, namely to insert it in subject-position in (31a) and make the embedded clause to finite (as in It seems (to his friends) that Paul enjoys commercials). Thus, when the numeration is exhausted (at the stage depicted in (31a), for example), the embedded subject Paul must raise out of the non-finite clause into the subject-position of the finite matrix clause (31b). 18 I present the data regarding pure Superiority for the sake of completeness. I will not have to say much about these constructions though the analysis proposed in section 3 can also be used to account for these constructions. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives12 Interestingly, a Wh-subject in the non-finite embedded clause underlies the same constraint: (32) a. * Seems (to his friends) who to enjoy commercials? b. Who seems (to his friends) to enjoy commercials? The embedded Wh-subject cannot stay in situ, nor can it simply move to an embedded SpecCP.19 It must move to matrix SpecCP. It may shed some light on Multiple Wh to see how raising contexts change the grammaticality status; let us insert expletive it in one case and raise the Wh in the other: (33) a. * Who does it seem to whom enjoys commercials? b. Who seems to whom to enjoy commercials? In a Multiple Wh-constructions where the matrix subject-position is filled by it, the Wh-subject may not move from its embedded position to matrix SpecCP; in a construction where the matrix subject-position needs to be filled through Raising, it may. As a matter of fact, it must raise regardless of whether the raised element stays in this position. But in (33b), the MLC is violated: the closest element that C0 can attract is not who but to whom. (The same applies to (33a).) Now consider (34): (34) a. ? To whom does it seem that who enjoys commercials? b. * To whom seems who to enjoy commercials? In accordance with the MLC, to whom moves to SpecCP in (34a), forming a (more or less) grammatical sentence. But if the same operation takes place in the former grammatical (33b), the construction is ruled out (34b)—despite accordance with the MLC (presumably because who fails to be Case-marked). The observation to take with us from this example is that NP-movement prior to Wh-movement may have an effect on the application of the MLC and hence the relevance of the Superiority Condition. 2.2.2 Takahashi (1993): Scrambling and Multiple Wh in Japanese The intuition that scrambling and Wh-formation may interact was also pursued by Takahashi (1993). The generalization that emerged from his study is roughly that Japanese scrambling comes in two types, both of which differ with respect to Wh-elements. Short scrambling in Japanese can 19 For present purposes, it is immaterial whether this position exists, or what the status of vacuous movement is. Kleanthes K. Grohmann 13 be viewed as involving A-movement, displaying many of its characteristics, unlike long scrambling: scrambling across clause boundaries should rather be analysed as A’-movement much in the spirit of Mahajan’s (1990) cross-linguistic characterization of scrambling phenomena (also see Corver and van Riemsdijk 1994 among many other works on scrambling).20 The following shows both constructions (Takahashi 1993:663-664): (35) a. Dare-ga nani-o tabeta no? who-NOM what-ACC ate Q ‘Who ate what?’ b. Nani-o dare-ga tabeta no? (36) a. John-ga dare-ni [Mary-ga nani-o tabeta to] itta no? John-NOM who-DAT Mary-NOM what-ACC ate COMP said Q ‘Who did John tell that Mary ate what?’ b. ?? Nanii-o John-ga dare-ni [Mary-ga ti tabeta to] itta no? c. Pizzai-o John-ga dare-ni [Mary-ga ti tabeta to] itta no? While one construction exemplifies the possibility of short scrambling of two Wh-elements (35b)—with the base-order as in (35a)—the other one shows that this is not possible crossing clause boundaries (36b). Of particular interest is the contrast between (36b) where the Wh-element long scrambles, yielding ungrammaticality/strong marginality, and (36c) which is an instantiation of long scrambling of a non-Wh-element. What the Japanese examples show is that Multiple Wh-constructions differ depending on whether short or long movement takes place. As this is also a crucial distinction for purposes of scrambling, an interaction does not immediately seem implausible. It is also clear that movement of the Wh-elements cannot be driven by standard feature-checking (i.e. that a strong Wh-feature needs to be checked) as Japanese Wh-formation is in situ and the Wh-feature thus weak. 20 The nature of scrambling is not topic of this paper (but see section 2.3). The data and judgements are taken from the literature, and to my understanding variations in acceptability exist. This might gives us a clue that the input of A-relations in Multiple Wh is not as straightforward as the data seem to suggest initially. While scrambling most likely affects Multiple Wh, it cannot be the only influence. It seems to be established that the Wh-elements do not move overtly for the purpose of Wh-checking; also, scrambling allows free order of all arguments. With this in mind, the following (Akemi Masuya, personal communication) is unexpected: (i) a. Dare-ga Mary-ni nani-o ageta no? Who-NOM Mary-DAT what-ACC gave Q ‘Who gave Mary what?’ b. Dare-ga nani-o Mary-ni ageta no? c. * Nani-o Mary-ni darega ageta no? Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives16 While (40a) shows that all definite arguments leave VP (cf. (41a)), (40b) shows the same for specific indefinites (cf. (41b)); the specificity is made explicit by ein bestimmtes ‘a specific’. (42) highlights the contrast between an indefinite non-specific argument preceding sentential negation versus following it.25 (On the non-specificity of non-interrogative was ‘what/something’, see Haiden 1995 and the discussion in section 3.3.2 below.) Scrambling can then be seen as movement out of VP into the Agr-complex. In spite of many attempts to account for scrambling more accurately, no perfect solution is known to me. I thus adapt one of the available accounts and assume the ordering of the two AgrOPs (AgrDOP and AgrIOP for direct objects and indirect objects, respectively) to be free.26,27 This yields the following structural configurations: (43) a. [AgrSP Der Martini hat [AgrIOP der Mariaj [AgrDOP das Buchk nicht [VP ti gegeben tj tk]]]] b. [AgrSP Der Martini hat [AgrDOP das Buchk [AgrIOP der Mariaj nicht [VP ti gegeben tj tk]]]] In minimalist spirit, overt movement takes place for checking strong features. Here I follow current assumptions that movement out of VP takes place to check strong (D-, ϕ- and/or) Case-features in AgrP. The exact properties of Agr are not relevant for current purposes. 2.3.2 The CP-Domain: Pre-Subject Scrambling as Topicalization Scrambling as construed above (i.e. as feature-driven A-movement) is not a uniform process in German. As claimed in previous work, I assume there to be a substantial difference between scrambling below the subject and scrambling above the subject, henceforth pre-subject scrambling (Grohmann 1996, 1997d, in preparation). This relies crucially on the assumption that the subject in 25 As a matter of fact, indefinites under negation morphologically amalgamate nicht and ein: (i) Der Martin hat der Maria kein Buch/nichts gegeben the Martin has the Mary no book/nothing given ‘Martin didn’t give Mary a book/anything’ I am not concerned here whether this is evidence in favour of the indefinite object having moved out of VP over negation or whether negation sits higher than the object; even in the latter case it is not clear that the indefinite has not moved out of VP (see Haiden 1995). 26 Free order of the AgrO-projections is an adaptation of Haeberli’s (1995) proposal that all Agr-projections may be freely ordered. Independently, however, I assume the subject to be in a fixed position in relation to objects, which I mark here as AgrSP. This projection must be set higher than either AgrOP (see Grohmann, in preparation). 27 Note that one possible alternative may be free ordering of the base-generated arguments in VP. It is not directly relevant to the present issues how middlefield-scrambling can be accounted for and I will hence simply adopt the idea that only AgrIOP and AgrDOP may appear in any linear order. A rich body of literature exists about the nature of scrambling and the role of base-generation versus A- or A’-Movement (see, among many others, the collection of papers in Corver and van Riemsdijk 1994). To briefly mention yet another approach, Richards (1997) adopts the framework of multiple specifiers (Ura 1994, Chomsky 1995) about which I have nothing to say at this point. Some arguments against a multiple specifier analysis of German scrambling can be found in Grewendorf and Sabel 1997. Kleanthes K. Grohmann 17 German sits in a unique position in unmarked clauses (SpecAgrSP in the present configurations), an assumption that seems to be borne out (Grohmann, in preparation).28 The subject is arguably in its assigned position, SpecAgrSP. It thus functions as an indicator for the IP-boundary: all material preceding the subject is outside IP, in the CP-domain. I will refer to this as the left periphery. Following Rizzi (1997), CP could be conceived as an articulated domain, much in the spirit of the finer structure of IP, developed ever since Pollock 1989. The most relevant functional projection in this domain is a recursive Top(ic)P, independently established for German in recent work (see Müller and Sternefeld 1993, Haftka 1995). As head movement is not of much concern here (especially verb movement and the V2- phenomenon, see fn. 31), the following are the major projections including their intrinsic features, immediately relevant for present purposes: (44) CP [Wh] — TopP* [Top] — AgrSP [Nom] — AgrI/DOP [Dat/Acc] — VP [θ] All arguments leave VP (where they are θ-marked) and move to the specifiers of the respective Agr- projections (for Case-checking);29 arguments scrambled over the subject are in addition endowed with a Top-feature, moving to the specifier(s) of TopP(s). The Wh-feature sits on the head of CP.30 I will leave open the process of positioning adjuncts, but nothing a priori prohibits their presence in TopP. Implementing these ideas, we would receive something like the following representations (leaving out negation or other adverbs) for instances of pre-subject scrambling:31 (45) a. [TopP Das Buchi hat [AgrSP der Martin [AgrDOP ti [AgrIOP der Maria gegeben]]]] b. [TopP Der Mariaj hat [TopP das Buchi [AgrSP der Martin [AgrDOP tj [AgrIOP ti gegeben]]]]] c. [TopP Das Buchi hat [TopP der Mariaj [AgrSP der Martin [AgrIOP ti [AgrDOP tj gegeben]]]]] 28 Note that in Chomsky 1995:sec. 4.10 and much current work, the view of Agreement projecting is dispensed with altogether in favour of an implementation of multiple specifiers (Kuroda 1988). The main objection to AgrP is that it carries no interpretable feature and is hence not “virtually conceptually necessary” at the interfaces. However, assuming the existence of an AgrP does not necessarily involve its survival at LF, and alternatives to deal with interpretability issues are feasible (see Zwart 1997b). I will adopt the existence of AgrP for expository purposes. 29 It is Case, as indicated here; I leave out the possibility that this move simply checks D-features (which may even be conceptualized as the EPP) on DPs rather than Case. There is recent literature on either possibility and even more, all irrelevant for the present purposes. 30 Note that Rizzi also suggests a recursive TopP to the left of CP (FocP), an assumption empirically supported by data from Romance languages, among others, and also by German (as shown in earlier work, see Grohmann 1997a). I will return to the relevance of this projection later. 31 In this paper, I am not concerned with the specifications of deriving V2. Although it undoubtedly plays a crucial role even in these constructions, I will remain agnostic about the nature of movement or whether movement plays a role at all. In the light of this, I will not indicate possible traces of the verb in derivations. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives18 With freely ordered AgrOPs, both objects satisfy the MLC in movement out of VP.32 In that configuration, they also satisfy the MLC for further movement to TopP. The same, of course, applies to adjuncts as well. A temporal adverbial such as gestern ‘yesterday’ would also be topicalized when it appears in front of the subject.33 It is also shown that the same process takes place in embedded contexts (see Grohmann 1996, 1997d). (46) a. [TopP Gesternj hat [AgrSP der Martin tj das Buch der Maria gegeben]] yesterday has the Martin the book the Mary given ‘Yesterday, Martin gave the book to Mary’ b. Er sagte, [ForceP dass [TopP die Mariaj [AgrSP der Martin tj geliebt hat]]] he said that the Mary the Martin loved has ‘He said that Mary, Martin loved’ In (46a), the sentential adverb in pre-subject position sits in TopP (where I leave out the exact derivation, cf. fn. 33); in (46b) the fronted object moves there.34 2.3.3 Grohmann (1997b): Scrambling and Multiple Wh in German What the previous presentations have shown is that A-movement may obviate Superiority effects in (Multiple) Wh-constructions. One clear case is Raising in English, the others are scrambling (Japanese) or nominal feature-checking (Spanish); in order to consider a similar application—namely, movement operations prior to Wh-movement—to German, the status of scrambling needs to be addressed as well as the requirement of it being A-movement. For reasons of space, I cannot indulge in a deeper treatment here. I would like to hold fast to the resulting observation that there are movement processes that occur prior to Wh-movement, and whether it is solely constrained to A-movement is not relevant. German scrambling is such a process, and as we have seen above, pre-subject scrambling is topicalization. The following will be an implementation of scrambling and Multiple Wh in German. 32 Norbert Hornstein (personal communication) reminds me that the assumption of freely ordered AgrOP is not immediately compatible with recent minimalist proposals concerning movement, such as Richards (1997). Note, however, that Richards’ approach crucially builds on the availability of multiple specifiers in the grammar—an assumption that is not shared here (at least not with respect to German; see also Grewendorf and Sabel 1997 on this). 33 A severe difficulty arises when we want to account for adjuncts, simply for the reason that no satisfactory theory of their syntactic (and semantic) behaviour exists. One of the main problems is their position in the clause structure: (where) are adjuncts base-generated, (where) do they move, what categorial status do they have, are they right- or left-adjoined or both? Some proposals towards the syntax of adverbs or adjuncts in general by Laenzlinger (1996), Alexiadou (1997), Ernst (in press), Cinque (forthcoming) may give us a clue. 34 The position of the complementizer is here indicated as ForceP, following Rizzi’s finer structure of CP. This will not be relevant here, though the distinction of a position for Wh-elements versus complementizers might be. Kleanthes K. Grohmann 21 The following will contain a discussion of D-linking in English and an extended approach to German Multiple Wh. It will become clear that Multiple Wh underlie stricter contextual restrictions, here understood as Discourse-Restricted Quantification (DRQ). I will then propose a syntactic approach to these semantic and pragmatic constraints. 3.1.1 Context in English and D-Linking As a starting point, D-linking can be defined as a restriction on (multiple) questions where a felicitous answer can only denote sets of referents for the Wh-element(s) that have previously been established in the discourse.38 Such a view of D-linking immediately conjures up a correlation with partitives where the set of referents for the NP is already established. Thus, Wh-phrases of the types which N(s) and which of the Ns are inherently D-linked.39 This accounts for the grammaticality of (50-51) on the one hand and the grammaticality difference in (1a) versus (1b) on the other; who, what etc. are not—and in general, cannot be40—D-linked. The two major discussions in the literature regarding D-linking are the works by Bolinger (1978) and Pesetsky (1987). Curiously, both crucially differ in what is the trigger for apparent Superiority violations with D-linked phrases (Comorovski 1996:85): Bolinger takes the D-linked property of the fronted Wh to be the crucial factor, while Pesetsky claims it is the D-linking of the non-fronted Wh (Wh-in-situ). The following two situations from Bolinger 1978:108 and Pesetsky 1987:109, respectively, show how context may D-link a bare Wh (emphasis from Comorovski 1996:84): (52) a. * What did who break? b. I know that among all the disasters in that kitchen, Jane scorched the beans and Lydia put salt in the ice tea; but whát did whó bréak? I know that somebody broke something, so stop evading my question. (53) a. I wonder where what goes b. I know that we need to install transistor A, transistor B and transistor C, and I know that these three holes are for transistors, but I’ll be damned if I can figure out from the instructions whére whát goes! 38 Pesetsky (1987:107f.) formulates it as “[w]hen a speaker asks a question like Which book did you read?, the range of felicitous answers is limited by a set of books both speaker and hearer have in mind.” Comorovski (1996:2) takes this as the basis for her elaboration when calling “discourse-linked those Wh-phrases whose range of felicitous answers is limited by a set of objects already referred to in the discourse or salient in the context of utterance.” 39 It is a widely held view that D-linking and partitivity are alike in their semantic (and ultimately syntactic) properties. I will not explicitly discuss this here but refer the interested reader to the literature (see e.g., Kiss 1993, Comorovski 1996, Wiltschko 1997 and references cited there for discussion). 40 D-linking of bare Wh-phrases is possible under heavy stress, for instance, or in even more restricted context. See Baker 1970 and Bolinger 1978, among others, and the discussion by Pesetsky (1987). Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives22 In both contexts, the fronting of the lower Wh is possible; but do these examples show that it is the D-linking of the fronted Wh (52b) or of the lower Wh (53b) that allows this? Presumably not; remarkably (as also pointed out by Comorovski), neither author employs examples with one D- linked and one bare Wh to support their claims. Besides, (53a) is good even without context (see section 2.1.3 above).41 Now consider the following sets of questions. (54) a. * Who did who kiss this week? b. * What did you persuade who to buy? (55) a. * Which girl did who kiss this week? b. * Which book did you persuade who to buy? While (54) repeats the by now well-known Superiority facts, (55) shows that if a D-linked Wh is fronted over a bare Wh, grammaticality does not automatically arise. On the other hand, if a non-D-linked Wh is fronted over a D-linked one (56), the question is acceptable.42 (56) a. Who did which boy kiss this week? b. What did you persuade which of your friends to buy? The data above show—if anything—that context plays a crucial role; in how much we can make any appropriate generalizations, however, differs from the individual viewpoint. Comorovski argues that the leftmost Wh must be D-linked. This carries over to bare Wh: 41 The other example Pesetsky (op. cit.:109) gives is (ib) which does indeed contrast with (ia): (i) a. * What did who do? b. I know what just about everybody was asked to do, but whát did whó (actually) do? While (ia) is not part of any specific context, a rough reference of both Wh-elements in (ib) has been introduced in the discourse. The crucial role of this will be explored in much more detail in the following. 42 It is well-known to anyone who studied Multiple Wh in some depth, that judgements are very intricate. My own (limited set of) informants agree with the judgements given here for (55-56). Comorovski (op. cit.:85), however, presents the following judgements to show that neither Bolinger nor Pesetsky were correct with their accounts: (i) a. ? Which book did how many people buy? b. ? What did which student read? The conflicting judgements on the various sets of data (i.e. fronted bare Wh and D-linked Wh-in-situ, fronted D- linked Wh and bare Wh-in-situ, fronted bare Wh and bare Wh-in-situ, fronted D-linked Wh and D-linked Wh-in-situ) can only lead to the conclusion that something else besides D-linking is also at stake in English. Kleanthes K. Grohmann 23 (57) a. It’s nice to have all those times scheduled, but when are you doing what? (#But what are you doing when?) b. It’s nice to have all those activities ahead of you, but what are you doing when? (#But when are you doing what?) Comorovski (1996:143) illustrates with these examples taken from Bolinger 1978 that wide-scoped Wh-phrases are always D-linked. Here we encounter a dilemma, however: wide scope is established by movement to CP (where it takes scope over the entire clause); on the other hand, Pesetsky argues that D-linking does not induce movement. I will return to this after an exposition of the German facts and an alternative approach to Multiple Wh. Compare lastly (57), where bare Wh-phrases may also be D-linked, with the situation depicted in (58). Bare D-linked Wh-phrases differ from inherently D-linked Wh-phrases, in that only the latter may appear in any order:43 (58) It’s nice to have all those activities scheduled for all those times, a. … but at which time are you going to do which activity? b. … but which activity are you going to do at which time? To summarize our discussion of English multiple interrogatives, context may obviate Superiority effects by D-linking Wh-phrases; to which extend it may is yet unclear. After a thorough discussion of the situation in German, we will be able to describe English Multiple Wh more accurately which will also lead us to a more satisfactory account of the facts. 3.1.2 Context in German and DRQ In German, Superiority effects cannot as easily be found as in English. I propose on the basis of the following data and contexts that D-linking occurs in a much stricter version, which I call Discourse-Restricted Quantification (DRQ): the quantificational force of both Wh-elements in a German Multiple Wh-construction is severely restricted by the discourse.44 43 But note the contextual difference between (57) where only one of the Wh-referents is mentioned and (58) where both possible referents are introduced, as illustrated in (i): (i) It’s nice to have all those activities scheduled for all those times, a. … but when are you going to do what? b. # … but what are you going to do when? If two bare Wh-phrases are used in the context of (58), only the order when…what is felicitous (presumably due to the proximity of the referent of the fronted Wh, namely all those times). 44 It is generally assumed that Wh-elements are quantificational. Under an interpretive approach to quantification (and hence Wh)—as suggested by Postma (1995), for example—a Wh does not carry its interpretation inherently but receives it; it is not unreasonable to assume that this reception is the result of checking a particular feature. While Wh-phrases are often taken to be indefinites, their interpretation may vary: they may be existential or quasi- Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives26 b. Ich weiß, daß wir Transistor A, Transistor B und Transistor C installieren müssen, und ich weiß, daß diese drei Löcher für Transistoren da sind, aber der Teufel soll mich holen, wenn ich aus der Anleitung herausfinden könnte, wo was hin geht! The judgements are as predicted, both in English and in German. While (52b) is rendered grammatical due to the context (in contrast to the ungrammatical (52a)), (62a) is not grammatical (felicitous) because it does not conform to DRQ: no reference has been made to the breakers or the broken items.48 Likewise, (53b) is okay (as is (53a))—thus not showing much for English in the first place—and the German equivalent (62b) is, too: reference to both, the where and what has been introduced in the discourse.49 The other contexts given above for English concern (58) (and (i) in fn. 45), translated here: (63) Es ist ja schön, all diese Aktivitäten zu all diesen Zeiten geplant zu haben, a. … aber zu welcher Zeit wirst du welche Aktivität machen? b. … aber welche Aktivität wirst du zu welcher Zeit machen? a. … aber wann wirst du was machen? b. … aber was wirst du wann machen? As both, a reference to time and to activities, has been made (all diese ‘all these’, referring to something known to speaker and hearer), (63a-d) are all well-formed, felicitous questions: with the subject which N-phrase preceding the object and vice versa as well as either order for the bare Wh- phrases. (All other examples from above are fine when DRQ is strictly observed.) 3.1.3 More Data on Contexts in German One might conclude that felicitous German Multiple Wh necessarily involve definite DPs or referring expressions, as can be seen in (60). This is not the case, however: (64) Situation III: A sex shop owner tells a reporter of items he sold that day and his customers. “I just sold a dildo, a magazine and a porno. A lesbian, an old guy and a pretty hooker bought all that stuff, but it really could be anyone coming in and buying stuff here.” 48 This question is only felicitous when DRQ is observed, such as in (i) where the breakers and the broken items are mentioned prior to the question aber was hat wer zerbrochen? ‘but what did who break?’: (i) … Ich weiß auch, daß Peter, Paul und Maria die Teller, die Tassen und die Schalen zerbrochen haben; aber… … I know also that Peter, Paul and Mary the plates, the cups and the bowls broken have; but… ‘… I also know that Peter, Paul and Mary broke the plates, the cups and the bowls; but…’ 49 As we have seen in section 2.1.3, there is no asymmetry in English (nor in German) with where/when and Wh- arguments. But this is irrelevant here, as reference to both Wh-elements has been introduced. Kleanthes K. Grohmann 27 a. Who bought what? b. * What did who buy? c. Wer hat was gekauft? d. Was hat wer gekauft? In this situation indefinite DPs denote the sold objects and the same acceptability/grammaticality judgements apply as in (60). But on top of the information of sold objects (cf. (59)), the buyers of these items are also mentioned in (64), thus priming for a felicitous double question in German. This supports the view that at least specificity is involved in D-linked contexts (see, among many others, Kiss 1993 or Comorovski 1996) and further that interrogative Wh-elements are indefinites (standardly assumed ever since Karttunen 1977). I will return to this discussion. The main conclusion one can draw from the previous examples is that a felicitous Multiple Wh-construction in German depends on the context which may depend solely on the situation or more accurately, as exemplified here, on the information provided by the statement priming for the question. The quantification of the Wh-elements is thus restricted by the discourse, in other words: the Wh-elements underlie Discourse-Restricted Quantification.50 This is not the case in simple Wh-questions which extends to German as well. Compare the following with (59) from above: (65) Situation I: (from above) “I sold a book today.” a. Who bought it? b. Wer hat es gekauft? 50 To come back to the remark from fn. 9 that the referent sets must be more than one for each Wh, it may be relevant to mention Uriagereka’s (1988:209ff.) observation regarding multiple questions, as also brought to my attention by Howard Lasnik (personal communication). He distinguishes between “incriminatory” and “inquisitory” questions: while the latter is the type of Multiple Wh that we are interested in here, the former gives a framework for those types of Multiple Wh where each Wh has one referent and the purpose of the question is to assign an ordering among them. Thus imagine a situation where two boys come crying to their mothers, obviously having beaten each other. If the mother asks (i), she knows exactly that one of the boys hit the other, but she does not yet know who started the fight: (i) Who hit whom? This question pairs the referents of the Wh-elements but does not intend to inquire more (i.e. it does not yield a list), whereas in the cases shown here, an exhaustive pair-list answer is intended. It might be the case that Uriagereka’s incriminatory question type extends further than the case illustrated here (or even in Uriagereka 1988). Note that in the German equivalent, only (i) is fine; the Wh-object fronted is ruled out. This clearly shows that these types of questions are different from the type of Multiple Wh under investigation here. It is not immediately clear in how much this distinction bears on the object of the present study. These are speculations, however, and should be part of a broader investigation than can be pursued here. Thus, let us constrain the present goal to inquisitory-type Multiple Wh where the intended answer pairs a list of Wh-referents containing more than one referent for each Wh exhaustively. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives28 In both cases, the simple Wh leads to the formation of a grammatical and felicitous question, although the set of referents is anything but known. This is the classical case of a Wh: one person does not know the referent set of some entity and inquires it from someone who knows; in this sense, the Wh is an indefinite expression. Let me finally show that the discourse restriction set up by the information provided not only applies to a Wh-subject and a Wh-object, but more generally to all kinds of double questions, involving any two Wh-elements: arguments (DPs and PPs alike) as well as adjuncts (temporal, locative or any other) and combinations of all of them.51 In (66), the question seeks to get a distributive answer for the direct and indirect objects of a ditransitive verb. (66) Situation III: (from above) “This guy gave away his mag collection, a special ring and handcuffs.” a. To whom did he give what? b. * What did he give to whom? c. # Wem hat er was gegeben? d. # Was hat er wem gegeben? (67) Situation III: (from above) “This guy gave away his mag collection, a special ring and handcuffs. Remember the biker lady? And this old fart and the young high school teacher? He gave it to them.” a. To whom did he give what? b. * What did he give to whom? c. Wem hat er was gegeben? d. Was hat er wem gegeben? The same restrictions we have observed above apply: both referent sets must be given in the discourse in German, but not in English (contrast with (67)). The same restriction holds for the combination of an argument and adjunct: 51 Note that all examples provided here involve matrix questions. As Norbert Hornstein (personal communication) points out to me, embedded questions possibly behave differently. As this study can only be preliminary in many respects, deeper considerations of embedded questions must be left for further research. The fact that topicalization in embedded contexts is not as well-formed as in matrix clauses (at least in English) may bear on this. Kleanthes K. Grohmann 31 With D-linked and non-D-linked Wh-elements involved in double questions, four possible orders arise, as already noted. These possibilities are furthermore enriched by (language-internal) asymmetries between subject and object, for example, or certain Wh-adjuncts. Concentrating on the subject-object asymmetries in English, the above data allow the following categorizations for fronted non-D-linked Wh-elements: (74) a. Non-D-linked Wh-SU… Non-D-linked Wh-OB Who bought what? b. Non-D-linked Wh-SU… D-linked Wh-OB Who bought which book? c. * Non-D-linked Wh-OB… Non-D-linked Wh-SU What did who buy? d. Non-D-linked Wh-OB… D-linked Wh-SU What did which girl buy? Superiority is obeyed in (74a) and (74b), where the Wh-subject is fronted; it is apparently violated in (74d), where the Wh-object is fronted; in this case, however, the Wh-subject is D-linked. That this is crucial can be seen from (74c) which is ungrammatical. The same list can be construed for fronted D-linked Wh-phrases: (75) a. D-linked Wh-SU… D-linked Wh-OB Which girl bought which book? b. D-linked Wh-SU… Non-D-linked Wh-OB Which girl bought what? c. D-linked Wh-OB… D-linked Wh-SU Which book did which girl buy? d. * D-linked Wh-OB… Non-D-linked Wh-SU Which book did who buy? The inverse is the case here: a D-linked Wh-object cannot be fronted over a non-D-linked Wh- subject (cf. (75c) vs. (75d)). In other words, English questions involving a Wh-subject and a Wh- of Multiple Wh cross-linguistically. While some suggestions are made here, deeper issues may involve mental representations of the respective Wh-elements in the lexicon of speakers. The current treatment can only be seen as one attempt to uncover the special role of D-linking in this respect. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives32 object seem to not permit a non-D-linked subject Wh to remain in situ. Interestingly, even this generalization can be violated:56 (76) a. * What did who buy? b. * Which book did who buy? c. Which of these books did who buy? In (76a), the bare Wh has no properties that allow it to be fronted over the Wh-subject, while the Wh in (76b) is at least D-linked; this, however, is not enough in English to allow fronting over the bare (= non-D-linked) Wh. (76c), on the other hand, shows that fronting of a partitive Wh-object over a non-D-linked Wh-subject is grammatical. The upshot of the intricate English data is then that the Superiority Condition rules out fronting of a lower bare Wh over a higher one; this movement can only occur to check a Wh- feature, and the higher Wh is closer to CP than the lower one (MLC). Superiority effects can be obviated by D-linking, where no Wh-movement applies; constraints on ordering Wh-elements underlie unselective binding (D-linking). Furthermore, partitive Wh-phrases may move even if another Wh is apparently closer to CP; this, I will show, can also be found in German (DRQ). I will return to all three accounts after a thorough discussion of a syntactic analysis of DRQ and subsequently “Wh-topics.” The observations concerning German Multiple Wh can be summarized as follows. If the possible referent sets for both Wh-elements has been introduced in the discourse, anything goes; if no or only one referent sets has been established, nothing goes (DRQ). Unlike English, German allows this special kind of D-linking for all Wh-elements. The task is now to (i) give a syntactic account for DRQ, (ii) find evidence for this approach from other languages, and (iii) establish a connection between the variation in Multiple Wh among languages. I suggest to (i) understand DRQ as topicalization, (ii) look at Chinese for Wh-topics, and (iii) consider the different syntactic realization of semantic and/or pragmatic factors across languages. 56 I would like to thank Jennifer Graham for pointing out to me the special role that partitives play. I will return to this contrast below. While it has been observed that D-linking and specificity, and even partitivity, are connected in some sense, there is no consensus about this interaction; furthermore, to my knowledge, the contrast in (76) has not been noted in the literature. (Kiss (1993) notes that partitives are easier to extract out of Wh-islands, as is well- known by now; Wh-islands will be addressed in section 4.2. I will return to the role that partitives play in general below. Kleanthes K. Grohmann 33 3.2 DRQ: Multiple Wh Is Topicalization After clarifying the issues involved with Multiple Wh-constructions and D-linking/DRQ, I will propose to analyse DRQ as overt topicalization, i.e. that in German Multiple Wh-questions all Wh- phrases move overtly to the specifier of (a recursive) TopP, situated between AgrSP and CP (as part of the CP-domain following Rizzi (1997) and others). As a consequence of how DRQ has to be interpreted, some (related) instances of English D-linking also involve topicalization. 3.2.1 Outset: The Problem It is by now pretty well established that in a simple Wh-question the Wh moves to SpecCP. Languages differ in this respect (see section 2.1.1) whether they move overtly (before Spell-Out) or covertly (at LF). This is also the standard view for multiple Wh-questions. However, D-linked Wh- phrases are not assumed to move to CP. One property of D-linking is that it does not require movement for interpretation, but rather that the D-linked Wh is interpreted in situ, via unselective binding (following Lewis (1975) and Heim (1982) regarding unselective binding for indefinites as well as Pesetsky (1987) and subsequent work for implementation to the syntax of Wh-elements). With the previous discussion in mind, some problems immediately arise. For one, it is not exactly clear what “unselective binding” means in a minimalist framework. Or more to the point, is it really enough to say that some Wh-phrases receive interpretation by movement to SpecCP, while others do not move anywhere at any stage? What is it that allows discourse to bind unselectively? It would be nice if we could pin-point the special discourse-properties down to something else. In the minimalist framework, where relevant properties are understood as morphological features, D-linking must correlate to some feature. While it has been argued that D-linking relates to specificity (Kiss 1993), others argue that it correlates to partitivity (Comorovski 1996).57 If a D- linked Wh would be a partitive Wh, we would not be able to account for the contrast in (75), repeated slightly extended below as (77-78): (77) a. * Which book did who buy? b. * Which books did who buy? (78) a. Which of the books did who buy? b. Which of these books did who buy? 57 David Pesetsky (personal communication) claims that the relevant property of D-linked Wh-phrases is not specificity but definiteness. However, I have not seen the relevant empirical data or theoretical support for this view. The standard definitions of both properties are vague enough to allow for this assumption; given that I have not seen proposals in favour of definiteness in the literature (unlike specificity or partitivity), I will ignore this for the remainder of the discussion. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives36 (83) a. [DP which [+Wh] of [DP those [+Top] students]] b. DP which D’ [+Wh] D0 PartP Part’ Part0 DP of those D’ [+Top] D0 NP students The bracketed structure in (83a) shows the unchecked features, namely those DP-features that need to engage in a feature-checking process in an appropriate configurations with another head bearing the same feature. This is clarified a little in (83b): partitivity itself is satisfied in its own phrase, part of a deeper DP-structure; specificity is probably saturated inside the lower (specific) NP students. These features are thus checked DP-internally (like nominal NP-features are checked DP- internally). Furthermore, I take [Top] to be a feature of the complement of Part0; in how much this assumption can be verified across a larger set of partitive construction is beyond the scope of this study.60 As a result, the two features construed with the two DPs remain to be checked via movement. As both DPs are part of the same DP-constituent, the assumption is that [Top] percolates to the top of the projection and is the feature that needs to be checked first. Applying the structure in (83b) to all partitive interrogative phrases, DRQed Wh-phrases contain two features that need to be checked via movement: a Wh- and a Top-feature. As the Top- feature needs to be checked first, the relevant functional projection to look at is TopP. The part that has been assumed to denote specificity and partitivity will thus be taken one step further: it represents given information—a property well-known from topicalized elements. 60 The correct, more detailed structure of these are not relevant here; no claim is made that this is enough of a representation. The important properties of (83b) to note are the two DP-features ([Wh], [Top]) and the (not unreasonable) assumption that all other features are satisfied. Kleanthes K. Grohmann 37 Analysing a DRQed Wh as a topic has syntactic consequences alongside the well-established semantic ones. Are these consequences desired at all? Do we need them? The answer to these questions will, of course, be “yes.” Before going on, let me briefly mention three cases in support of this view, one of which we have seen already (with the other two to be outlined in more detail below). I take these observations to be good reasons to pursue an approach to Multiple Wh based on topicalization. Details will be laid out in the following. Firstly, we have seen that contextual restrictions apply to the felicity—and ultimately grammaticality—of Multiple Wh-constructions. Thus different semantic and pragmatic properties directly lead to two different syntactic outputs (as described in detail in section 3.1 above). Secondly, there is good evidence in Chinese that certain constructions involving a Wh are actually instances of Wh-topicalizations. On top of the evidence in favour of such an analysis, a closer look to the discourse restrictions upon these constructions shows striking resemblances with those operative in German. I will lay out Wu’s (1996) study in some detail in section 4.1.1 below. Thirdly, Wh-extraction out of Wh-islands seem to be cross-linguistically only possible with “heavy” or D-linked Wh-phrases (see Comorovski 1996:ch. 5 for discussion). Looking at this phenomenon, we will be able to make generalizations about the possibility of extraction on the grounds of different structural positions—and hence, different syntactic processes—involved. I will return to this issue in section 4.2 below. 3.2.3 The Rough Syntax Let me propose a rough outline how to implement topicalization to German Multiple Wh.61 The restriction on possible referents has the flavour of a topic; namely, it refers to given information. As implied by the definition of DRQ, this property seems to take scope over the quantificational force of the Wh-element: it restricts it (see fn. 44 above). In a (minimalist) framework that expresses all relevant information in terms of formal features, topichood is an instantiation of positioning the topic in such a way that it can be licensed, namely by entering it in a specifier-head relationship with a topic-feature. The formal feature [Top] heads its own projection. The locus of this is presumably inside the CP-domain (following Rizzi (1997)) and the appropriate structural configuration looks as follows: 61 Please bear with me. This section presupposes the existence of Wh-topics and their presence in German Multiple Wh. I will lay out their syntax here, then expand on its problems and consequences. In section 3.3.2, I will show that one objection to this approach can be refuted in its own terms (Wiltschko 1997). Section 4.1.1 will contain a convincing discussion that Wh-topics exist (Wu 1996). The presentations following will support the crucial role that topicalization plays in a range of constructions. The upshot is that while the direct positional evidence for both Wh-elements to sit in TopP may not be as strong as one could hope for, consequences of this approach are. I hope to return to the former in future research, while lending even more support to the latter issue as well. Please remember that the present lines are inquiries in the strictest sense. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives38 (84) TopP* XP Top’ [Top] Top0 … [Top] The topic-feature which is part of the feature-bundle of XP (the topic phrase) must be checked off against the topic-feature of the head; this is indicated in (83) by strikethrough typeface.62 As shown by Rizzi, languages may employ more than one topic per sentence (thus having multiple topic projections, indicated as TopP*); they may also differ with respect to the position of TopP* relative to CP: in some languages TopP precedes CP, in others TopP follows CP—and yet other languages may have TopP* preceding and following CP; in principle, there could also be languages where topichood is not checked in its own projection in the left periphery. German has the option of moving a topic to a position preceding as well as following CP (such as Left Dislocation, as proposed in Grohmann 1997a), with preference for the latter; in both, English and German, a topic may also appear base-generated before CP: we can witness such constructions in Hanging Topics (see Grohmann 1997a for discussion and the collection of papers in Anagnostopoulou et al. 1997). In the case of a Wh-phrase subject to DRQ, the restriction is expressed on the Wh-phrase and can be conceptualized at least as follows (cf. (83) above): (85) [Wh [Top]] Thus, whenever the set of referents (that is, referents that are part of a felicitous answer to a given question) is restricted, this information is encoded on the Wh-phrase and needs to be checked off. Moreover, the encoding can be thought of as a topic-feature; as the Top-feature needs to be checked off before the Wh-feature, it is the former feature only that needs to be checked (see (83b) and discussion above). This drives DRQed Wh-phrases to move to the specifier of TopP, prior to further movement to satisfy the Wh-feature on C0. The following shows a sample derivation (cf. (48c)): 62 For present purposes, further information such as the instantiation of topic-comment structure, or even focus- background will be irrelevant here; for proposals, see among many others Vallduví 1990. Kleanthes K. Grohmann 41 Wh-elements in German Multiple Wh are Wh-topics and undergo a derivation in accordance to DRQ. This means that the underlying, deeper structure associated with these Wh-elements is invariably that of partitive Wh-phrases, i.e. the one shown in (83b). DRQ is then a more powerful version of D-linking: while the latter has the force to interpret Wh-elements, the former assigns deeper syntactic structure based on semantic and pragmatic properties.65 But English may also employ this strategy: namely, in exactly those cases, where the Wh is inherently partitive (83a). (88) a. Which of these students bought which of these books? b. Which of these books did who/which student/which of these students buy? (89) CP DP i/DPj C’ [Wh] C0 TopP [Wh] ti/tj Top’ [Top] Top0 IP [Top] ti/DPi I’ I0 VP ti V’ V0 DP/tj a. Wh-SU bought Wh-OB b. Wh-OB did Wh-SU buy On analogy with the previous derivation, (89) represents the derivations for (88).66 The partitive Wh needs to check both, the Wh-feature and the Top-feature. The generalization, then, is that all 65 This is just one way of interpreting the facts, i.e. that the properties of the clausal type (multiple interrogation, either as D-linking in some contexts or as DRQ in all contexts) directly influences the properties of the Wh- elements. Let us stick to this picture of the interrogative world for expository purposes. 66 Some details are completely irrelevant here: I thus label the subject-position in English IP and employ a very simple VP-structure, which suffices to show that the subject moves from its VP-position to a higher position, before moving to TopP and CP (analogously to German) in one case, and the object moves overtly to TopP—as in Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives42 German Wh-elements in Multiple Wh-constructions (i.e. under DRQ) have the pragmatic, semantic and syntactic properties of partitives/topics, while only overt partitive Wh-phrases have these properties in English. Further restrictions apply which I will return to. Let me briefly mention additional evidence—tentative, though, they may be—from English that the general Wh-Topic approach suggested here can be extended to cover certain facts in English, too.67 (90) a. What kind of bike, it was? b. Hey, hey LBJ, how many did cigs you smoke today? If some specially D-linked Wh-phrases in English may also contain [Top], one might analyse (90a) as an “incomplete” derivation where for some reason the last step, i.e. that of checking the Wh- feature, was left out. Likewise, (90b) can be understood as checking of the Top-feature by the whole Wh (how many cigs) and further Wh-checking by only the interrogative part (how many); here, the topicalized part (cigs) stays in TopP. 3.3 Extensions of Wh-Topics The above exposition reveals the direction that the idea of DRQ conceived of as topicalization of Wh-phrases heads in, but it needs refinements. Some of these will be provided here. 3.3.1 Subjects and Multiple Wh: Some Consequences One question that immediately arises is whether we can decide which Wh-phrase is topicalized or whether there can be more than one Wh-topic. Given the conception of DRQ for double questions as presented in (61), we would expect both Wh-elements to move to SpecTopP. The sample derivations shown above build on this assumption also. This is precisely what the following discussion will be concerned with. To yield a derivation that conforms to the MLC it is the Wh-phrase from the lowest AgrP that moves to the highest TopP. This is what we have seen in (87) already.68 German—and then to CP in the other. It is also irrelevant which features are checked when, especially whether the object checks Case- or D-features on its way to TopP. 67 These examples were found by the author. (79a) was overheard from a policewoman in College Park, MD when inquiring about a stolen bike; by the time of uttering the question it was already established that the topic of the discourse was a specific bike. (79b) was found in The Smoking Life by Ilene Barth (Columbus, MS: The Genesis Press, 1997) on page 80 as cited here. If nothing else, the reader is encouraged to smile at this point;-) 68 Recall that AgrOP is assumed to be ordered freely. The MLC, then, straightforwardly forces the higher object to move to the lower TopP (where the lower Top0 is introduced first, attracting the element in the higher AgrOP). Kleanthes K. Grohmann 43 (91) a. [CP Wenj hat [TopP tj [TopP wemi [AgrSP der Martin [AgrIOP ti [AgrDOP tj vorgestellt]]]]]? whom has to-who the Martin introduced ‘Whom did Martin introduce to who yesterday?’ b. [CP Wemj hat [TopP tj [TopP weni [AgrSP der Martin [AgrDOP ti [AgrIOP tj vorgestellt]]]]]? The two Wh-phrases in (91a) are ordered direct object—indirect object. The surface derivation is derived by ordering AgrIOP above AgrDOP. The opposite is the case in (91b). Here, we are confronted with another question. If German Multiple Wh is always an instance of topicalization (driven by DRQ), should the subject not also move to TopP? The reasoning behind this is, of course, that if two non-subject Wh-phrases occur with the subject in between and both Wh-phrases are in TopP, we need to account for the subject-position. Consider (23a), repeated as (92a), and an alternative to (91) as in (92b): (92) a. Wen hat Martin warum geküßt? b. Wen hat Martin wem vorgestellt? For cases such as (92), I do indeed propose that the subject moves higher to the specifier of a further TopP. We assume the subject-position to be fixed, an assumption that is certainly shared by other linguists (starting in the minimalist framework at least with Zwart 1993 where the subject needs to move to AgrSP for feature-checking); if the subject is not tied to one structural position (e.g., AgrSP), then it is at least with respect to its relation to the other arguments (Grohmann, in preparation). But in cases such as (92), it is plausible that the subject moves further, once the relevant features are checked by AgrS0.69 I thus take the subject to be inherently marked with the property of (thematic) topichood which it can “check off” in AgrSP. The reason for the inverted commas is that I take this checking process to be inherent, not formal. The underlying assumption is the thematic dominance of the subject which manifests itself by virtue of being clause-initial. When a clause is not subject-initial, this inherent thematic checking is overridden: in this case, the subject has to move to the specifier of a TopP in order check the relevant thematic feature; and only in this case: following Zwart (1997a and earlier work), subject-initial declarative clauses do not invoke the CP-structure. A Wh-subject may be interpreted as “Wh-in-Agr,” without further movement for exactly this reason. But when there is one Wh-phrase preceding and one following the subject, AgrSP does 69 This is also motivated independently: as is well-known, the subject always bears a certain notion of topichood, in the vague sense of “thematic topic” as outlined here (for example, see Erteschik-Shir 1997 for an elaborate discussion and consequences for her focus-structural approach to synthesize syntactic and semantic conditions, to name just one recent study; also see Comorovski 1996 for a similar concern regarding subjects and Wh-questions) which in the common cases receives its interpretation in the subject position. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives46 (97) a. daß Hans endlich einmal was gut/richtig machen will that Hans finally once what well/correctly make wants ‘that Hans finally wants to do something well/right’ b. ?? daß Hans endlich einmal richtig was machen will c. * daß Hans endlich einmal gut was machen will This shows that the claim that any position to the left of manner adverbs renders arguments specific, cannot be maintained; non-interrogative (“existential”) was may never be specific showing that (i) scrambling out of VP is not only restricted to specificity and (ii) existential interpretation is not dependent on a projectional domain but on the domain of the existential operator.74 Another consequence of Haiden’s study is that manner adverbs do indeed sit low in the clause. Building on this finding, one would predict—under current assumptions regarding Multiple Wh-constructions—that double questions cannot be formed with either Wh-element following a manner adverb. (98) a. Wer hat was richtig gemacht? b. Was hat wer richtig gemacht? (99) a. * Wer hat richtig was gemacht? b. * Was hat richtig wer gemacht? Recall that the desired interpretation of Multiple Wh is a pair-list reading. While the questions in (99) may not bear this reading, they may serve as Echo questions. Postma (1995:106) makes a similar observation. He employs the Dutch equivalents (netjes ‘neatly’) to show the impact on Wh-elements with respect to a pair-list interpretation based on relative distribution. His data and conclusions carry over directly to German: (100) a. ?? Welcher Schüler hat richtig welche Aufgabe(n) lösen können? Which pupil has right which exercise(s) solve can ‘Which pupil could solve which exercise(s) correctly?’ b. Welcher Schüler hat welche Aufgabe(n) richtig lösen können? (101) a. * Welcher Schüler hat richtig was lösen können? b. Welcher Schüler hat was richtig lösen können? 74 Diesing (1992) proposes VP as the invariant domain for existential interpretation, while Postma (1995) suggests to link it to an existential operator. Kleanthes K. Grohmann 47 The Multiple Wh-constructions in the b-examples are both grammatical and felicitous: a pair-list reading can be construed. This is not the case in the a-examples; here, the only possible interpretation is that of an Echo question. The preceding discussion has shown that Wiltschko’s data show Superiority effects even in German; it has also shown, however, that Wiltschko’s analysis is not sufficient: she rules out the ungrammatical constructions on the grounds of the lower Wh’s failure of being D-linked; D- linking occurs through scrambling. As other authors have shown, however, even the supposedly non-scrambled Wh-elements in Wiltschko’s corpus have scrambled. Wiltschko’s approach can be reconciled with the present analysis in a straightforward manner. As briefly touched upon already, DRQ might be seen as a movement-variant of D-linking. Thus, elements may be D-linked by not moving at all (inherent D-linking as laid out by Pesetsky) or by moving to a topic position. English does not make use of the movement-variant,75 but German does so (solely). Hence, Wiltschko’s D-linked phrases are actually topicalized, following DRQ. To conclusively show that a high TopP in the left periphery is involved (and at the same time account for Wiltschko’s apparent counterexamples) shall be a topic for further research; for the time being, I conclude that Wiltschko’s analysis cannot accurately account for her data either.76 As a last word for the time being on this issue, note that (95b) may be rendered grammatical when a manner adverb follows the lower Wh: (102) a. Was hat denn schon oft wer gut gesehen? b. Was hat denn schon oft wer richtig gemacht? Take another set of examples from her study (Wiltschko 1997:117-8), concerning Pesetsky’s notion of “aggressively non-D-linked” Wh-elements: (103) a. Wer zum Teufel hat wen gesehen? who to-the devil has who seen ‘Who the devil saw who?’ b. * Wen zum Teufel hat wer gesehen? (104) a. Who the hell caught what b. * Who caught what the hell? Pesetsky (1987:124f., fn. 20) presents the contrast in English between an aggressively non-D- linked Wh-subject (104a) and an aggressively non-D-linked object (104b). Aggressive non-D- 75 Presumably only with partitive Wh-elements, though this is not crucial. 76 See also fnn. 61 and 72. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives48 linking also exists in German (103), as Wiltschko convincingly shows, and patterns with English in that the Wh-phrase in question may not be scrambled. We can capture the ungrammaticality of another of Wiltschko’s examples—namely, (103b) containing an aggressively non-D-linked Wh—differently: the Wh-phrase wen zum Teufel need not—and hence, under minimalist assumptions, cannot—be topicalized, as it is clearly non-D- linked. A non-D-linked Wh must sit in SpecCP in order to be interpreted and such movement evokes a completely different interpretation of the question (which is not relevant for this study).77 The following sentences can be accounted for accordingly (Wiltschko 1997:118): (105) a. * Wer zum Teufel hat wen zum Teufel gesehen? who to-the devil has who to-the devil seen ‘*Who the devil saw who the devil?’ b. * Wen zum Teufel hat wer zum Teufel gesehen? Aggressively non-D-linked Wh-elements must move to SpecCP overtly; German C0 can only attract one Wh prior to Spell-Out, hence ruling out the constructions in (105). 3.3.3 Wh-Adjuncts in Multiple Wh-Constructions: Some Speculations Take for one Huang’s (1982) null preposition hypothesis that aims at accounting for the syntactic behaviour of Wh-adverbs such as where and when versus why and how.78 Only the former may be preceded by an overt preposition such as from where, since when. We can schematize this as follows with the null preposition represented as p (from Kitahara 1997:101): (106) a. [PP p where/when] b. * [PP p why/how] Huang uses this fact to propose that where/when pattern like why/how in the syntax, while they behave like who/what at LF. (See also Tsai 1994 for a discussion of this in Chinese.) As Norbert Hornstein (personal communication) points out, it is not immediately clear how this might account for their asymmetry with respect to Superiority effects (but see Kitahara 1997:101ff. for some suggestions). 77 Recall that the German Multiple Wh-constructions investigated here are those that follow DRQ. The case of aggressively non-D-linked Wh-elements might blur the object of study: it may in fact not be a “real” multiple interrogative, in the sense of Uriagereka 1988, for example. In this sense, it is plausible that no topics are involved and consequently (103a) conforms to the Superiority Condition, while (103b) does not: according to the MLC, the first Wh that C0 attracts is wer zum Teufel and wer, respectively. 78 Jianxin Wu (personal communication) informs me that this distinction was also intended to capture the Chinese data, which has subsequently been modified. The exact behaviour of why/how are still pretty much a mystery. Kleanthes K. Grohmann 51 the introduction—van Riemsdijk (1997)—for the validity of this approach), Kuno and Takami’s left-dislocated elements are, in fact, hanging topics. Put crudely, hanging topics differ from left- dislocated elements in not agreeing in Case with the resumptive pronoun. I argue that topics and left-dislocated elements move to TopP, while hanging topics are base-generated in TopP. The following paradigm and the corresponding simplified representations show this: (112) a. Diesen Manni hat Maria ti geküßt this-ACC man has Mary kissed ‘This man, Mary kissed’ b. Diesen Manni, deni hat Maria ti geküßt this-ACC man the-ACC has Mary kissed ‘This man, Mary kissed him’ c. Dieser Mann, deni hat Maria ti geküßt this-NOM man the-ACC has Mary kissed ‘This man, Mary kissed him’ In (112a), the topic moves from its AgrOP-position to TopP where it checks [Top]; all other features have been checked at AgrOP. The left-dislocated element is also a topic (i.e. endowed with [Top]); however, it moves from its Agr-position to an intermediate position where it spells out as the demonstrative pronoun (presumably for V2-reasons) before it moves on to TopP. In the hanging topic-construction, on the other hand, the demonstrative pronoun is the base-generated object that moves for the usual reasons and the hanging topic is generated in TopP; crucially, this element does not agree in Case with the pronoun. In English, left dislocation does not exist, but the comparable constructions involve hanging topics.84 Furthermore, this TopP is available in both languages preceding CP. Only in German, though, is this position a possible landing site for movement; in English it may only serve as a position for base-generated elements (i.e. hanging topics). (113) a. This man, why did Mary kiss him? b. * This man, why did Mary kiss? 84 Note that Case-agreement is different from German: the fact that the hanging topic may also bear accusative Case in English is rather a reflex of morphology (with accusative being the default Case). Every dislocated element is marked accusative, just like all hanging topics in German are marked nominative (default Case in this language). See Grohmann 1997a for thorough discussion. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives52 (114) a. Diesen Mann, warum hat Maria den geküßt?85 b. Dieser Mann, warum hat Maria den geküßt? These facts can account for the differences that Kuno and Takami note without positing different positions for Wh-elements in the following way. Interjections sit in such a position that they intervene movement over them in English but not in German. Thus, if an element occurs in a position preceding the interjection in English it must be base-generated. The analogy is drawn to the distinction of base-generated hanging topics and movement-created left dislocation. These show also that a left-peripheral position blocks movement.86 Returning to the present topic, we now have evidence that left-peripheral positions vary in German and English with the possibility of movement. The syntax of interjection has, to my knowledge, not been the topic of much recent research. We can see from the limited data that it must be in a fixed position in the CP-domain. Interjections presumably sit right below the Wh- position (or even adjoined to C0) and somehow intervene with movement to that position in a similar vein that we observe in hanging topics. The conjecture is thus that hypothesis from (107) can be retained and these restriction account for the differences observed by Kuno and Takami.87 3.3.4 Multiple Wh and Quantifier Interaction: A Reaction to Beck (1996) Wu (1996) shows that scope asymmetries exist in Chinese when Wh-elements and quantifiers interact (see below for more on Wu’s study). In particular, a topicalized Wh-element yields only an individual reading when fronted over a quantifier. Interestingly, Beck (1996) also notes a similar peculiarity for German. She presents data that show interpretive differences in the interaction between quantifiers and Wh-phrases, and proposes an analysis based on the Minimal Quantified Structure Constraint.88 Consider (115), taken from Beck 1996:1 and 6: 85 With the Wh in CP, V2 has not to be saturated in any special form. I take the low pronoun to be a Spell-Out for Case-reasons. Note that topics are not allowed without the resumptive pronoun: (i) * Diesen Mann, warum hat Maria geküßt? I have to refer the interested reader to Grohmann 1997a for more discussion and evidence. 86 Another argument in favour of base-generation of why and how comes from the observation that in conjunction with interjections, they may only modify the clause they are in, as pointed out to me by Norbert Hornstein: (i) a. * Why, man, did you say Bill left b. Why, man, did you say Bill left 87 It is clear that the previous discussion is just another tentative approach. The details to this approach would be beyond the scope of this paper and I have to leave them open for further considerations, as so many issues. 88 The following gives a straightforward account for Beck’s data throughout. While lending strong support to the present analysis, it also does away with the various constraints that Beck proposes. A more careful investigation of her data may be needed, but the most relevant parts will be dealt with here rather smoothly. Kleanthes K. Grohmann 53 (115) a. # Wen hat niemand wo gesehen? whom has nobody where seen ‘Where did nobody see whom?’ b. # Wann hat niemand wen eingeladen? when has nobody whom invited ‘When did nobody invite whom?’ The two Wh-elements are argument and adjunct in both cases. In (115a), the Wh-adjunct wo occurs low, while wann appears high in (115b); both positions are fine under normal circumstances (cf. (28)). What makes these cases special is the presence of the quantifier niemand in between. It seems to render the sentences ungrammatical, or rather, deviant (marked by ‘#’): no coherent meaning can be assigned to the sentences in (115), as Beck remarks. Compare with (116): (116) a. Wen hat wo niemand gesehen? b. Wann hat wen niemand eingeladen? Here, the quantifier appears lower than both Wh-elements. Both sentences are grammatical and can be interpreted distributively. Under the proposal laid out here, the difference between (115) and (116) can be accounted for directly: as we have seen in section 3.3.1, a non-Wh-subject must move overtly to TopP in case it is preceded and followed by a Wh-element as it could otherwise not be interpreted as the highly thematic subject. This implies for (115), that the quantifier niemand would be topicalized. We know independently, however, that certain quantified expressions cannot be topicalized:89 (117) a. * Nobody, Karl likes b. Most people, Karl likes (118) a. * Niemanden mag Karl b. Die meisten Menschen mag Karl The lack of topic-properties of niemand in (115) rules out the derivation that it would receive under the present proposal. The subject-quantifier cannot move to TopP, and thus wo cannot move 89 I cannot indulge here in a deeper discussion of the exact reason for this restriction. Only some quantifiers can be topicalized (maybe along the weak/strong-dichotomy of quantifiers following Milsark (1977) and much subsequent work), and for the present purpose it suffices to note that some quantifiers such as nobody cannot topicalized. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives56 Again, a partitive Wh moves to CP; for theory-internal reason, it has to stop at TopP on the way. This is the sort of vacuous movement that is hard to show but assumed throughout. We can only give the circumstantial evidence from (123b) where the partitive element of the Wh remains at this intermediate position, on analogy with the previous examples. In (123a), the subject does not have to raise to TopP, as it is not overtly filled; niemand may receive interpretation in AgrSP just fine. In (123b) TopP above the subject is filled, forcing illegitimate movement of niemand to TopP. As predicted, (122) is grammatical with a strong quantifier as the subject, and so is (123).92 (124) a. Wen hat jede Studentin von den Musikern getroffen? b. Was hat jeder für Bücher gelesen? The strong quantifiers in (124) may move to TopP without any further ado. The approach as just outlined can thus account for all of Beck’s data without further assumptions. The role of quantifiers and topicalization may need some more attention, and so does Beck’s discussion on apparent intervention effects of jeder ‘every’. The upshot of this discussion, however, is that a topicalization-approach to Multiple Wh (driven by DRQ) seems feasible not only on its own grounds, but also with respect to other relevant phenomena observed in the literature. Furthermore, this section lends strong support to the proposal that under certain circumstances subjects must topicalize overtly. 4 Cross-Linguistic Variation, Learnability and More Support for Wh-Topics In the last section, I will present evidence from other languages; the most prominent one—and the one which this study leans on heavily—is Wu 1996 for Chinese. I will then discuss the role of Wh- islands to some extent. After looking at cross-linguistic evidence for Wh-topics, it will be more generally considered how the D-linking/DRQ-dichotomy from English and German can be realized across languages. I will close with some remarks on learnability and child language. 4.1 The Availability of Wh-Topics in UG This section will contain an elaborate discussion of the Chinese findings by Wu (1996) that allow us to construe the notion of a Wh-topic; I will also give additional evidence from Japanese. 92 For the sake of completeness, the version of (123b) without a quantified subject is also grammatical: (i) Was hat Luise für Bücher gelesen? Kleanthes K. Grohmann 57 4.1.1 Chinese Wh-Topics In Chinese questions there is no overt movement of the Wh-element(s). Presumably, the Wh- elements move to SpecCP at LF in both single and double questions.93 This is considered the standard approach to Chinese interrogatives ever since Huang 1982. I will now present some interesting findings by Wu (1996) who concludes that in some cases Wh-elements may move. These cases are strictly discourse-restricted in that they are instances of topicalization. After citing his empirical evidence, I will recapitulate Wu’s speculations on a possible interaction between Wh-movement and topicalization—two grammatical processes that seem at first glance to contradict each other but may well be conceived as being related. The general idea is very much in line with the present proposal. First, consider standard examples of Wh-formation in Chinese where no movement takes place at all, as presented by Wu (1996:178): (125) a. Shei mai-le shenme? who buy-ASP what ‘Who bought what?’ b. Ni quan shei bu yao mai shenme dongxi? you persuade who not buy what thing ‘You persuade who not to buy what?’ In (125a) we have an instance of a subject- and Wh-object, while (125b) parallels the construction where Pure Superiority could possibly arise (by movement of the lower Wh-element). In neither construction can we observe movement taking place. But compare (125) to (126) below (from Wu 1996:177): (126) a. Shenmei shei mai le ti? what who buy ASP ‘*What did who buy?’ b. [Shenme dongxi]i ni quan shei bu yao mai ti? what thing you persuade who not buy ‘*What did you persuade who not to buy?’ 93 Aoun and Li (1993) and Watanabe (1992) suggest movement of parts of the Wh-phrase (or an empty Wh- operator) in the overt syntax, parameterizing Wh-movement in a different way. I will adopt the traditional view, although nothing much hinges on this assumption. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives58 We can see that the constructions in (125) are the equivalents of (126) with the exception that the lower Wh-element is fronted in both.94 This gives rise to Superiority violations in English but not in Chinese. In an interesting discussion, Wu reaches the conclusion that the fronted Wh-elements in (126)—and comparable cases—must be topicalized. For economical reasons within the minimalist framework, the movement of the Wh cannot be for Wh-feature-checking purposes: the standard Wh-checking at LF is more economical than overt checking. One of the most convincing pieces of evidence he presents is the following restriction on Multiple Wh (adapted from Wu 1996:181): (127) Situation: Zhangsan went to the supermarket to do some shopping. a. Zhangsan mai-le shenme? Zhangsan buy ASP what ‘What did Zhangsan buy?’ b. # Shenme Zhangsan mai-le? In the given context, only (127a) is an appropriate question. The situation does not imply that Zhangsan actually bought anything. The question in (127b), on the other hand, can only be posed if there is a referent to the Wh-element, i.e. if Zhangsan has really bought something such as a number of items of a shopping list. (Compare this with the felicity of questions for German.) (128) Situation: Zhangsan went to the supermarket and bought items from a mutually known list. Shenme Zhangsan mai-le? At the face of it, German and Chinese are different in that DRQed Wh-phrases always topicalize overtly in German, while in Chinese it is only one Wh that moves (also, see fn. 94 on 94 Wu notes that they are not equivalent in their meaning: while (126a) may be interpreted in a distributive or an individual reading, (126b) may only be assigned the latter. But this is a difference not directly relevant here: this section serves to show that the conception of certain Wh-elements as topics is not as alien as it may sound at first sight; in how much languages vary with respect to the semantics of Wh-topics needs to be established in further research. Wu (op. cit.:182) also points out that (ib) may only have a distributive meaning (unlike (ia)), a statement that he qualifies somehow in his fn. 2 on the same page according to which some speakers may get a distributive reading. (i) a. What did everyone buy? b. Which book did everyone buy? In order to make clear predictions about the possibilities of distributivity and topicalization of Wh-elements (and quantifiers), more work needs to be done. Kleanthes K. Grohmann 61 these circumstances are very restricted. But it shows that Wh and topicalization are not mutually exclusive by nature.97 Miyagawa (1987:186) notes that while wa is ungrammatical in simple interrogatives, it may occur in those questions that severely restrict the set of referents. Here, wa is interpreted thematically and thus refers to a definite individual: (132) a. * Dare-wa kita no? who-TOP came Q ‘Who came?’ b. Dare-wa kite, dare-wa konakatta no? who-TOP come-GER who-TOP didn’t-come Q ‘Who came, and who didn’t?’ Miyagawa shows more instances of the connection of the topic marker and Wh-elements. Note that this is also reminiscent of what Wu (1996) has shown for Chinese and what section 3 was concerned with for German. The upshot of this is, again, that a connection between Wh-formation and topicalization need not be ruled out a priori; furthermore, it is found cross-linguistically. It should be noted that the nature of wa on Wh-elements differs from the remainder of this paper. While the contrast between German and English showed different constraints by the discourse (further supported by the Chinese data), something else is going on in Japanese; wa is a contrastive topic-marker and thus induces a different interpretation than a D-linked topic. Further research will have to take discourse restrictions into account as I have shown for German. 4.1.3 Wh-Topics in General The limitations set for this project have not allowed a more thorough survey of Wh-topics across languages; this task remains for future research. Interestingly, evidence from child language acquisition might support the present approach (deVilliers et al. 1996, Thornton 1995) which will be laid out in section 4.4; incidentally, non-verbal languages might do so, too.98 The outcome of this will also have to be addressed in another project. In sum, the cross-linguistic observations in general, and the thorough application to German in particular, go hand in hand towards a theory of Wh-topics that might turn out to be found more frequently than previously assumed. 97 Judgements regarding the acceptability of wa vary; it should also be noted that wa in these cases can only mean contrastive topic. This might be additional evidence that a focus-part is involved in these constructions, maybe even responsible for Wh-interpretation across languages (as briefly mentioned above). 98 In a recent study, Neidle et al. (1997) are concerned with Wh-formation in American Sign Language (ASL). Their findings seem to suggest that ASL employs a sentence-initial Wh-topic in addition to a sentence-final “real” Wh-element. Robert Lee (personal communication) brought the interesting observations from ASL to my attention. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives62 4.2 Wh-Islands and Wh-Topics Comorovski (1996) connects Cinque’s (1990) characterization concerning D-linking and the possibility for Wh-extraction out of a Wh-island (see her chapter 5 for detailed discussion) into her general account of D-linking. 4.2.1 Some Basic Data Comorovski’s (1996:162) data from Swedish are taken from Maling 1978 and Engdahl 1980 which are shown below, respectively: (133) a. * Vad visste ingen vem som skrev? what knows no-one who wrote ‘What does no one know who wrote? b. Sven undrar vilken bok alla studenter minns vilken författare som skrev Sven wonders which book all students remember which author wrote ‘Sven wonder which book all students remember which author wrote’ While a bare Wh may not be extracted out of a Wh-island (133a), a D-linked one may (133b). Rizzi (1978) noted a similar property of Italian, where Multiple Wh-constructions are generally ruled out. Again taken from Comorovski 1996:163, the relevant data are shown here: (134) a. ?? A chi non ti ricordi quanti soldi hai dato? to whom not you remember how-much money have-you given ‘To whom don’t you remember how much money you gave?’ b. A quale dei tuoi figli non ti ricordi quanti soldi hai dato? to which of your sons not you remember how-much money have-you given ‘To which one of your sons don’t you remember how much money you gave?’ Original speculations regarding the “heaviness” of Wh-island-extractable Wh-phrases seem to point to the phenomenon of D-linking, as fairly well established by Cinque and Comorovski.99 With respect to the present approach—where multiple Wh-phrases are analysed as topics—this might have far-reaching consequences. With this in mind, does Chinese—where Wu (1996) has conclusively shown that Wh-topics exist—allow Wh-extraction out of Wh-islands? 99 Cinque refers to the difference of D-linked Wh vs. non-D-linked Wh as a difference in “referentiality.” In the terms of partitives, this referential property follows immediately (see also 4.4.3 below where a further explicit reference to referentiality made by Thornton (1995) is reviewed). Kleanthes K. Grohmann 63 (134) a. Shenme dongxii Zhangsan xiang zhidao Lisi mai mei mai ti? what thing Zhangsan want know Lisi buy-not-buy ‘What did Zhangsan want to know whether Lisi bought?’ b. * Shenmei Zhangsan xiangxin Lisi tou le ti de yaochuan? what Zhangsan believe Lisi steal DE rumour ‘What is x such that Zhangsan believes the rumour that Lisi stole x?’ As Wu (1996:74) shows a Wh-topic may be extracted out of a Wh-island (134a); (134b) illustrates with a CNP-island that strong islands are still boundaries for extraction.100 One can imagine how the by now familiar connection between D-linked Wh-phrases and Wh-topics would account for this within a general explanation why only these may be extracted and not bare, non-D-linked Wh-phrases; namely, the D-linked Wh uses embedded TopP as an “escape hatch” to move out of the Wh-islands. (I will lay out the details below.) 4.2.2 Slavic and the Position of TopP Bulgarian also has a TopP in the left periphery, independently from Multiple Wh. The standard assumption is that projection dominates CP (or FocP in Rizzi’s (1997) terms). This has been argued for by Rudin (1985) and is mentioned in Richards 1997:111f. who cites the following: (135) a. [TopP Ivan vcera [CP kakvo kupi]]? Ivan yesterday what bought ‘What did Ivan buy yesterday?’ b. [TopP Vcera Ivan [CP kakvo kupi]]? With respect to Wh-elements moving to this projection, Richards (1997:112) notes that only a D-linked Wh may move there: (136) a. * [TopP Koj vcera [CP kakvo kupi]]? who yesterday what bought ‘Who bought what yesterday?’ b. [TopP Koja zena vcera [CP koja kniga kupi]]? which woman yesterday which book bought ‘Which woman bought which book yesterday?’ 100 As Comorovski (1996) sets out to account for, extraction of a D-linked Wh is possible not only out of Wh- islands but (often) out of all weak islands, as Wu also shows for Chinese (negative islands, factive islands). On the other hand, strong islands can never be moved out by Wh-elements. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives66 At first sight, this might pose a problem: if the language exhibits Wh-topics elsewhere (namely in Multiple Wh, conforming to DRQ), this is rather unexpected, especially with the cross- linguistic data in mind we have seen above. This is even more puzzling as non-interrogative topics may be extracted out of Wh-islands. (142) a. Bücher weiß ich, wer gelesen hat books know I who read has ‘Books, I know who read’ b. Dieses Buch weiß ich, wer gelesen hat Topicalized plural nouns (142a) or DPs with a demonstrative (142b) can be extracted out of a Wh- island (see Müller and Sternefeld 1993, Culicover 1996 for more). Obviously, they must get somehow to TopP of the matrix clause. So what could prohibit DRQed Wh-elements to move? What Italian, Romanian and Bulgarian have in common is that all languages allow TopP to precede CP.104 In particular, they all make use of this position in non-interrogative contexts. English, on the other hand, has this position available but the only evidence we have seen comes from base-generated hanging topics. In German, this position can also be filled, as we have witnessed from the examples of left dislocation. I suggest to look at these constraints more closely. One way of interpreting the differences between English and German pre-CP topic-positions is the following: pre-CP TopP cannot be filled through overt movement in English; on the other hand, in German movement can only target it if is subsequently filled. If this characterization is correct, it should be possible in principle to evoke the projection in English if the element moves further; likewise, it cannot be activated in German if the element does not stay. Can we motivate this somehow? Recall from section 3.3.3 that the resumptive pronoun in left dislocation-constructions is assumed to be the Spell-Out of the trace of the left-dislocated element. In Grohmann 1997a, 1998 I suggest that the Spell-Out occurs in order to satisfy constraints on Case-expression (low pronoun) or on V2 (high pronoun). Especially the latter condition might explain why left dislocation can only be found in V2-contexts, namely in matrix clauses and embedded under so-called “bridge” verbs that optionally may select for an embedded V2-clause. Now consider (143), which seems to be a very strange exception at first glance: 104 I have not looked at the possibility of the ordering [TopP Top 0 [ CP C 0…]] in Swedish or in Chinese rather than only allowing [CP C 0 [TopP Top 0 …]]; both might be feasible, however, given the following discussion. Kleanthes K. Grohmann 67 (143) a. Welche Bücheri weißt du, diei welche Studenten gekauft haben? which books know you the which students bought has ‘Which books do you know that which students bought them?’ b. * Welche Bücher weißt du, welche Studenten gekauft haben? c. * Welche Bücheri weißt du, diei haben welche Studenten gekauft? d. * Diese Bücheri weiß ich, diei diese Studenten gekauft haben The grammaticality of (143a) comes as a huge surprise, given what we know about left dislocation. Left dislocation concerns a topic and its co-referent demonstrative pronoun; there is no (obvious) topic here. However, there is a Wh dislocated which is co-referent with a demonstrative pronoun and this particular Wh has been analysed throughout as a Wh-topic. This contrasts with (143b), where there is no demonstrative pronoun related to the Wh-topic. Also, as just noted, V2-context is crucial for left dislocation; it is not present here. This contrasts with (143c) where the same construction is set up within a V2-context. Lastly, (144d) shows that a non-interrogative topic may not be left-dislocated, violating the standardly assumed constraints. Within the present proposal, (143a) receives a straightforward analysis.105 105 I dispense with illustrating those parts of the derivation that are not relevant here. I will thus only show the derivational steps that are crucial to derive the left-dislocated Wh and the Wh-topic. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives68 (144) CP Welche Bücheri C’ [Wh] C0 IP [Wh] … TopP diei Top’ [Top] Top0 CP [Top] welche Studentenj C’ [Wh] C0 TopP [Wh] tj Top’ [Top] Top0 IP [Top] tj… ti… The overall constructions is a Multiple Wh; both Wh-elements thus bear [Top]. Both also bear [Wh] which in the normal case one Wh checks overtly, the other covertly. The derivation runs normally until the lower IP is formed. Top0 enters the derivation and attracts the closest element, welche Studenten.106 The Wh-subject thus checks [Top]. Next, C0 attracts the closest element bearing a Wh-feature; again, the Wh-subject moves and checks [Wh]. It is only here that a further Top0 comes in the picture. As welche Studenten has already checked [Top], welche Bücher is the closest element with an unchecked Top-feature; thus, the Wh-object is attracted, moves and checks its feature [Top]. After the matrix IP is formed, another C0 arrives. There is only one element with a Wh-feature, namely the Wh-object. While in simple Multiple Wh-constructions there is only one C0 and hence only one [Wh] can be checked overtly, in the case of two clauses a second C0 is available, thus allowing the last move. Interestingly, however, the TopP dominating CP may not be 106 Once the details concerning Wh-subjects are completely clarified, we can give a rule-driven account why in this case not the Wh-object is attracted; the apparent optionality we have already seen also holds for this example: (i) Welche Studenten weißt du, die welche Bücher gekauft haben? Kleanthes K. Grohmann 71 4.2.4 Results After an extensive discussion of the data and possibilities of Wh-extraction out of Wh-islands across languages, we can confirm the observation that only D-linked Wh-elements may be extracted to some extent. Under the view proposed here, these elements need to check a Top-feature on their way to the initial position of the matrix clause. It is thus more accurate to say that only Wh-topics can be extracted out of Wh-islands.111 As the data have shown—and especially the discussion of German—it cannot be the Top- feature that allows extraction; rather the Top-feature allows movement beyond the lower Wh, yet still inside the embedded clause. It is the Wh-feature on matrix C0 that forces further movement. We can now modify the generalizations from (138) as follows: (152) a. [CP Wh i [… [IP … [TopP ti … [CP Wh … [IP … ti…]]]]] b. * [CP Wh i [… [IP … [CP Wh … [IP … ti…]]]]] c. * [TopP Wh i [… [IP … [TopP ti … [CP Wh … [IP … ti…]]]]] d. * [CP Wh i [… [IP … [CP Wh … [TopP ti … [IP … ti…]]]]] e. * [TopP Wh i [… [IP … [CP Wh … [TopP ti … [IP … ti…]]]]] The only acceptable way to escape a Wh-island is by topicalizing above the lower Wh and further Wh-move to the matrix CP (152a). All other possibilities are ruled out: long movement to matrix CP over a filled embedded CP (152b), movement to matrix TopP (153b) and extraction out of embedded TopP situated below CP to either matrix CP (152d) or matrix TopP (152e). 4.3 Cross-Linguistic Differences Among Multiple Wh I will now turn to cross-linguistic considerations of D-linking in the bigger framework outlined here.112 I propose a tripartite classification of languages and the felicity constraints on Multiple Wh. We already have three types of languages with respect to Wh-movement to CP (overt, covert and mixed). The present classification is similar in spirit: do languages overtly topicalize multiple Wh-phrases, covertly or mixed? This classification is very tentative; moreover it just looks at some obvious syntactic possibilities (fronting of Wh-subject vs. Wh-object) based on pragmatic (= discourse) restrictions. It does not make any deeper predictions as to what might be going on in other languages. It is, however, useful to get a theory-independent grip on these restrictions. 111 Admittedly, the fact that (certain) D-linked Wh-phrases are endowed with [Top] in all languages has not been established here; the prediction is clear, though, based on the present proposal. It is also most certainly and smoothly in accordance with all data presented in this study. 112 This project allows only for a very limited and crude look into discourse restrictions across languages. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives72 4.3.1 Chinese and German German and Chinese Multiple Wh differ in some respects. For one, all Wh-elements obligatorily move overtly in German but not in Chinese. Secondly, a distributive reading can be applied to any felicitous Multiple Wh in German, while it can only be assigned to those constructions in Chinese where both Wh-elements stay in situ. But the contextual influence on Wh-topics is the same in both languages. This sub-section is concerned with finding similarities between these two languages that cannot be found in the other languages under investigation (see below). In this respect, it is important to bear in mind that both languages (i) exhibit Wh-topics and, more importantly, (ii) underlie the same contextual restrictions on the felicity of multiple interrogatives. We have seen the contrast between English and German in some detail in section 3.1—let us now turn to the similarities between German and Chinese given the same circumstances.113 Consider thus the following situations (verified by Jianxin Wu, personal communication):114 (153) Situation I: A salesman comes home and speaks to his wife. “I sold a few things today.” a. # She mai-le shenme? b. # Shenme mai-le she? c. # Wer hat was gekauft? d. # Was hat wer gekauft? (154) Situation II: A salesman comes home and speaks to his wife. “I sold a car, a fridge and a TV today.” a. # She mai-le shenme? b. # Shenme mai-le she? c. # Wer hat was gekauft? d. # Was hat wer gekauft? In neither language is it possible to ask a Multiple Wh if reference has been made to only one Wh- element. That this is the right conjecture can be witnessed from the following: 113 For the sake of exposition, I only present data for the felicity of Multiple Wh involving a Wh-subject and a Wh- object, here and in the following. 114 Here and in the remainder of the section, the questions in all languages are translated as Who bought what? in the a- and c-examples, and What did who buy in the b- and d-examples, similar in essence to (59). Kleanthes K. Grohmann 73 (155) Situation III: A salesman comes home and speaks to his wife. “I sold a car, a fridge and a TV to an old man, a young woman and a teenager.” a. She mai-le shenme? b. Shenme mai-le she? c. Wer hat was gekauft? d. Was hat wer gekauft? If both Wh-elements are specified to some degree (basically following DRQ), German and Chinese allow a felicitous Multiple Wh. 4.3.2 Spanish and Hebrew The second type of discourse restrictions on Multiple Wh can be observed in Spanish and Hebrew. Although we have only seen limited data from these languages, the important aspect to note is that both languages allow the Wh-object to be fronted over the Wh-subject. The syntax of Multiple Wh in Hebrew and Spanish shall not be the topic of this paper; rather, I am interested in showing that the contextual restrictions applying to Multiple Wh are the same in both languages, while they differ from German and Chinese on the one hand, and from English on the other. These two languages differ from German and Chinese in that not the referent sets of all Wh- phrases must be established in order to ask a Multiple Wh, but only of one.115 Employing the same situations as above, here are the results. (The following data are confirmed by Danny Fox and Idan Landau for Hebrew and Juan Carlos Castillo for Spanish.): (156) Situation I: A salesman comes home and speaks to his wife. “I sold a few things today.” a. Mi kana ma? b. # Ma kana mi? c. Quién compró qué? d. # Qué compró quién? (157) Situation II: A salesman comes home and speaks to his wife. “I sold a car, a fridge and a TV today.” a. Mi kana ma? b. # Ma kana mi? c. Quién compró qué? d. # Qué compró quién? 115 In this respect, they pattern much like English. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives76 Under a cue-based approach to learnability (Lightfoot, forthcoming), unambiguously simple (= Degree-0, Lightfoot 1991) positive data enable the learner to figure out the correct structure (and hence correct derivation) for the construction in question. With respect to the Wh-adjunct asymmetry in English and the question of movement vs. generation, the examples with interjections plausibly serve as cues: (161) a. Why, child, did you eat my dinner? b. How, child, can you eat so quickly? (162) a. * When, child, did you eat my dinner? b. * Where, child, did you hide my dinner? While (161) shows that some Wh-elements are fine in the pre-interjectional position, the ungrammaticality of others (162) shows that this contrast is due to the prohibition of moving to this position. Of course, (162) is not a possible input for the learner. But hanging topics are, especially those constructions where the hanging topic precedes a Wh; hanging topics furthermore are base- generated—no movement ever targets a pre-CP-position.118 The child has thus evidence that this left-peripheral position can only be filled via base-generation. Let us thus assume that the learner can differentiate between base-generation and movement of Wh-adjuncts. In German, all four constructions are fine. Furthermore, the contrast between hanging topics and left dislocated elements (that does not appear in English) allows the learner to differentiate movement and base-generation correctly. Crucial for the German learner is that all these constructions can be possible evidence, as they are grammatical.119 4.4.2 Cues for Wh-Topics But a more interesting issue—which also has farther-reaching consequences—is the difference among languages with respect to contextual restrictions on multiple interrogatives. Why languages should differ in such restrictions should be a topic for a different study; there seems no obvious explanation based on the (limited) observations this study could provide. A related question, however, can be given a preliminary explanation: what causes the syntactic variation given the contextual restrictions in Multiple Wh? Let us start with the most restricted case: DRQ which is at work in German, for example. The contextual restriction is that the reference of both Wh-elements must have been introduced prior to asking a Multiple Wh. Syntactically, this is expressed by movement to a topic-position in this 118 I refer to the discussion in sections 3.3.3 and 4.2.3 above. 119 Variation among German dialects with respect to allowing warum and wie in a position other (= lower) than CP (as found by Haider (1996)) may be deduced from these facts, too; in order to positively support this hypothesis, though, more relevant data needs to be collected from speakers of that dialect. Kleanthes K. Grohmann 77 language where this “referentiality” can be formally checked. Aside from other reasons, evidence for this movement can be found in the structure of the Wh itself: while the topic-feature may be encoded on a functional head attracting this movement, it must have a reflex coded onto the Wh. This reflex, I argue, roots inside the more delicate structure construed for the Wh: these Wh- elements behave like partitive interrogatives and hence consist underlyingly of the same structure that partitives do. The topic-feature can then be seen as a requirement of the partitive. In German all Wh-elements in Multiple Wh contain the partitive—whether it is overtly expressed or not. The simplest case is that of an overt partitive Wh: here the learner gets a cue from the peculiar behaviour, namely that partitive interrogatives may be split and intervening material is constrained. Consider (122-124) again, shown here in a variation: (165) a. Welches von diesen Bildern gefällt dir am besten? b. Welches von diesen Bildern gefällt jedem Kind am besten? c. Welches von diesen Bildern gefällt keinem Kind am besten? (166) a. Welches gefällt dir von diesen Bildern am besten? b. Welches gefällt jedem Kind von diesen Bildern am besten? c. # Welches gefällt keinem Kind von diesen Bildern am besten? This is the paradigm that every adult speaker of German knows. The learner, however, can deduce from possible positive evidence such as (166a) that the partitive Wh may be split into the interrogative part and the partitive part. It thus receives a cue that some Wh-phrases check a Top- feature as well as a Wh-feature, and that such phrases may be split up.120 (166b) gives the learner a cue that some quantifiers may intervene between the Wh-part and the Top-part of the partitive Wh. Knowledge of the behaviour of quantifiers will make available the rest of the paradigm; for example, as soon as the child learns that weak quantifiers may not be topicalized, (166c) can be deduced. It then derives the constructions in (166) correctly by checking the topic-feature prior to checking the Wh-feature.121 As the behaviour of partitive Wh-elements in German is practically the same for D-linked and bare Wh-phrases in multiple interrogatives, the learner will assign the syntax of partitives to all Wh- 120 I assume that the topic-feature is universal to partitives, and if a learner comes across a partitive and correctly analyses it as such, it knows that [Top] is the relevant feature. Without a thorough discussion of language acquisition, I refer the reader to the works of Crain and Thornton or Lightfoot, for example, as well as the references cited there. The general assumption here is something like the Strong Continuity Hypothesis, i.e. that we can assume maximal knowledge for minimal input. 121 The intricacies involved here predict that at the beginning stages, mistakes may be made. These mistakes are still within the boundaries of UG, however, so that a learner would never use strategies that are not allowed; it simply mis-analyses them to some degree, often related to economy conditions. Some of these actual mis-analyses will be shown below from English. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives78 phrases; in other words, it will assume a more intricate structure for all Wh-phrases based on the pattern observed from partitives.122 On the other hand, the learner of English will be exposed to different syntactic behaviour between partitives and other Wh-phrases. One unambiguously piece of positive evidence is the contrast observed in (139), repeated here as (167):123 (167) a. * How much moneyi do you wonder who to give ti to? b. ? How much of the moneyi do you wonder who to give ti to? Suppose the learner detects that partitive Wh-phrases in English differ in their syntactic behaviour from other Wh-phrases; suppose also that the learner correctly identifies this difference with the obligatory movement through TopP. The result is that the learner has now a cue that there are Wh- phrases with more intricate structures that require topicalization. Analogously to possible variation among German dialects, the English learner might then get cues that allow it to extend this structure to other Wh-phrases. Just as the German learners may vary—depending on evidence—whether they should assign the partitive structures to D-linked Wh-phrases and then (in the ideal, standard case) also to bare Wh-phrases (see fn. 123), the English learner may get evidence to assign the partitive structure to D-linked Wh-phrases.124 4.4.3 Wh-Topics and Mis-analyses in Child Language While learners accrue linguistic knowledge in a rapid way, “mistakes” may be made on the way when the learner tests out freshly accumulated data. It is thus predicted that with the intricate behaviour of questions, learners produce mis-analyses to some extent. If this approach and the general picture of language acquisition (such as presented in Lightfoot or Crain and Thornton) are on the right track, irregularities can be predicted; furthermore, it can be—to some extent—predicted what kinds of mistakes will occur and what kinds will not. 122 This view can then be extended to variation among dialects. I predict that if there are any restrictions in multiple interrogative contexts, the most liberal behaviour will be that of partitives, followed by “heavy” (= D-linked) Wh- phrases and bare Wh-phrases being most restricted. If this is so, the learner will implement these differences by not assigning the deeper (= partitive) structure to bare Wh-phrases first, and then possibly, extend it to “heavy” ones. 123 Admittedly, I am not an expert on child language. While (167b) may be an unlikely candidate of positive evidence that a child may be exposed to, there are numerous other examples that show different syntactic behaviour of partitives, in “simpler” constructions. 124 I am not aware of English dialects that show the same syntactic behaviour of partitive Wh-phrases not only with D-linked Wh-phrases, but also with bare ones. If the latter variety does exist, the learner of that particular dialect will thus assign the partitive structures to these elements too. Kleanthes K. Grohmann 81 5 Conclusion128 The goal of this paper was to investigate the nature of multiple interrogatives; in particular I went at length to propose a syntactic account within the minimalist framework (Chomsky 1995) for constructions that are restricted by the discourse—in severity as well as in variation. The initial thesis that Wh-movement may be influenced by movement processes prior to Wh-movement turns out to be borne out. Unfortunately, the exposition of this thesis took a long time and needed much space. Fortunately, many different issues could be investigated. With evidence from a number of languages, I set out to consider movement processes prior to Wh-movement in German. The obvious candidate is scrambling, or A-movement in general, following the general path that Boskovic (1993) walked for Spanish (and Hebrew), Takahashi (1993) for Japanese and Hornstein (1995) for English. I myself also went there for German (Grohmann 1997b). There is, however, an end to this path. Movement for the purpose of already- known features prior to Wh-movement is not enough. It became clear that discourse plays a crucial role. Pesetsky (1987) accounted for this with a theory of D(iscourse)-linking. I took the rather unclear details of D-linking as the starting point to re-consider the facts and suggest an elaboration. Finding that German Multiple Wh underlie much stricter discourse restrictions than English, I propose that in German Multiple Wh-constructions both Wh-elements are obligatorily D-linked in a special sense: they must conform to D(iscourse) R(estricted) Q(uantification). In general terms, this means that the referent sets of both Wh- elements must have been introduced in the discourse in order to formulate a felicitous double question. Technically, I propose that DRQ is instantiated by movement to the left periphery (Rizzi 1997) of both Wh-elements. Thus, DRQ fits out a Wh with a feature [Top], to be checked at TopP. The structure assumed here for partitives helps to visualize this abstract and at first glance ad hoc feature: internally to the DP-structure of a partitive, [Top] is present, namely on the partited DP (the DP following the partitioner). German Wh-elements have thus to be assigned the more intricate structure of partitive interrogatives in Multiple Wh, even if they are only bare Wh-phrases. Once both Wh-elements have checked [Top], the higher Wh moves to CP to check [Wh], in accordance with the MLC. This approach may carry over to a lesser degree to English D-linked Wh-phrases. Especially in instances where partitive Wh-phrases differ from simpler ones, this approach makes the right predictions. One such instance is extraction out of Wh-islands (inter alia Kiss 1993, Comorovski 1996). The result of a deeper discussion of Wh-extraction out of Wh-islands is that D-linked Wh- phrases may be extracted, when they can be assigned the (abstract) partitive structure; or in other 128 Given the many open questions raised and pointed out throughout this study, this might actually be a misnomer; call it “preliminary summary.” Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives82 words: Wh-topics may move out of Wh-islands. The landing site is matrix CP and the jumping site is a high TopP, situated above the embedded CP. The initially counterintuitive notion of a Wh-topic deserves special consideration. Wu (1996) conclusively shows that there are instances in Chinese Wh-formation where a Wh is fronted; this cannot be the result of checking [Wh]. Wu shows that it is in fact topicalization. In many instances, the notion of a Wh-topic has been explored, employed and elaborated. It designates those Wh- elements that have an underlyingly deeper structure, equipped with a topic-feature (partitive). In the course of the proposal, some apparent counterexamples and other difficulties have been pointed out, addressed and sometimes even solved. One particularly nasty issue is that of (Wh- )subjects. On the grounds of thematic properties intrinsic to subjects, I show that in some problematic cases, the subject topicalizes also. Moreover, this topicalization is a different process than normal topicalization, as it is more an information-structurally driven property than some morphological feature that motivates this move. While capable of handling the data, this approach throws up a number of other problems. These, as many others, need to be subjected to scrutiny in future work. With respect to variation, suggestions were advanced that parameterize English D-linking and German DRQ to some degree, also allowing for further variation. Important considerations come from child language (which, even in English, seems to wonderfully go hand in hand with the present proposal; see Thornton 1995) and learnability issues. Unfortunately, the number of issues followed up here required a density that might blur some of the more important points. Fortunately, I am confident that these points are relevant indeed. A unique example of Wh-extraction out of Wh-islands in German has been observed and assigned a straightforward explanation within the given framework. Many more nice correlations have been found, too—one of which is that the input of scrambling in obviating German Superiority effects results most naturally as a corollary of my proposal, rather than the driving force; that the latter is not warranted for is apparent: scrambling is not a unified phenomenon with respect to one particular property of the moved elements. It can be, however, topicalization, namely in those instances in German where objects are scrambled above the subject. This extensive study proposes to take discourse restrictions on Multiple Wh-constructions across languages seriously into syntactic considerations.129 The suggestion made here is that such restrictions may be due to the nature of structurally more complex Wh-phrases; this structure, in turn, forces movement prior to Wh-movement which is understood as topicalization to license the contextual factors. While this partially extensive study captures a wide range of data and (at first glance unrelated) phenomena, it has always been pointed out that more work needs to be done. 129 (A personal note: this sentence sums up the past 82 pages most accurately.) Kleanthes K. Grohmann 83 References Abraham, Werner, Samuel D. Epstein, Höskuldur Thráinsson and C. Jan-Wouter Zwart, eds. 1996. Minimal Ideas: Syntactic Studies in the Minimalist Framework. Amsterdam: John Benjamins. Abraham, Werner and Kleanthes K. Grohmann, eds. 1997. Groninger Arbeiten zur germanistischen Linguistik 40. Rijksuniversiteit Groningen. Alexiadou, Artemis. 1997. Adverb Placement: A Case Study in Antisymmetric Syntax. Amsterdam: John Benjamins. Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts, eds. 1997. Materials on Left Dislocation. Amsterdam: John Benjamins. Aoun, Joseph, Norbert Hornstein and Dominique Sportiche. 1981. Some Aspects of Wide Scope Quantification. Journal of Linguistic Research 1: 69-95. Aoun, Joseph and Y.-H. Audrey Li. 1993. Syntax of Scope. Cambridge, MA: MIT Press. d’Avis, Franz-Josef. 1996. On Wh-Islands in German. In U. Lutz and J. Pafel, eds. On Extraction and Extraposition in German. Amsterdam: John Benjamins. Baker, C. Lee. 1970. Notes on the Description of English Questions: The Role of an Abstract Question Morpheme. Foundations of Language 6: 197-217. Beck, Sigrid. 1996. Quantified Structures as Barriers for LF-Movement. Natural Language Semantics 4: 1-56. Bolinger, Dwight. 1978. Asking More Than One Thing at a Time. In H. Hiz, ed. Questions. Dordrecht: Reidel, 107-150. Boskovic, Zeljko. 1993. On Certain Violations of the Superiority Condition, AgrO, and Economy of Derivation. Ms., University of Connecticut, Storrs. [Appeared in a revised version, 1997, Journal of Linguistics 33: 227-254.] Boskovic, Zeljko. 1998. Wh-Phrases and Wh-Movement in Slavic. Ms., University of Connecticut, Storrs. Castillo, Juan Carlos. In Press. The Syntax of Container-Content Relations. In E. Murguia, A. Pires and L. Quintana, eds. University of Maryland Working Papers in Linguistics 6. Chierchia, Gennaro. 1991. Functional Wh and Weak Crossover. Proceedings of WCCFL 10. Stanford: CSLI, 75-90. Chomsky, Noam. 1973. Conditions on Transformations. In S.R. Anderson and P. Kiparsky, eds. A Festschrift for Morris Halle. New York: Holt, Reinhart and Winston, 232-286. Chomsky, Noam. 1980. On Binding. Linguistic Inquiry 11: 1-46. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 1998. Minimalist Inquiries: The Framework. Ms. (Draft), Massachusetts Institute of Technology, Cambridge. [To appear in R. Martin et al., eds.] Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives86 Kuno, Susumo. 1982. The Focus of the Question and the Focus of the Answer. Papers from the Parasession on Nondeclarative Sentnces. Chicago Linguistic Society, 134-157. Kuno, Susumu and Ken-Ichi Takami. 1993. Grammar and Discourse Principles: Functional Syntax and GB Theory. Chicago: University of Chicago Press. Kuroda, Sige-Yuki. 1988. Whether We Agree or Not: A Comparative Syntax of English and Japanese. Lingvisticae Investigationes 12: 1-47. Laenzlinger, Christopher. 1996. Adverb Syntax and Phrase Structure. In A.-M. Di Sciullo, ed. Configurations: Essays on Structure and Interpretation. Somerville, MA: Cascadilla Press, 99- 127. Lewis, David. 1975. Adverbs of Quantification. In E. Keenan, ed. Formal Semantics of Natural Language. Cambridge: Cambridge University Press, ???. Lightfoot, David W. 1991. How to Set Parameters. Cambridge, MA: MIT Press. Lightfoot, David W. Forthcoming. The Development of Language. Oxford: Blackwell. Mahajan, Anoop. 1990. The A/A-bar Distinction and Movement Theory. PhD Dissertation, Massachusetts Institute of Technology, Cambridge. Maling, Joan. 1978. An Asymmetry with Respect to Wh-Islands. Linguistic Inquiry 9: 75-89. Martin, Roger, David Michaels and Juan Uriagereka, eds. In Press. Step by Step: Essays on Minimalism in Honor of Howard Lasnik. Cambridge, MA: MIT Press. Milsark, Gary. L. 1977. Toward an Explanation of Certain Peculiarities of the Existential Construction in English. Linguistic Analysis 3: 1-29. Miyagawa, Shigeru. 1987. Wa and the Wh-Phrase. In J. Hinds, S. K. Maynard and S. Iwasaki, eds. Perspectives on Topicalization: The Case of Japanese ‘wa’. Amsterdam: John Benjamins, 185-217. Moltmann, Friedrike. 1991. Scrambling in German and the Specificity Effect. Ms., Massachusetts Institute of Technology, Cambridge. Müller, Gereon. 1997. Optional Movement and the Interaction of Economy Constraints. In C. Wilder, H.-M. Gärtner and M. Bierwisch, eds. The Role of Economy Principles in Linguistic Theory. Berlin: Akademie Verlag, 115-145. Müller, Gereon and Wolfgang Sternefeld. 1993. Improper Movement and Unambiguous Binding. Linguistic Inquiry 24: 461-507. Muromatsu, Keiko. In Progress. The Syntax of Classifiers. PhD Dissertation, University of Maryland, College Park. [Working Title.] Neidle, Carol, Judy Kegel, Benjamin Bahan, Debra Aarons and Dawn MacLaughlin. 1997. Rightward Wh-Movement in American Sign Language. In D. Beerman, D. LeBlanc and H. van Riemsdijk, eds. Rightward Movement. Amsterdam: John Benjamins, 247-278. Kleanthes K. Grohmann 87 Pesetsky, David. 1982. Paths and Categories. PhD Dissertation, Massachusetts Institute of Technology, Cambridge. Pesetsky, David. 1987. Wh-in situ: Movement and Unselective Binding. In A.G.B. ter Meulen and E. Reuland, eds. The Representation of (In)definiteness. Cambridge, MA: MIT Press, 98-129. Pesetsky, David. 1998. Lifestyles of the Which and Famous: English is Bulgarian. Talk presented at Colloquium Series, University of Maryland, College Park (March 6). Pollock, Jean-Yves. 1989. Verb Movement, UG and the Structure of IP. Linguistic Inquiry 20: 365- 424. Postma, Gertjan. 1995. Zero Semantics: A Study of the Syntactic Conception of Quantificational Meaning. PhD Dissertation, Leiden University/HIL. The Hague: Holland Academic Graphics. Reinhart, Tanya. 1995. Interface Strategies. Ms., OTS, Rijksuniversiteit Utrecht. Richards, Norvin. 1997. What Moves Where When in Which Language? PhD Dissertation, Massachusetts Institute of Technology, Cambridge. van Riemsdijk, Henk. 1997. Left Dislocation. In E. Anagnostopoulou, H. van Riemsdijk and F. Zwarts, eds., 1-10. Rizzi, Luigi. 1978. Violations of the Wh-Island Constraint in Italian and the Subjacency Condition. Montreal Working Papers in Linguistics 11. [Reprinted in Issues in Italian Syntax, 1982, Dordrecht: Foris.] Rizzi, Luigi. 1997. The Fine Structure of the Left Periphery. In L. Haegeman, ed. Elements of Grammar: A Handbook of Generative Syntax. Dordrecht: Kluwer, 281-337. Rudin, Catherine. 1985. Aspects of Bulgarian Syntax: Complementizers and Wh-Constructions. Columbus, OH: Slavica. Rudin, Catherine. 1988. On Multiple Questions and Multiple Wh-Fronting. Natural Language and Linguistic Theory 6:445-501. Stowell, Tim and Filippo Beghelli. 1994. The Direction of Quantifier Movement. Paper presented at GLOW, Vienna. Takahashi, Daiko. 1993. Movement of Wh-Phrases in Japanese. Natural Language and Linguistic Theory 11: 655-678. Thornton, Rosalind. 1995. Referentiality and Wh-Movement in Child English: Juvenile D- Linkuency. Language Acquisition 4: 137-175. Tsai, W.-T. Dylan. 1994. On Nominal Islands and LF Extraction in Chinese. Natural Language and Linguistic Theory 12: 121-175. Ura, Hiroyuki. 1994. Varieties of Raising and the Feature-Based Bare Phrase Structure Theory. MIT Occasional Papers in Linguistics 7. Uriagereka, Juan. 1988. On Government. PhD Dissertation, University of Connecticut, Storrs. Syntactic Inquiries into Discourse Restrictions on Multiple Interrogatives88 Uriagereka, Juan. 1993. Specificity and the Name Constraint. In C.A. Mason, S.M. Powers and C. Schmitt, eds. University of Maryland Working Papers in Linguistics 1: 121-143. Uriagereka, Juan. In Press. A Minimalist Dialogue. Cambridge, MA: MIT Press. Vallduví, Enric. 1990. The Informational Component. PhD Dissertation, University of Pennsylvania, Philadelphia. deVilliers, Jill, Thomas Roeper and Jürgen Weissenborn. 1996. Superiority: Syntax or Semantics? Talk given at What the Child Has to Say about Linguistic Theory, Rijksuniversiteit Utrecht. Watanabe, Akira. 1992. Wh-In-Situ, Subjacency and Chain Formation. MIT Occasional Papers in Linguistics 2. Wilder, Chris, Hans-Martin Gärtner and Manfred Bierwisch, eds. 1997. Economy Principles in Linguistic Theory. Berlin: Akademie Verlag. Wiltschko, Martina. 1997. D-Linking, Scrambling and Superiority in German. In W. Abraham, ed. Groninger Arbeiten zur germanistischen Linguistik 41, 107-142. Wu, Jianxin. 1996. Wh-Topic, Wh-Focus and Wh-In situ. In J.C. Castillo, V. Miglio and J. Musolino, eds. University of Maryland Working Papers in Linguistics 4: 173-192. Zwart, C. Jan-Wouter. 1993. Dutch Syntax: A Minimalist Approach. PhD Dissertation, Rijksuniversiteit Groningen. Zwart, C. Jan-Wouter. 1996. “Shortest Move” Versus “Fewest Steps”. In Abraham et al., 305- 327. Zwart, C. Jan-Wouter. 1997a. Morphosyntax of Verb Movement: A Minimalist Approach to the Syntax of Dutch. Dordrecht: Kluwer. Zwart, C. Jan-Wouter. 1997b. Transitive Expletive Constructions and the Evidence Supporting the Multiple Specifier Hypothesis. In W. Abraham and E. van Gelderen, eds. German: Syntactic Problems—Problematic Syntax. Tübingen: Niemeyer, 105-134.