The MLC and Derivational Economy Gisbert Fanselow Introduction There is a certain tension between the role which the Minimal Link Condition (MLC, (1)) plays in at least the minimalist theories of syntax, and the existence of numerous (apparent or real) counterexamples such as (2) that arise in multiple questions. For such questions, the MLC seems to imply strict superiority effects. In particular, wh-objects should not be able to cross whsubjects on their way to Spec,CP. More often than not, this prediction fails to be observed. Put differently, the question arises as to why the MLC is respected strictly by head movement, and more of less so by A-movement, while it is a fairly poor predictor for grammaticality when the proper way of carrying out operator movement is at stake. (1) Minimal Link Condition (MLC) α cannot move to γ if there is a β that can also move to γ and is closer to γ than α (2) Constructions violating the superiority condition a. which book did which person read? b. was hat wer gelesen what has who read “what was read by whom?” (German) If correct, this characterization of the problem already suggests a solution: the MLC must be interpreted as a principle that is sensitive to interpretation/ expressivity (cf. also Kitahara (1993), (1994), Reinhart (1995), Sternefeld (1997)). Whenever it does not make a semantic difference whether the MLC is respected or not, the MLC must be obeyed strictly. However, the MLC is never (by itself) able to block a movement operation that is inevitable for expressing a certain meaning. Consequently, to the extent that head movement does not have any semantic effects, the MLC governs head movement 74 Gisbert Fanselow in a strict and exceptionless way. To the extent that different ways of carrying out operator movement are crucial in establishing different semantic relations, the MLC effects we observe in this domain are modulated by considerations of interpretation. Originally, the idea that the MLC decides between those structural alternatives only that have identical meanings was motivated by data involving different scope assignments to wh-operators (see sect. 1). The present paper argues that the required meaning identity must also involve distinctions of information structure (sections 3 and 4), which explains why many (if not most) languages are like German in not showing simple superiority effects at all. Languages like English and Bulgarian fit into such a picture as well – there is no variation among languages in this respect. Furthermore, we concur with Sternefeld (1997) in the claim that the MLC must be applied in a cyclic rather than global fashion (section 2.4), and we argue that it involves reference to LF-identity rather than meaning identity in a broad sense. 1. The MLC and wh-phrase scope The MLC is a core principle of current syntactic theorizing, and has been made responsible for a wide variety of syntactic generalizations, such as the Head Movement Constraint of Travis (1984), the intervention effects restricting A-movement to subject position (Chomsky 1993, 1995, Stepanov 2001, this volume), and the superiority effect governing the formation of multiple questions. In spite of the important role it plays in determining whether syntactic computations are formally correct, some aspects of multiple questions require that the MLC is sensitive to the interpretation of the structures or derivations that it compares. Before we discuss this fact, let us consider some simple superiority effects in English. Object wh-phrases cannot cross c-commanding subject wh-phrases (3), as was observed by Kuno and Robinson (1972). Haider (this volume) argues that the contrast in (3) involves a grammatical constraint that bans wh-phrases occupying the subject position of finite clauses (such as the Empty Category Principle of Chomsky 1981). (3) Simple subject-object asymmetry a. (It does not matter) who bought what b. (It does not matter) *what who bought _ The MLC and derivational economy 75 Independent of whether such a factor contributes to making (3b) worse than (3a), the special status of the subject position cannot be the only source for superiority effects: wh-objects must not cross wh-subjects even when the latter are lexically governed, as in (4). Likewise, a wh-object from a lower clause cannot cross a wh-object from a higher clause on its way up to Spec,CP (5). The interaction of clausemate objects yields identical intervention effects, as evidenced by the contrasts in (6). (4) Subject-object asymmetry not involving proper government a. who do you expect _ to do what? b. *what do you expect who to do _ (5) Biclausal object-object-asymmetry a. who do you persuade _ to do what b. *what do you persuade who to do _ (6) Superiority effects among objects a. *what did you give who _ b. who did you give _ what c. what/which check did you send _ to who d. *who(m) did you send what/which check to _ As Hendrick and Rochemont (1982) correctly point out, data such as (4) – (6) are incompatible with the view that the superiority effect can be completely reduced to the ECP or a similar principle. What is called for is an account along the lines originally proposed by Kuno and Robinson (1972): A wh-DP a cannot cross a structurally higher wh-DP b when moving to Spec,CP. This generalization derives from the MLC in (1) straightforwardly. One notorious difficulty of purely formal accounts of the superiority condition derives from the fact that pairs of wh-phrases that take different semantic scope need not obey the MLC, as (7) illustrates (see, e.g., Huang 1982, Lasnik and Saito 1992). If the lower occurence of who in (7a) takes matrix scope, the sentence is fine, although the movement of what across who fails to obey the MLC. If the lower who takes scope over the complement clause only, (7a) is as ungrammatical as (3b). The effect is not confined to clausemate wh-phrases. Of ten English native speakers (all linguists) that I consulted, seven accepted (7c), and five did not even find (7d) objectionable. 76 (7) Gisbert Fanselow Absence of superiority effects for wh-phrases with different scope a. who wonders what who bought? b. who wonders who bought what? c. who wonders what John persuaded who to buy __ ? d. who wonders what John told who that he should buy __ ? According to Golan (1993), Kitahara (1993), and Reinhart (1995, 1998), such facts suggest that the MLC must be interpreted as an economy constraint related to LF-outputs (meanings). Whenever there is no other way to express a certain meaning, the MLC need not be respected. Let us consider (7a) in more detail. Overt movement of a wh-phrase to Spec,CP fixes its scope. A wh-phrase merged in a complement clause can thus take matrix scope under two conditions only: it moves to the Spec,CP position of the matrix clause, or it stays in situ, and gets scope-bound by an element in the matrix clause. It must not, however, be placed into the Spec,CP position of the complement clause, and still take matrix scope. Thus, the subject of the complement clause who can take matrix scope in (7a) only if it stays in situ. In other words, it can take matrix scope only if crossed by the lower wh-phrase what targeting the complement clause Spec,CP position. The meaning (8a) of (7a) simply cannot be expressed differently – (7b) means something else (viz. (8b)). Whether the MLC is respected or not is irrelevant when the structural alternatives differ in interpretation. (8) a. For which persons x,y: x wonders what y bought b. For which person x, and for which z: x wonders who bought z In contrast to what holds for (7), the two derivational alternatives in (3) do not yield different interpretations: there is only one scope option available for the two wh-phrases. In such a situation (and only in such a situation), the MLC filters out derivations that are not in line with it. Further English constructions illustrating that the application of the MLC depends on the interpretation arrived at will be presented in sections 2.1. and 2.3. Given that the wellformedness of (7a) is of some theoretical importance, it is surprising that little evidence from other languages has entered the discussion of the interpretation-sensitivity of the MLC. According to one of my informants (Koyka Stoyanova, p.c.), (9a,b) are as fine in Bulgarian as they are in English if the second occurence of koj is stressed, but in her dialect, the order kakvo koj is grammatical in simple multiple questions, too. The MLC and derivational economy 77 Penka Stateva, my second Bulgarian informant, does not accept the order kakvo koj in a simple clause, and rejects (9) as well. No contrast such as the one between (3) and (7) exists in Bulgarian. The absence of this contrast will be explained in section 4.1: we argue there that the ordering restrictions among Bulgarian wh-phrases are not caused by the MLC. The ungrammaticality of (9) in some dialects therefore does not bear on the issue of the interpretation sensitivity of the MLC. (9) Anti-superiority in Bulgarian a. #koj se chudi, kakvo koj who wonders what who “who wonders what who bought?” kupi? bought b. #na kogo kaza, who.dat you-tell kupi? bought kakvo what koj who For other languages, it is not much easier to construct relevant evidence, because the simple superiority effect exemplified in (3) is not a widespread phenomenon. The following data from German, however, provide further evidence for the interpretation sensitivity of the MLC. (2b) has already shown that the formation of multiple questions is not affected by the MLC in German (at least superficially) when clausemates are involved, but it has been claimed frequently that a wh-phrase from a lower clause cannot cross a matrix wh-word. (10) Superiority for non-clausemates in standard German a. *wen hat wer wh.acc has who.nom b. gehofft, dass hoped that Irina Irina einlädt invites wer hat gehofft, dass Irina wen einlädt “who has hoped that Irina will invite who?” There are reasons to doubt, however, that the ungrammaticality of (10a) (in the standard dialect) is caused by the MLC. Superiority effects disappear when the wh-phrases are discourse-linked in the sense of Pesetsky (1987). However, (10a) does not improve in the standard language when d-linked wh-phrases are used. Thus, what rules out (10a) must be different from the MLC. 78 Gisbert Fanselow (11) *welchen dass which.acc that Studenten Irina hat welcher einlädt Professor gehofft, student.acc Irina has which.nom professor invites hoped “which professor has hoped that Irina invites which student?” In less restrictive dialects (such as the one spoken by the author), all sentences in (12) are acceptable up to a certain degree, but (12a) and (12b) have different interpretations. If (12a) is completely wellformed at all, the sentence allows a single-pair interpretation only. A pair-list-reading is available for (12b) only, i.e., for the structure which violates the MLC. In addition the “scope-marking” construction (12c) allows the pair-list-reading as well. (12) Nonstandard German: Subordinate clause wh-elements crossing matrix wh-phrases a. (?)wer hat who has gehofft, dass hoped that b. wen hat who.acc has wer who.nom c. was what wer who hat has Irina Irina gehofft, hoped wen who dass that einlädt invites Irina Irina gehofft, wen Irina hoped who.acc Irina einlädt invites einlädt invites How can these data be understood?1 In quite a number of languages, in situ wh-phrases cannot take scope out of the minimal (finite) clause they are contained in. Hindi is a case in point (see Mahajan 1990). The scope of an in-situ wh-phrase must be determined by linking it to a higher wh-phrase, or to a scope marker. The linking might be arrived at in various ways (binding, covert movement), but the important observation concerning Hindi and other languages is that linking is subject to strong locality requirements. In contrast to what holds for overt movement (= wh-scrambling in the case of Hindi), finite CPs are barriers for the linking relation. Consequently, (13) is ungrammatical because the lower occurence of kis-ko must be linked to a whphrase or a scope marker, but cannot be so because it is embedded in an island for linking. The MLC and derivational economy 79 (13) Clauseboundedness of the binding of in situ wh-phrases in Hindi *Raam-ne kis-ko kahaa ki Sitaa-ne Raam-erg who.dat told that Sita-erg “who did Ram tell that Sita saw who?” kis-ko dekhaa who saw? Let us now come back to (12). First, we want to explain why (12a) is out with a pair-list interpretation. This follows if (the relevant version of) German resembles Hindi in that finite CPs are barriers for the scope linking of in situ wh-phrases. Consequently, wen cannot be scope-linked to wer in (12a), which renders the structure ungrammatical under the intended interpretation. Finite clauses are not, however, barriers for overt movement. Therefore, there is a way of constructing a Logical Form for (12) in which both whphrases take matrix scope, viz. by moving the wh-element from the complement clause into the matrix-Spec-CP position, and by scope-linking the matrix subject to the matrix Spec,CP position. This is what has happened in (12b). None of the relations established there is in conflict with locality requirements – but the MLC is violated. Apparently, this MLC-violation is licensed because the relevant Logical Form cannot be arrived at in a different way – the structure (12a) respecting the MLC is incompatible with the locality of the licensing of wh-phrases in situ. (12) illustrates the same phenomenon as (7), but in a rather different context. The other examples in (12) illustrate two further points. (12c) shows that German is like Hindi in having a wh-scope-marking construction, in which a scope marker (was) rather than the real wh-phrase appears in Spec,CP. (12c) is well-formed in all dialects of German, and expresses a pair-list interpretation. A minor point illustrated by this example is that finite clauses are islands for scope taking in German only for wh-phrases that do not occupy a Spec,CP position (note that the lower wh-phrase is fronted in the complement clause). There are various ways of analysing the construction (see, e.g., the contributions in Lutz, Müller and von Stechow 2000), but details are irrelevant for the more important point: long wh-movement in (12b) and wh-scope marking in (12c) yield the same interpretation, but the wh-scope-marking construction (12c) avoids an MLC violation, in contrast to (12b). This shows that the sensitivity of the MLC to interpretation cannot involve a simple, “global” concept of meaning identity. If it would, the wellformedness of (12c) should imply that the MLC is able to rule out (12b). Given (12c), no MLC-violation is necessary for expressing the “meaning” of (12b). The MLC must therefore not be sensitive to “meaning identity” in a global sense. Rather, the identity of interpretation that is rele- 80 Gisbert Fanselow vant for the applicability of the MLC must be a matter of identical (or closeto-identical) Logical Forms. The LF of (12c) is different from the one of (12b) (see in particular Fanselow and Mahajan (2000) for arguments), and therefore, (12c) does not count when the grammaticality of (12b) is established. Haider (1997: 221) exemplifies the claim that complement clause whphrases may cross matrix wh-phrases in German with examples such as (14). To me, (14) invites a single-pair answer only, so that (14) is not fully comparable to the multiple questions discussed so far. Furthermore (14) involves apparent movement from a V2-complement clause, and the theoretical status of such an operation is quite unclear, see Reis (1996, 1997) for arguments that the construction is parenthetical. I therefore refrain from discussing such examples in more detail. (14) Superiority violations in a construction with extraction out of a V2 complement wemi Bild hat wer verkauft]? gesagt [ei habe who.dat hat who.nom said picture sold sie ei ein has.subjunctive she a “who said she had sold a picture to whom?” Our argumentation presupposes that single-pair interpretations of multiple questions (for which (12a) seems marginally acceptable) have a derivation different from the one for multiple questions with a pair-list reading. This claim is supported by the observation that further constructions are ungrammatical with a pair-list reading, but acceptable under a single-pair interpretation. E.g., most native speakers of German (including the author) reject (15) as a question asking for pair-lists, but the single pair interpretation is fine. (15) Multiple adjunct question with a single pair interpretation wie hat er es warum geschrieben how has he it why written “how did he write it, and why” Examples such as (7) show that the applicability of the MLC depends on the interpretation of the structure that it would block. German data such as The MLC and derivational economy 81 (12) constitute further evidence for this. At the same time, the data in (12) shows that the MLC is not sensitive to “meaning” in a global sense – rather, it is the nature of the LF that a movement operation creates that determines whether the MLC must be respected. 2. The MLC and expressivity The strongest conclusion one can draw from from the discussion in the preceding paragraph is that requirements of semantic expressivity always override the MLC. A structure violating the MLC is ungrammatical only if the Logical Form it would express can be arrived at with a structure respecting the MLC. In this section, we defend this strong conclusion against potential counterexamples, and discuss how the MLC can be applied in local fashion. First, we discuss the interaction of the MLC with the that-trace filter. Section 2.2 focuses argument-adjunct asymmetries, while section 2.3 is dedicated to nestedness effects, which have been related to the MLC. Finally, we will briefly discuss what a cyclic application of the MLC might look like. 2.1. Interactions with the ECP As one of the anonymous reviewers has pointed out, the absence of a contrast in (16) might pose a problem for the idea that the MLC applies only if that does not prevent a certain interpretation from being expressed: (16) Two wh-phrases merged in a finite complement clause a. *who do you think that _ bought what b. *what do you think that who bought _ (16b) violates the MLC, so its ungrammaticality is expected. However, the constellation that respects the MLC, viz., (16a), is ungrammatical as well because of a that-trace-filter violation. In contrast to what we saw in section 1, the MLC violation of (16b) is not tolerated by the grammatical system of English, in spite of the fact that this renders the interpretation of (16b) inexpressible. The absence of a contrast in (16) does not show, however, that the MLC is able to block structures even if the competing structure respecting the 82 Gisbert Fanselow MLC violates a further condition on LF. Aoun et al. (1987) and others have argued that the principle Q responsible for the that-trace effect applies at PF, and not at LF. Consequently, Q cannot interact with the MLC: the MLC applies to LFs, and compares derivations that yield (close-to-) identical LFs. It is blind to what happens in other branches of the derivation. A structure that has an optimal LF and is accepted by the MLC need not be in line with further PF-requirements, rendering the LF unpronouncable. Given this relevance of PF-constraints, (16) does not exclude an interpretation of the MLC that compares different ways of arriving at essentially the same LFs – while it falls in line with an other conclusion arrived at in section 1: the MLC is not a principle that takes care of “expressivity” in a literal sense. The following observation leads to a modification of our analysis of (16), which leaves the crucial point intact, however: the MLC responds to the need of respecting further LF-constraints, but it is blind to what happens in the PF-branch of grammar. Haider (this volume) argues that there is an extra constraint banning wh-phrase occupying the specifier position of a finite IP in English. The constraint is independent of the MLC, since it shows its force even in constructions that do not involve a crossing wh-dependency, as was already observed by Chomsky (1981). Interestingly, as (17) illustrates, the relative degree of (un-)acceptability involves dimensions such as discourse-linking (see 17b), and, as Bresnan (1972) has observed, in situ wh-subjects are much better when they appear in subjunctive clauses. (17) Wh-subjects in situ a. *who believes that who loves Irina? b. ?who believes that which man loves Irina? c. ?who demands that who be arrested? The ungrammaticality of (16b) might therefore also be caused by the presence of an in situ wh-subject in a finite clause quite independent of the MLC. Given the contrasts in (17), one would expect that structures like (16b) improve if, e.g., the complement clause appears in the subjunctive mood. In such a construction, the overt movement of the subject of the complement clause still implies a that-trace filter violation, but the additional ban against in situ wh-subjects is now much less strict. According to Anthony Green and Sue Olsen (p.c.), (18a) is indeed much better that (16b). The MLC and derivational economy 83 (18) Missing superiority effect for extraction out of a subjunctive complement a. (?)what do they require that who buy? b. *who do they require that buy what c. *what do you expect who to buy d. who do you expect to buy what If the contrast between (16b) and (18a) generalizes, we have a further example from English that shows that the MLC does not block a construction (viz., (18a)) if the structure that conforms to the MLC (viz., (18b)) violates a different principle. The contrast between (18a) and (18b) would force upon us the assumption that the that-trace filter banning overt subject movement in fact applies at LF, and not at PF. Otherwise, its effects would not be visible to the MLC, as necessary for (18a). Consequently, the PF-located constraint that is invisible to the MLC (as required for (16)) is rather the further ban against in situ wh-subjects argued for by Haider (this volume) and not the that-trace filter. It should finally be noted that the contrast between (18a) and (18c) is due to the fact that the MLC-respecting competitor is well-formed in the case of (18c), but not in the case of (18d). 2.2. Adjuncts Multiple questions with adjunct wh-pronouns constitute a second domain that is relevant for the status of the MLC as an economy constraint. None of the structures in (19) is grammatical – although there is no other (monoclausal) way of expressing the intended interpretations. (19) Adjunct effects in English a. b. c. d. *who came why *why did who come *who spoke how *how did who speak? The MLC clearly picks (19a,c) rather than (19b,d), and correctly so in the light of (20). (19a,c) are blocked by some requirement (see, e.g., Haider, this 84 Gisbert Fanselow volume, Reinhart 1995, Hornstein 1995, among many others) that excludes the adjuncts how and why in any position but Spec,CP. (20) a. who spoke when? b. who spoke in what way? Again, the question arises as to why the MLC cannot be overriden in this context – yielding (19b,d), which do not violate the strong constraint against how and why appearing in situ. Note that (21) is ungrammatical: only one out of some twenty linguists with English as a native language who I consulted accepted the sentence with a downstairs interpretation of how. Unlike what we saw in the preceding section in the context of (18), the ungrammaticality of (19b,d) can not be explained in terms of an additional constraint filtering out wh-phrases in the subject position of non-subjunctive clauses. (21) *how does the police demand that who be treated _ We will propose two accounts of (19) that allow us to maintain that the MLC is ignored when a certain LF cannot be constructed otherwise. As Haider (this volume) has pointed out, adjunct effects of the sort exemplified in (19) are absent in OV languages, as (22) illustrates. This observation excludes the idea that (19a,c) are ungrammatical on simple semantic grounds. (22) Missing Adjunct Effects in OV-languages a. wie het hoe gedaan b. wer es wie gemacht hat how done who it heeft (Dutch complement question) (German complement question) has Haider suggests that higher-order wh-operators such as how and why must c-command the head of the phrase they are applied to. Higher order adverbs range over events, so how and why should c-command the element that situates the proposition in time, i.e., how and why must c-command the (finite) verb. This condition is fulfilled in (22a,b), but not in (19a,c). Movement of the finite verb to Comp does not render wh-adjuncts in situ ungrammatical in Dutch or German. This is in line with the general observation that verb second movement is invisible at the level of Logical Form, either because it is reconstructed, or because it applies in the phonological component of grammar. The MLC and derivational economy 85 (22) Missing Adjunct Effects in OV-languages c. wie heeft d. wer hat who has het es it hoe wie how gedaan gemacht done (Dutch matrix question) (German matrix question) The account suggested by Haider (this volume) cannot be fully correct, however, because Swedish is not in line with it. All of my five informants accepted (23a), and three of them found (23b) grammatical, in spite of the VO-nature of Swedish. (23) Missing adjunct effects in Swedish a. vem skrattade varfoer who laughed why b. Det spelar ingen roll vem som skrattade varfoer it plays no role who that laughed why “it does not matter who laughed for what reason” Similarly, Richards (2001: 18–19) reports adjunct effects for the SOV language Tibetan. Therefore, a different solution is called for. Rizzi (1990: 47) has proposed that certain wh-adjuncts (corresponding to sentence-level adverbs) are based-generated in Comp. One way of translating this proposal into the current discussion consists of the assumption that certain wh-elements are required to appear at the left periphery of clauses on a language particular and item-specific basis. Because of (22) – (23), this idiosyncrasy of how and why (and French pourquoi) cannot be reduced to semantic considerations alone. One way of spelling this idea out lies in the assumption that the MLC applies cyclically (see below for details), while the constraints forcing how and why into Spec,CP are representational principles checking the wellformedness of completed Logical Forms. The MLC would therefore apply prior to the constraints affecting higher order wh-phrases, with the desired effect: the MLC picks (19a,c), and these sentence are blocked at too late a point in the derivation for undoing the impact of the MLC. The account sketched so far predicts the data as judged in (19) and (24). The MLC forces the subject to move to Spec,CP in a multiple question involving subjects and adjuncts, but the resulting structure is blocked because why and how cannot appear in any position but Spec,CP. On the other hand, when adjuncts interact with objects, the MLC will make (24b) block (24a). (24b) is also in line with the requirement that English wh-adjuncts appear at the left periphery. 86 Gisbert Fanselow (24) Adjunct-object interaction in multiple questions a. *what did Bill buy why b. why did Bill buy what Hornstein (1995: 147–149) reports further data such as (25) that may in fact lead to a simpler analysis. If his judgements are correct, why (unlike its Dutch, German and Swedish counterparts) cannot appear at all in multiple questions, quite independent of the position it appears in. (25) wh-adjuncts blocked in multiple questions a. *I wonder why Bill left when b. *I wonder why Bill lives where c. *I wonder why which person came d. *I wonder why you bought what e. *why does John expect who to win If Hornstein is correct, wh-adjuncts come in two varieties. German wie “how” and warum “why” are linked to a semantic representation that makes them eligible for multiple questions, whereas how and why cannot appear there. Under such an account, all sentences in (19) are simply gibberish, and we need not care about what the MLC would predict for them. Whether this simplification is tenable or not depends on the status of (24b). If grammatical, this sentence is incompatible with the idea that why cannot appear in multiple questions. The simplification thus presupposes that (24b) involves an “illusion of acceptablity” (Hornstein 1995: 148). We need not settle the issue here, because the idea that the MLC is an economy constraint can be maintained in the account discussed earlier as well. 2.3. Nestedness A third domain sheds light on the question of whether the MLC is sensitive to LF-identity or not: nestedness effects. It has been suggested that the nestedness effect can be derived from the MLC, see Richards (2001) for a detailed proposal. If this suggestion is correct, the application of the MLC could not be confined to structural candidates yielding the same LF. The MLC and derivational economy 87 In English, the interaction of two wh-phrases moving to two different Spec,CP positions is governed by a nestedness effect (see Fodor 1978, Pesetsky 1982): the dependencies formed by the two wh-chains must not cross – one path must be embedded in the other. The nestedness condition is respected, even when it blocks the expression of a certain interpretation, as it does in (26b) and (27b).2 (26) Nestedness Effects a. ?Which violinj do you wonder which sonatai to play _i on _ j b. *Which sonataj do you wonder which violini to play _i on _ j (27) a. ?Whatj did you decide [whoi [to persuade ti [to buy tj]]] (Oka 1993: 255, (2a)) b. *Whoi did you decide [whatj [to persuade ti [to buy tj]]] (Oka 1993: 255, (2b)) The constraint responsible for nestedness is respected even though the meanings of (26b, 27b) are different from the one expressed by (26a, 27a). Such observations are relevant for the present discussion to the extent that claims made by Richards (2001) and others are correct that the nestedness condition reduces to the MLC. If it does, (26) and (27) would not be in line with the idea that the MLC is ignored when a certain LF could not be formulated otherwise. Under what conditions does the MLC imply nestedness effects? Consider an abstract representation such as (28a), with two wh-phrases that both could be attracted by either of CompA and CompB. When the derivation reaches the point at which CompB attracts a wh-phrase (at which the specifier of CompB must be filled by a wh-phrase), a “blind” application of the MLC implies that wh1 only can move, forming (28b). At a later stage in the derivation, CompA attracts (the specifier of CompA must be filled by a wh-phrase). Let us confine our attention to a situation in which wh1 has already reached its scope position in (28b). Therefore, it cannot undergo further movement. What will happen in such a situation? (28) a. [CompA … [CompB [ .. wh1 .. [ wh2 … ]]]] b. [CompA … wh1 [CompB [ .. wh1 .. [ wh2 … ]]]] c. [wh2 CompA … wh1 [CompB [ .. wh1 .. [ wh2 … ]]]] 88 Gisbert Fanselow If the MLC applies blindly irrespective of whether the LF it generates is wellformed or not, then only wh1 can move to CompA (wh2 cannot move because of intervening wh1), which implies that the derivation breaks down, because a wh-phrase is required to move that must not do so. In this way, aspects of the wh-island condition might be derived, see Chomsky (1995). This would constitute a case in which the MLC rules out a meaning that cannot be expressed otherwise. It is not advisable, however, to derive the wh-island effect from the MLC. In spite of the fact that it respects the superiority condition, English is sometimes quite liberal with respect to wh-islands, as the status of e.g. what do you wonder how to fix suggests. German respects the wh-island condition, but fails to show superiority effects. The two phenomena simply are not correlated with each other. If wh1 is frozen in its position in (28b), i.e., if it cannot move further, and if that is taken into account in the computation of MLC effects, then wh1 does not constitute a b is the sense of (1) repeated below that could go to CompA (1) Minimal Link Condition (MLC) α cannot move to γ if there is a β that can also move to γ and is closer to γ than α Therefore, wh2 can move to CompA (as in (28c)). The derivation leading to (28c) is well-formed, yielding a nested structure, because the lower of two Comps (which attracts first) only attracts the higher of two wh-phrases if movement respects the MLC. In this way, the nestedness condition is derivable from the MLC. Obviously, this reduction of the nestedness condition to the MLC presupposes that the applicability of the MLC does not depend on the existence of a different way of constructing the intended Logical Form. The LF (28d) is different from (28c), so that the fact that (28c) cannot be arrived at in a derivation respecting the MLC is irrelevant for the wellformedness of (28d). (28) d. [wh1 CompA … wh2 [CompB [ .. wh1 .. [ wh2 … ]]]] If one wants to stick to the idea that the MLC triggers nestedness effects, one has to offer alternative accounts of the data presented sections 1 and 2.1 that suggest an interpretation-sensitivity of the MLC. There are, however, good reasons for not endorsing such an MLC-based account of nestedness. Superiority and nestedness do not go hand in hand, as one would expect if the two phenomena were due to the same principle of UG. The MLC and derivational economy 89 For example, Swedish respects the nestedness condition (see Maling and Zaenen 1982: 238f) although it fails to show superiority effects, see (29) and (ii) in endnote 3. Thus, at least in Swedish, nestedness cannot be reduced to the MLC. (29) Absence of superiority effects in Swedish Vad what koepte bought vem who At least certain varieties of Spanish (see (30)) and Catalan exemplify what appears to be an anti-nestedness effect for extractions from wh-clauses: the wh-phrase that is merged in the higher position must also be the one moved to the higher of the two Spec,CP slots. Thus, wh-subjects and wh-indirect objects may cross wh-objects, but not vice versa. (30) Anti-nestedness in Spanish a. *qué libros which books no not quién who ha has leido read b. quién who sabes qué libros you know which books ha has leido read no not c. a quién to who sabes you know no sabes qué libros not you know which books ha has devuelto returned Celia Celia d. *qué libros no sabes a quíen ha devuelto Celia Likewise, Richards (2001: 27) claims that there is an anti-nestedness effect in Bulgarian. Again, the constraint seems uncorrelated with superiority, since simple superiority effects are observed in Bulgarian only, and not in Spanish. (31) Anti-nestedness in Bulgarian a. Koj1 se Who self opitvat da razberat kogo2 try to find out whom b. *Kogo1 se opitvat da razberat t1 e ubil t2 is killed koj2 t2 e ubil t2 In any event, it is hard to draw firm theoretical conclusions from such contrasts, since there is considerable individual variation among speakers of 90 Gisbert Fanselow Bulgarian (see Richards 2001: 28) and of Spanish (at least among the speakers we consulted). This variation suggests that processing factors contribute to generating (anti-)nestedness effects (see also Fodor 1978). Furthermore, nestedness effects have properties are different from those of superiority. Norwegian shows nestedness effects, but only if three (or more) dependencies are involved (Maling & Zaenen 1982). This is unexpected from an MLC perspective: the addition of a third wh-phrase eliminates superiority effects in English. Likewise, at least in English, there is no discourse-linking influence on nestedness: (26b) is bad although both wh-phrases are d-linked in the sense of Pesetsky (1987). Superiority effects fail to show up, however, when the wh-phrases are d-linked. To sum up, there is a number of reasons for not deriving (anti-)nestedness from the MLC. 2.4. Cyclic application of the MLC Our discussion corroborated the view that the MLC is an economy constraint: it does not apply when the relevant LF cannot be generated without violating it. The MLC is, however, insensitive to the issue of whether other components of grammar (such as PF) might prevent the structure selected by it from surfacing. The target LFs that the MLC compares must be very similar to each other. Otherwise, we could not understand the data discussed in section 1: the availability of a wh-scope-marking constructions was shown to be irrelevant for the applicability of the MLC in a structure involving long wh-movement. From a conceptual point of view, the MLC should be a derivational principle that applies when a phrase moves, or when a phase is completed. A cyclic application of the MLC may be called for on empirical grounds as well: if the principles that block in situ wh-subjects in English non-subjunctive clauses, and wh-adjuncts in non-left peripheral positions do not apply to PF, but rather at LF, then we must guarantee that the application of the MLC is not affected by them. This would hold if the MLC is applied cyclically, while the two constraints are representational restrictions on completed LFs. The simplest (but insufficient) way of applying the MLC cyclically and capturing interpretation effects at the same time works with the assumption that attracting Comps come with some index that must be shared by the whphrase to be attracted. The index indicates the target scope of the wh-phrases. Comp can attract a wh-phrase only if the indices borne by the two elements are identical. Therefore, under a strict interpretation of (1), a wh-phrase can The MLC and derivational economy 91 skip another wh-phrase if they have different indices. See, e.g., Sternefeld (1997) for a discussion. What3 can move across the wh-subject in (32a), since who bears the scope index A of the matrix Comp. (32) a. whoA CompA wonders whatB CompB whoA bought _ b. whoA CompA wonders whoB CompB bought whatA c. whoA CompA wonders whoB CompB bought what d. *who CompA wonders whatB CompB whoB bought In such a model, the MLC can be hard-wired into the definition of movement (as proposed by Chomsky 1995): the attracting Comp always triggers the movement of the closest wh-phrase with the same index. Exceptions to the MLC such as (32a) are more apparent than real: whoA cannot be attracted by CompB at all. While being attractive from a conceptual point of view, this model does not account for a number of data we have considered. In (33), the wh-phrases must bear the same index, because they take scope over the same proposition (viz., the whole sentence). Therefore, if Comp attracts the closest wh-phrase with the same index, the sentences in (33) cannot be generated at all – contrary to what is necessary. (33) a. ?what do they require that who buy? b. wen who.acc hat has wer who.nom gehofft, dass hoped that Irina Irina einlädt invites For (33a), it might suffice to assume that the that-trace effect is hard-wired into the definition of movement as well, so that who is invisible to the attracting matrix Comp in (33a). Such a solution cannot be applied for (33b), however, since matrix subjects easily reach Spec,CP in German questions. (33b) and (to a lesser extent) (33a) thus show that a local version of comparing different derivations cannot be avoided in a successful theory of the MLC. This can be made precise as follows. Let us assume that wh-phrases in situ receive their scope by being bound (as was first suggested by Baker (1970), see Dayal (2003) for an overview of non-movement theories of wh-phrases in situ), and that the binding process is itself cyclic. When the syntactic object (34a) has been constructed, a wh-phrase must move to the specifier of CompA if CompA has a feature attracting a wh-phrase. There are four derivations to be considered, 92 Gisbert Fanselow then: either wh1 or wh2 moves to the specifier of CompA, and the wh-phrase remaining in situ may or may not be scope-bound by Comp or the other whphrase. (34) a. [CompA --- [wh1 - - - [ wh2 - - - ]]]] Suppose that the in-situ wh-phrase is not scope bound after movement within (34a). Then we arrive at the representations (34b,c), which are different from each other. Therefore, the MLC-respecting structure (34b) cannot block (34c), if an application of the MLC presupposes that the relevant LF can be generated otherwise. (34) b. [wh1 [CompA --- [wh1 - - - [ wh2 - - - ]]]]] c. [wh 2 [CompA --- [wh1 - - - [ wh2 - - - ]]]]] Wh-phrases that are not scope-bound at all are illegal at LF. Consequently, the two partial derivations in (34b,c) will end up as grammatical only if the wh-phrase left unbound so far is later bound by a higher Comp, or by a higher wh-phrase. This is exactly what happens in (32a,b). These examples show that neither of (34b,c) should be able to block the other. If the derivation proceeds beyond (34b,c), the cyclic nature of wh-binding implies that the scope of the in situ wh-phrase must not be confined to the domain of CompA. Suppose now that the in situ wh-phrase is scope bound after movement. This yields the representations (34d,e): (34) d. [wh1 [Σ CompA --- [wh1A - - - [ wh2A - - - ]]]]] e. [wh 2 [Σ CompA --- [wh1A - - - [ wh2A - - - ]]]]] The two syntactic objects in (34d,e) are certainly not identical, but they differ in a specific way only. The presence or absence of a phonetic matrix should be irrelevant for the constitution of a Logical Form. If we abstract away from the distribution of phonetic features in a syntactic object (and call the result a “partial Logical Form”), then the categories Σ are fully identical in (34d,e). Consequently, the MLC is applicable if it sensitive to the identity of the partial Logical Forms under construction, and if it selects the most economical one of the legal derivations. Normally, the MLC will pick (34d) and block (34e) because the closest phrase must be attracted. However, if there is a factor that applies cyclically and renders (34d) illegal, the MLC will let (34e) pass, since there is no better competing structure left. The that-trace filter The MLC and derivational economy 93 (33a) and the locality requirements for binding (33b) are examples of factors that imply a vacuous application of the MLC. 3. Pragmatic effects In the majority of languages, there are no simple superiority effects for clausemate wh-phrases. The purpose of this section is to integrate the description of these languages into our interpretation of the MLC. Section 3.1 presents the core facts, discusses potential processing influences, and contains further remarks on argument-adjunct asymmetries. Section 3.2. refutes the idea that the absence of simple superiority effects is due to a relaxed definition of closeness, while section 3.3 argues that we also cannot be content with the proposal that the superiority violations are absent because scrambling may precede wh-movement. The economy account envisaged here is discussed in section 3.4. 3.1. The absence of simple superiority effects: some general remarks In a surprisingly large number of languages, intervention effects of the kind exemplified in (3) do not show up in single clauses. Consider, e.g., the examples given in (35), all illustrating (apparent) violations of (1). Other languages belonging to this group are Mohawk, Kashmiri, Malayalam, and the Slavic languages except Bulgarian. (35) Apparent violations of the MLC for clausemate arguments a. Vad what koepte bought vem who (Swedish) b. hva# what keypti bought hver who (Icelandic) c. qué what dijo said quién who (Spanish) d. co what kto who robił did e. nani-o what dare-ga tabeta no who ate f. was what hat has wer what (Polish) gesagt said (Japanese) (German) 94 Gisbert Fanselow Two remarks are in order before we can discuss possible analyses for (35). First, it is often hard to determine whether a language tolerates superiority violations or not. When I asked 22 Dutch linguists via the internet to rate (36), five accepted it and seven found it questionable, while ten speakers rejected the sentence. It not very plausible that this judgment pattern lends support to the claim that there is a categorial difference between, say, Dutch and German with respect to superiority. Likewise, it is not obvious what the marginality of (37) implies for the status of superiority in French. (36) Dutch superiority #ik weet niet wat wie I know not what who “I do not know who has bought what” gekocht bought heeft has (37) French superiority ?Je me demande à qui a I me wonder to whom has “I wonder who has talked to whom” parlé talked qui who Instead of forcing (36) and (37) into one or the other category, the graded nature of such MLC violations should figure in the analysis of the construction.3 This is particularly true in the light of experimental findings concerning judgements by linguistically naive informants. We compared structures such as (38a) and (38b) in a questionnaire study and found a highly significant difference between multiple questions that respect the MLC and those that do not. Structures violating the MLC were rated worse than those respecting it (4.8 vs. 2.34) on a 1-6 scale (1: perfect, 6: completely ungrammatical) by linguistically naive informants. (38) a. Wer besucht wen in der Villa? who visited whom in the villa b. wen besucht wer in der Villa? 2.34 4.80 Given that the syntax literature states more or less unanimously that German lacks simple superiority effects, such findings are a bit surprising at first glance, but they are in line with those obtained by Featherston (2002a,b), and they reappeared in a very similar shape in our questionnaire studies concerning Polish and Russian. The MLC and derivational economy 95 The key to an understanding of this difference between the syntacticians’ wisdom and empirical findings lies in the observation that acceptability judgements are influenced by a variety of factors, among them being processing difficulty. Object initial structures are harder to process than their subject-initial counterparts (as was already shown by Krems 1984 and Frazier and Flores d’Arcais 1989, see also Hemforth 1993, among many others), and it seems to be for exactly this reason that object-initial structures are in general rated worse than subject initial ones in German, irrespective of whether a potential superiority violation is involved or not (see Featherston 2002b). The rating difference between (38a) and (38b) is thus not a proof that there is some underlying MLC-based superiority effect in German, but if this line of reasoning is correct, it is hard to see on what basis one would have to assume a grammatical rather than a processing account for the rating profile for Dutch (36). The second remark concerns the reappearance of argument-adjunct asymmetries in structures violating superiority. Wh-Objects may cross whsubjects in Swedish (35a), but wh-adjuncts do not have such a freedom: my five informants unanimously rejected (39b), and accepted (39a) only. (39) Swedish adjunct superiority a. Vem who skrattade laughed varfoer why b. *Varfoer skrattade vem German, on the other hand, imposes no real restrictions on multiple questions involving warum, ‘why’. We asked 17 non-linguist native speakers of German to rate the grammaticality of (40). 15 of these accepted (40a), and 10 found (40b) grammatical as well. (40) Absence of superiority effects for German adjuncts a. wer lachte warum who laughed why b. warum lachte wer Presumably, this contrast is related to a further difference between Swedish and German. Multiple questions involving two adjuncts were unanimously rejected by the Swedish informants. German shows something reminiscent of a superiority effect in such multiple questions : (42a) was accepted by 9 96 Gisbert Fanselow of 17 informants, while (42b) was judged as grammatical by three informants only. To my ears, (42a) allows a pair-list reading, while (42b) is restricted to a single pair/echo interpretation. See Haider (this volume) for an analysis of languages (not necessarily true for German) in which multiple questions must not involve two adjuncts. Below, we will comment on the apparent superiority effect in (42). (41) Swedish multiple questions involving two adjuncts a. *Varfoer bettedde why behaved sig refl barnen the children hur how b. *Hur betedde sig barnen varfoer (42) German multiple questions involving two adjuncts a. Warum why benahmen behaved sich refl die the Kinder children wie? how? b. *Wie benahmen sich die Kinder warum? 3.2. The absence of simple superiority effects: caused by low subject positions? At least two types of formal accounts for the absence of superiority effects in (35) can be found in the literature, and we will discuss them in turn before we consider a pragmatic explanation. First, the definition of “closeness” central to the MLC might be modified, so that two phrases can be “equidistant” from a target position even if one of them asymmetrically c-commands the other. Second, additional movement operations might reverse the ccommand relations between wh-phrases before wh-movement. Whether a wh-phrase α may cross another wh-phrase β c-commanding α depends on the definition of closeness in (1): If the MLC is defined as in (43), crossing is excluded in general, but if closeness is made precise in a more liberal way, as in (44), the MLC does not restrict the movement of phrases within the same maximal projection. (43) MLC: Strict Version α cannot move to γ if there is a β that can also move to γ and that c-commands α The MLC and derivational economy 97 (44) MLC: Liberal Version α cannot move to γ if there is a β that can also move to γ and that asymmetrically m-commands α Suppose, then, that MLC effects are computed relative to (44). Whether a whobject may be moved across a wh-subject then depends on the hierarchical position of the subject. Subjects are base-generated in the VP. If the subject moves to Spec,IP as in (45a), it asymmetrically m-commands the object. Therefore, an object cannot pass it on its way up to Spec, CP. If the subject stays in VP, as in (45b), the condition for the application of (44) is not met, so that the presence of a wh-subject does not interfere with the preposing of a wh-object. (45) a. [ IP subject [verb phrase [V object]] b. [ IP [verb phrase subject [V object]] (44) thus links the presence or absence of simple superiority effects to an independent parameter, viz., the location of the subject. Indeed, subjects need not move to Spec, IP in many of the languages (among them Spanish or German) that disrespect superiority. The “free inversion” of subjects and verbs in Spanish has always been taken as evidence that Spec,TP can be filled by an empty pleonastic pro, which allows the subject to stay in the verbal projection. (46) Free Inversion in Spanish le regalaron los estudiantes un libro her gave the students a book “the students gave her a book as a present” The view that thematic subjects need not leave the VP in German either is corroborated by constructions in which the VP precedes the second position auxiliary, as was noted by Haider (1986, 1990, 1993): The subject can be part of such VPs (47b,c), a fact suggesting that it need not move to Spec, IP in overt syntax. (47) a. [Mädchen geküsst] hat er noch girls kissed has he not “he has not yet kissed any girls” nie yet 98 Gisbert Fanselow b. [Häuser gebrannt] haben hier noch houses burnt have here yet “houses have never burnt here” c. [Mädchen geküsst] haben ihn noch girls kissed have him yet “girls have not kissed him yet” nie never nie never Thus, there is independent evidence that (45b) is a legal constellation of German and Spanish. An MLC formulated as in (44) will not prevent the object from moving across the subject in (45b). In contrast, subjects must go to Spec,IP in English. Here, (45a) is the only constellation that can underlie multiple questions such as (3). Even in its liberal version (44), the MLC prevents an object from crossing a subject. The choice between (44a) and (44b) is thus a good candidate for an explanation of crosslinguistic variation concerning simple superiority effects (see, e.g., Haider, this volume). While such an approach successfully captures basic superiority facts, more complex data are not readily explained along these lines. Consider Icelandic first. One may want to relate the absence of superiority effects in this language to (44), since the existence of so-called transitive expletive constructions in Icelandic suggests that thematic subjects may be placed into lower positions than in English (see, e.g., Bobaljik and Jonas (1996)). Haider (2000, this volume) observes that movement to Spec,IP is an option for thematic subjects in Icelandic, German, and Spanish, and notes that an explanation of the absence of superiority effects in terms of a low subject position predicts that one should observe English-type asymmetries whenever the position of adverbial material makes it clear that the thematic subject occupies a high position. Haider cites contrasts such as (48) (which he attributes to Ottósson (1989), and H. Sigurdsson, p.c.) as evidence for the claim that this prediction is borne out: (48) Superiority effects in Icelandic and different subject positions a. Hva# hefur what has hver who gefi# given börnunum? the-children? b. *Hva# hefur hver oft gefi# börnunum? what has who often given the-children? “who has often given what to the children?” The availability of a low position for the subject hver in the verbal projection explains the grammaticality of (48a). In (48b), however, the subject precedes The MLC and derivational economy 99 oft ‘often’, i.e., it precedes an element adjoined to VP, and occupies a high position in the clause. The ungrammaticality of (48b) suggests, then, that the position of the thematic subject is the crucial factor governing superiority effects, as Haider argues. The argument presupposes, however, that (48b) becomes perfect when the order of the subject and the adverb is reversed. According to my informant (-orsteinn Hjaltason), this expectation is not fulfilled. Rather, we get the following array of relative judgements: (49) Superiority effects in Icelandic and different subject positions a. Hva# what hefur has hver who gefi# börnunum? given the children b. ?Hva# what hefur has hver who oft often gefi# given börnunum? the children c. ?Hva# what hefur has hver who oft often gefi# given hverjum? whom? d. *Hva# hefur oft hver gefi# börnunum? e. Hva# what hefur has hva#a fa#ir oft which father often gefi# given börnunum? the children f. *Hva# hefur oft hva#a fa#ir gefi# börnunum? (49a) is grammatical because Icelandic shows no superiority effects. (49b) is less acceptable, but this effect is not eliminated by the addition of a third wh-phrase (49c), as it should be if the phenomenon is related to the MLC. Most importantly, the structure becomes fully ungrammatical when the order of the subject and the adverb is reversed. The status of (49d) is quite unexpected, because the order of subject and adverb seems to imply a low position for the former. The ungrammaticality of (49d) is matched by the one of (49f), which involves a d-linked wh-phrase. Whatever may be responsible for the contrasts in (48) and (49) – the MLC is not likely to come into play. Similarly, the grammaticality of (50a) might be related to the low position occupied by the unaccusative subject in this example.4 However, the structure does not degrade dramatically when the subject is placed into the slot preceding the object pronoun (that is, when it presumably moves to Spec,IP): five out of a total of eight Dutch linguists I consulted found (50b) completely unobjectionable. 100 Gisbert Fanselow (50) Different subject positions and Dutch superiority a. wanneer when is is hem him wat what overkomen happened b. wanneer is wat hem overkomen The position of the subject is also not completely irrelevant for the wellformedness of multiple questions in German. Consider the contrasts in (51)5, in which a non-subject has been placed in front of a wh-subject. Such constructions fail to be fully grammatical (to different degrees) when the subject precedes a clitic object pronoun (51b,d), an unstressed object pronoun (51f), or one of the particles like denn which have been claimed to mark the VP boundary in German (51h) (see Diesing 1992, Meinunger 1995, for a discussion of VP boundaries). (51) Different subject positions in German multiple questions a. wann hat’s wer gesehen when has it who seen b. ?*wann hat wer’s gesehen “who saw it when?” c. wem who.dat hat`s has it wer who gegeben given d. ?*wem hat wer’s gegeben e. wem hat es wer gegeben f. ?*wem hat wer es gegeben “who gave it to whom” g. was hat what has denn ptc. wer who gesagt said h. ?*was hat wer denn gesagt “who said what” Multiple questions are less grammatical when a wh-phrase crosses a whsubject that has moved to Spec,IP. It is tempting to explain such contrasts in terms of the assumption that the wh-subject asymmetrically c-commands the trace of the wh-object in the ungrammatical examples, so that the MLC (44) would block the bad structures. The MLC and derivational economy 101 Such an analysis is not convincing, however. It does not take into account the fact that the same or similar contrasts show up in constructions for which the MLC cannot be relevant. German wh-words are ambiguous between an interrogative and an indefinite interpretation. The restrictions on the placement of interrogative wh-subjects exemplified in (51) are exactly mirrored by comparable restrictions on the placement of indefinite wh-subject pronouns, as (52) shows. The indefinite subjects in (52) share the distribution of wh-phrases, but they do not interact with any other element in the clause in terms of the MLC. Therefore, the MLC cannot explain (52), and it would be strange if it accounted for the same distribution of data in (51). (51) and (52) show that German syntax imposes restrictions on the placement of subjects that are not definite. The MLC is not responsible for these. (52) Effects of the subject position for indefinite pronouns a. dann hat’s wer gesehen then has it someone seen “then, someone saw it” b. ??dann hat wer’s gesehen c. dem hat`s wer gegeben him.dat has it someone given “someone gave it to him” d. ?*dem hat wer’s gegeben e. dem hat es wer gegeben f. ?*dem hat wer es gegeben g. hat denn wer angerufen has ptc. someone called “did someone call?” h. ?*hat wer denn angerufen? A central prediction of an account of (absent) superiority effects that exploits differences in the placement of subjects is not borne out: in a number of languages that fail to show superiority effects (German, Icelandic, and perhaps Dutch), the actual position of the subject does not influence the grammaticality of multiple question in the expected way. 102 Gisbert Fanselow 3.3. The absence of simple superiority effects: caused by scrambling? A second attempt of capturing (35) assumes that the object in fact c-commands the subject at the point of derivation when movement to Spec, CP is carried out. Under this circumstance, the MLC does not have to be relaxed in order to explain (35): Given that the order object > subject is in principle always grammatical in a German (53a,b), the question arises whether (53c) really is not in line with even the strictest version of the MLC. After all, (53c) might be derived from (53d) rather than (53e). In the former case, the highest wh-phrase is moved to Spec,CP in (53c), as predicted by the MLC. (53) Object-subject order in German and the MLC a. dass fast jeden jemand that nearly everyone.acc someone.nom “that someone had called nearly everyone” angerufen called hatte had b. dass fast jeden wer that nearly everone.acc someone.nom “that someone had called nearly everyone” angerufen called hatte had c. wen hat wer who.acc has who.nom “who has invited whom?” eingeladen invited d. hat [wen [wer eingeladen]] e. hat [wer [wen eingeladen]] In other words, (53c) might be grammatical because additional movement operations (scrambling) can change the c-command relations established by Merge.6 If the object can in general be placed in front of the subject, structures such as (53d) can be derived in which the wh-object c-commands the wh-subject. Even in its strictest version, the MLC cannot block the subsequent movement of the wh-object to Spec, CP. See, e.g., Fanselow (1998, 2001), Haider (1986), Wiltschko (1998), among others, for different versions of this account. According to Fanselow (1998), the contrasts in (54) corroborate the view that apparent violations of superiority are licensed by scrambling. Certain wh-phrases such as wen von den Studenten (54a) or was für Frauen (54d) can either move to Spec,CP as a whole, or be split up in simple and multiple questions (54b,e). In the latter case, only the wh-part of the phrase under- The MLC and derivational economy 103 goes fronting, whereas the remaining part is stranded. The stranded material indicates the position from which the phrase has been attracted to Spec, CP. The ungrammaticality of (54c,f) suggests, then, that a wh-phrase cannot cross another one in German, either. Objects may undergo overt wh-movement in multiple questions, but only if movement starts in a position c-commanding a wh-subject. (54) Superiority and Splitting von den Studenten hat heute wer eingeladen? a. wen who.acc of the students has today who.nom invited b. wen hat [von den Studenten] heute wer heute eingeladen? c. *wen hat heute wer abends von den Studenten eingeladen who has today who in the evening of the students invited “who has invited which of the students today (in the evening)s” heute d. was für Frauen hat wer what for women has who.nom today “who has invited which kind of women today” eingeladen invited e. was hat für Frauen wer heute eingeladen f. ??was hat wer für Frauen heute eingeladen Pesetsky (2000) points out that contrasts such as the ones in (54) find an explanation in terms of the intervention effects analysed by Beck (1996), see also Mathieu (2002). (55) shows that the parts of a discontinuous wh-phrase must not be separated by any kind of operator in German. An intervention account can explain (54) and (55) at the same time, while the MLC-based explanation for (54) cannot be easily extended to (55). (55) Intervention effects and Split noun phrases a. was hat er für Frauen nicht what has he for women not “what kind of woman did he not meet?” getroffen met b. *was hat er nicht für Frauen getroffen Pesetsky’s observation certainly establishes that data such as (54) cannot be used to show that object wh-movement cannot originate below a wh-subject 104 Gisbert Fanselow in a multiple question. Notice, however, that (54b,e) still show that whextraction of an object may start in a position c-commanding a wh-subject. Reference to (53b), i.e., to the grammaticality of structures in which the object occupies a higher position than the subject, thus seems to be in general a sufficient7 (though not a necessary) condition for the absence of simple superiority effects in a language. Unfortunately, the scrambling solution for (35) cannot be applied in all languages in which superiority effects are absent, because quite a number of them (Swedish, Icelandic, French, Dutch) do not have free constituent order generated by scrambling! 8 3.4. Pragmatics In spite of its shortcomings, the scrambling account has an attractive feature: it implies that the choice between the object- and the subject-initial versions of a multiple question is never arbitrary in the languages that tolerate (35). Scrambling can place an object in front of a subject only if the latter is more focal than the former. Therefore, the scrambling account of missing superiority predicts that apparent superiority violations are acceptable under certain pragmatic circumstances (those that would license scrambling) only. This prediction is borne out. The pragmatic conditions of use of (56a) and (56b) are different. They require different “sorting keys” (Comorovski 1996). Answers to (56b) are well-formed if the object of the clause represents a contrastive topic. There are no comparable restrictions on the wellformedness of (56a). (56) Absence of superiority in German a. wer who.nom hat has wen gesehen who.acc seen b. wen hat wer gesehen “who has seen whom?” This pragmatic dependency becomes evident when one considers the minitexts in (57). The a.- and b. examples introduce the referents of the subject and the object, respectively, as known to the speaker. These referents constitute the “sorting keys” for the multiple questions a’ and b’, they are discourselinked (see Pesetsky 1987). (57a) can only be continued by (57a’), and (57b) only by (57b’). The MLC and derivational economy 105 (57) Discourse influence on superiority violations in German wir we haben have bereits herausgefunden already found out a. wer jemanden gestern anrief, und wer nicht who.nom someone.acc yesterday called and who.nom not b. wen jemand gestern anrief, und wen nicht who.acc someone.nom yesterday called and who.acc not Aber wir sind nicht eher zufrieden, bis wir auch wissen But we are not earlier content until we also know a’. wer who.nom WEN who.acc angerufen called hat has b’. wen WER angerufen hat In other words, a wh-object can precede a wh-subject in German if the former is more topical than the latter. Out of the blue wh-questions allow subject > object order, only. This is particularly clear when the predicate is symmetric (such as treffen, “meet”) as in (58), so that discourse-linked differentiations of subjects and topics are very hard to imagine. (58) Superiority effects in out of the blue contexts Erzähl mir was über die Party. “Tell me something about the party” a. Wer hat who.nom has wen who.acc getroffen? met? b. ??Wen hat wer getroffen “who met who?” Steinitz (1969) was the first to observe that modal or sentence level adverbs resist reordering in the interest of information structure. The adverbial “superiority” effects discussed in (42) can be accounted for in these terms. The languages that lack simple superiority effects do not differ in this respect: constituent order reflects information structure. Different types of operations conspire to guarantee that focal information is preceded by topical one: scrambling (German, Japanese, Polish), topicalization to Spec, CP (Swedish, Icelandic, German), or subject placement in Spec, IP or VP (Spanish, German). In the most parsimonious account, these operations are driven by a constraint C-INF that requires that topical material c-command 106 Gisbert Fanselow focal elements (but more luxurious theories of information structure would have the same effect). If C-INF plays a role in determining the well-formedness of partial LFs in multiple questions as well, then (34d) (repeated here for convenience) is able to block (34e) only if this does not prevent a particular distribution of focality/topicality among the wh-phrases from being expressed within the limits imposed by C-INF. If the higher degree of topicality of wh2 must be expressed, (34e) can be chosen. Information structure overrides the MLC. (34) d. [wh1 [Σ CompA e. [wh 2 [Σ CompA 4. ----- [wh1A - - - [ wh2A - - - ]]]]] [wh1A - - - [ wh2A - - - ]]]]] The Nature of Exceptions While the absence of simple superiority effects in the interest of information structure is a widespread phenomenon, it is far from being universal, as evidenced, e.g., by the relevant English data. The contrast between English and German in the formation of multiple questions might be indicative of the different importance the languages attribute to C-INF: in German, its effects are stronger than the MLC, while it is the other way round in English. The interaction of the MLC and the constrains of information structure would thus be reminiscent of an optimality theoretic framework (for OT accounts of MLC-effects, see, e.g., Müller 2001, and the constributions by Hale and Legendre, Lee, and Vogel, this volume). We will argue, however, that such a conclusion is not warranted. The MLC is never stronger than C-INF. 4.1. Bulgarian Roumanian and Bulgarian are languages cited frequently when one wants to substantiate the claim that superiority effects are not confined to English. (59) Simple superiority effects in Roumanian and Bulgarian a. cine ce cumpara who what buys b. *ce cine cumpara c. koj kogo vizda who whom sees d. *kogo koj vizda The MLC and derivational economy 107 The Slavic languages other than Bulgarian such as Czech, Polish or Russian allow superiority violations, however. The seminal study of Rudin (1988) initiated an impressive series of studies that try to account for this and other differences among the Slavic languages, cf., Błaszczak and Fischer (2002) for an overview. The proposal advanced by Bo‰koviç (2002) (see also Bo‰koviç 1997, Stepanov 1998) is the most interesting one in the context of the preceding section. According to him, wh-phrases move to specifier positions defined in terms of information structure (focus) in Polish, Russian, or German, while movement targets a pure [+wh] specifier in Bulgarian or English. This might fit into the preceding discussion in the following way: when phrases move to [+wh]-specifier, only the attracting feature is grammatically visible, so that additional features of information structure will not interfere with the application of the MLC. However, when XPs are attracted to heads defined in terms of information structure, it is the distribution of the pertinent features that determines how attraction is carried out. It is doubtful, however, that a model drawing a sharp line between Bulgarian and the other Slavic languages is adequate. The intuition represented in (59c,d) is not shared by all native speakers of Bulgarian: two of the five native speakers that I have consulted accept a sentence such as kakvo koj pravi? “what who did” provided that koj “who” is stressed. It is not obvious, then, that the judgement pattern for Bulgarian multiple questions is qualitatively different from the one for German or Dutch. Even if we disregard the empirical issue of whether (59d) is really ungrammatical in Bulgarian (and not just rejected by some speakers, similar to what holds for German or Polish superiority violations), the contrast between (59c-d) is not identical with the one we find in English. A number of differences to English come out clearly. In Bulgarian, strict superiority effects can be found for animate subjects only. When the subject is inanimate, and the object animate, both orders are fine, as Billings and Rudin (1996: 38) have observed. (60) Absence of superiority effects with inanimate subjects of transitive verbs a. Kogo whom.acc kakvo what.nom b. Kakvo kogo e udarilo? “What hit whom?” e CL udarilo? hit 108 Gisbert Fanselow No superiority effects show up with psychological predicates, as (61) illustrates. Sometimes, subject-initial sentences even seem worse than sentences beginning with the dative wh-phrase: (61) Absence of superiority effects with psychological predicates a. Koj na kogo mu who.nom whom.dat CL-dat 3.sg (literally) “Who is likeable to whom?” xaresva? is-pleasing b. Na kogo koj mu xaresva? c. ??Kakvo na kogo mu what.nom to whom.das CL-dat.3.sg (literally) “what is likeable to whom?’ xaresva? is-pleasing d. Na kogo kakvo mu xaresva? Superiority effects are therefore restricted to external arguments of transitive verbs, and even for them, the only defensible generalization is the one offered by Billings and Rudin (1996: 46) “If the wh-external argument is human (i.e., koj), then it must appear first in the wh-cluster.” That such a constraint on wh-clusters may be necessary quite independent of any considerations of superiority is suggested by the fact that Bulgarian differs from English with respect to ternary questions as well. Kayne (1983), Hornstein (1995), Pesetsky (2000) and others have observed that superiority need not be respected in ternary questions: even the lowest whphrase can be fronted. (62) Absence of superiority effects in English ternary questions a. what did who buy where? b. what did who persuade who to buy Cancellation effects due to the addition of a third wh-phrase exist in Bulgarian, too (see (63)), but the examples used in the literature and the intutions of my informant Penka Stateva suggest that the liberalizing effect never affects subject koj. The MLC and derivational economy 109 (63) Restricted liberalization of superiority in Bulgarian a. Koj kogo kakvo e who whom what is “Who asked whom what?’ pital? asked b. Koj kakvo kogo e pital? c. Koj kogo kak e who whom how is “Who kissed whom how?” tselunal? kissed d. Koj kak kogo e tselunal? e. Koj who kogo kŭde whom where e is vidjal? seen f. Koj kŭde kogo e vidjal? Recall also that the ordering restrictions of Bulgarian koj do not show the interpretation-sensitivity of the English superiority effect. The judgements for (9) – repeated for convenience – seem to correlate with the jugdements for simple kakvo koj kupi. If the MLC would be responsible for the ungrammaticality of (59d), it would be unclear why the condition is not interpretation-sensitive in Bulgarian, whereas it is in English and German. (9) Anti-superiority in Bulgarian a. #koj se chudi, kakvo koj who wonders what who “who wonders what who bought?” kupi? bought b. #na kogo who.dat kupi? bought kaza, you-tell kakvo what koj who The MLC is thus not a likely cause for the ordering restrictions in Bulgarian. A simple account can, however, be formulated in terms of the fact that Bulgarian is a multiple fronting language. One of the crucial insights of Rudin (1988) was that the peculiarities in the behavior of Bulgarian (as compared to other Slavic languages) can be related to the fact that Bulgarian is a “multiple filler” language: all wh-phrases must be preposed in a multiple question (unless they are discourse-linked). Suppose that sequences of whpronouns form a cluster, and that the morphophonological realization of this cluster is subject to the kind of rules that also govern the linear arrangement 110 Gisbert Fanselow of sequences of clitics and inflectional affixes. As Bonet (1991), Halle (1992) and Noyer (1992) show, the order of elements in such clusters cannot be exclusively predicted from syntax. Rather, independent principles of morphology are needed, a view that is well-established nowadays in the theory of distributed morphology (Halle and Marantz 1993,1994). There is independent evidence that the composition of wh-phrases in clusters is governed by non-syntactic principles in Bulgarian. Billings and Rudin (1996: 43) suggest that (64b) is ungrammatical because *na kogo kogo ‘to whom whom’ violates a ban against consecutive wh-homophones. In colloquial Bulgarian, na kogo can be replaced by na koj, in which case both orders of the objects are fine: (64) Phonological restrictions in wh-clusters in Bulgarian a. Koj kogo who.nom whom.acc na kogo e to whom.dat CL pokazal? showed b. *Koj na kogo kogo e pokazal? c. Koj kogo who.nom whom.acc na koj to who.dat e CL pokazal? showed d. Koj na koj kogo e pokazal? “Who pointed out whom to who?” It natural to assume that further templatic constraints determine the arrangement of wh-pronouns in the cluster, among them a requirement that koj must come first in a truly transitive construction. This requirement implies the contrast in (59). Since it is a PF constraint, considerations of expressivity will not play a role, as required. Some observations from other languages lend support to the view that cluster formation is crucial in establishing ordering restrictions that resemble (but fail to be) superiority effects. Languages in which cluster formation is optional are of particular relevance here. In Yiddish, multiple fronting of whphrases to a position preceding the verb is possible, but not mandatory. In wh-clusters, word order is strict (65a,b) but it is free when only one wh-phrase is placed into preverbal position (65c,d), see Hoge (2000) for discussion. Likewise, in Hebrew, superiority can be violated only if the verb is placed between the two wh-phrases, although inversion in wh-questions is not necessary as such (66): The MLC and derivational economy 111 (65) Multiple questions in Yiddish a. ver who vemen whom hot has kritikirt? criticised b. *vemen ver hot kritikirt? c. ver who hot has vemen whom kritikirt? criticise d. vemen hot ver whom has who “who criticised whom?” kritikirt? criticised (66) Superiority in Hebrew a. ma kana mi what bought who b. *ma mi kana For obvious reasons, wh-pronouns cannot form a continuous cluster when they are separated by a verb. The data in (65) and (66) can be captured easily in a model that allows for templatic ordering restrictions of wh-phrases which apply when syntax is spelt out. Grewendorf (1999, 2001) and Hoge (2000) account for superiority in Bulgarian by cluster formation as well, but in a fairly different way. 4.2. English English superiority effects are difficult to account for in the model we propose. This is not necessarily a negative aspect: superiority effects in English are distributed in a very complex way, for which it is not clear at all how it could be captured in a simple MLC account. Intervention effects disappear in English when the wh-phrases allow a context-related interpretation. Pesetsky (1987) shows that (67a) is fine because it has a “discourse-linked” interpretation: a wh-phrase is discourselinked if its interpretation relates to a contextually given set of objects and persons, from which one tries to pick a relevant one with the wh-phrase. Thus, the d-linked wh-phrase in (67a) generates s contrastive topic for the answers, as it does in German. As Bolinger (1978) observes, proper contexts even license the absence of intervention effects for wh-pronouns, as in (67b). 112 Gisbert Fanselow (67) Absence of superiority effects in certain contexts in English a. which book did which person read b. I know what everyone was supposed to do. But what did who actually do? However, reference to (67) allone does not explain why (68a) sounds bad to the English ear, while its one-to-one translation into German (68b) is grammatical. (68) a. *what b. was will who see wird wer sehen The key to an understanding of this contrast lies in the observation that who is a topic in (68a), while wer can be focal in (68b), and bear focal stress. If wh-pronouns are inherently indefinite, and constitute bad topics, the different status of (68a,b) can be understood. A number of facts support this view. First, the acceptability of a crossing structure depends of the degree to which the subject wh-phrase can be interpreted as a referential category, as discourselinked, as a potential topic. (69) Crossing effects as a function of the potential topicality of the subject a. what did a friend of who say to Bill? b. what did whose friends say to Bill? c. *what did each friend of who say to Bill (Hornstein 1995: 147) d. *what did how many men buy? Second, Erteschik-Shir (1997: 190) observes that object initial questions are in general relatively bad in English when the subject is an indefinite, a weak quantifier. Obviously, the unacceptability of (70b) (with a non-generic nonspecific reading of a boy) cannot be explained in terms of the MLC. But if indefinites are bad as such in the subject position of questions, one does not need to additionally invoke the MLC. The MLC and derivational economy 113 (70) Non-referential subjects in wh-questions in English a. what did two boys find? b. *what did a boy find? c. which book did two boys find? d. ?which book did a boy find? Summing up, there is reason to believe that the difference between (68a) and (68b) stems from the fact that a wh-subject must be topical in English when it is in situ, while this does not hold for German. Zubizarretta (1998) develops a prosodic theory for accent and focus placement in English which implies that the predicate will be in focus in double questions of English in which the subject is left in situ. Erteschik-Shir (1997) proposes a model of the syntax-information structure interface which also implies topichood for the subject when certain formal dependencies are built up in a clause. In the interest of space, I will not try to assess the merits of these approaches, but confine myself to pointing out that the connection between topichood and in situ wh-subjects apparently need not be stipulated for English. Explanations borrowed from Zubizarretta and Erteschik-Shir may help explaining the status of (68a) – but do they also fit the general model we try to defend here, viz. that the MLC is an interface economy constraint that blocks structures only if their (partial) LF can be arrived at in a more economical way? What is the proper way of expressing questions in which an object wh-pronoun is the sorting key for answers? It is worthwhile to compare the constellations which lead to crossing effects with wh-pronouns in English with those that do not: (71) Structural constellations leading to crossing effects: passive a. who bought what? a’. *what did who buy? a”. what was bought by whom? b. who did you give _ what b’. *what did you give who _ c. what did you give _ to whom c’: *who did you give what to _ 114 Gisbert Fanselow The contrasts in (71) are related to the fact that English expresses information structure distinctions in a way different from scrambling and topicalization. (71) shows that English bans crossing wh-pronouns primarily in those contexts in which it offers an alternative way of making a lower (wh-) phrase more topical than the higher one. For subjects and objects, this alternative way is the passive construction. The conditions of information structure that license counterparts to (71a’) in German are therefore not inexpressible in English. Rather, they imply the use of a passive. (71b-c) illustrate that one can front both the direct and the indirect object in a multiple question, but the options (related to information structure) are linked to the dative alternation. (71b) is unobjectionable because it is in line with the MLC. The MLC-violation in (71b’) would have to be motivated on grounds of information structure (who being more focal than what), but in a dative shift construction, the inner object (who) must be more topical than the outer object. Therefore, (71b’) is ill-formed on pragmatic grounds. (71c) is grammatical since it conforms to the MLC. The information structure requirements that would license the MLC-violation in (71c’) are those that trigger the dative shift alternation. (71c’) is illicit because the proper way to express its information structure is (71b). In an OT-framework, one may feel tempted to explain the data in (71) by assuming a grammatically visible competition between active and passive sentences, or between the constructions V NP PP and V NP NP at the point when the MLC is evaluated, but a more conservative solution is also at hand: we can assume that the information structure constellation needed to override the MLC in (71a’, b’, c’) cannot be linked to the construction in question in English (because of the structural alternatives passive and dative shift). Other constellations do not yield a crossing effect. English has no special way of expressing information structure interactions of objects and adverbs and adverbial PPs. There being no restrictions on the distribution of topicality, the information structure needed to override the MLC in either (72a) or (72a’) can linked easily with to the sentences, so that both ways of formulating the multiple question are wellformed. (72) Constellations without crossing effects a. what did you see where? a’. where did you see what? b. to whom did you give what? b’. what did you give to whom? The MLC and derivational economy 115 The absence of a contrast between (72b) and (72b’) forces upon us the assumption that the construction V NP [to NP] comes in two varieties: to may be a dative marker, or the head of a PP. If information structure restrictions favoring the dative alternation affect the former version only, the absence of a contrast is predicted. Alternatively, we may assume that wh-PPs may always cross wh-DPs. For English, the approach just sketched implies that the topical nature of in situ subjects in multiple questions must be the blocking factor for sentences with wh-pronouns in subject position. (73) a. who arrived when? b. *when did who arrive 5. Concluding Remarks In the theory defended here, the MLC is a constraint that applies cyclically in a derivation: if more that one category can be attracted to a certain position P, only the one closest to P can move. However, the MLC cannot prevent a movement operation from applying if that movement step is inevitable in generating the (partial) LF-representation in question. Given that considerations of information structure play a role in this context, the fact that the MLC decides between syntactic objects with the same partial LF only renders the principle quite weak in the domain of operator movement. The predictions are quite different for head movement, if head movement does not have semantic effects. Consequently, the two syntactic objects in (74) (with A and B being heads attracted to X) do not yield different partial LFs, because they differ in the location of the phonetic matrix of A and B only. In the model advocated here, this is equivalent to saying that nothing will prevent the MLC from blocking (74b). (74) a. [[X A ] [ … A … [… B … ]]] b. [[X B ] [ … A … [… B … ]]] Phrasal A-movement has semantic consequences in many theories, and the pragmatic implications of different options of filling the subject position are obvious. The current proposal therefore implies that MLC-effects should be influenced by considerations of interpretation in the domain of A-movement as well, i.e., one should be able to observe apparent MLC-violations. This prediction is borne out. E.g., Hestvik (1986) observes that both objects can 116 Gisbert Fanselow be attracted to the subject position in the passive version of double object constructions in Norwegian: (75) Passive formation in Norwegian double object constructions a. det there ble was gitt given ham him en a gave present b. han he ble was gitt given en a gave present c. en gave ble gitt ham The standard assumption concerning English is that the direct object must not cross the indirect one in the passive of a double object construction, but this does not characterize all dialects of the language. After all, McCawley (1988: 79) observes that (76) sounds acceptable to speakers of British English. (76) a car was sold my brother __ for $200 by Honest Oscar Phrasal A-movement thus seems to have properties comparable to the one of operator movement with respecr to the MLC. One needs to identify the interpretive conditions that license (75c) or (76), and offer an account as to why information structure does not seem to modulate MLC-effects in certain languages or dialects of languages (such as American English). German shows that additional formal aspects comes into play that do not figure in operator movement: both objects may be promoted to subject status in a passive construction, but different auxiliaries are used for the promotion of direct and indirect objects: (77) Passive formation in German double object constructions a. jemand someone.nom b. ein a.nom stiehlt steals Schlüssel key dem the.dat wird is Kind child dem the.dat c. das Kind bekam einen the.nom child got a.acc “someone stole a key from the child” einen a.acc Kind child Schlüssel key gestohlen stolen Schlüssel key gestohlen stolen The MLC and derivational economy 117 Similarly, noun phrases with an oblique Case must not move to the subject position in many languages, and they may be skipped by A-movement to Spec,IP (see, e.g., Stepanov, this volume). There is no comparable array of facts with A-bar movement. The data show that the application of the MLC is not only sensitive to questions of identity of (partial) Logical Forms, but also constrainted by purely formal factors. A discussion of these is beyond the scope of the present paper. I have argued that the MLC must be considered an economy constraint that compares (partial) derivations and selects the one that fulfils checking requirements with the shortest movements possible. However, the set of candidate derivations which the MLC compares is constrained by formally encoded expressivity conditions: a derivational step B leading from structure S* to a partial LF S is blocked by the MLC only if S can also be reached from S* in a way that respects the MLC. Acknowledgments The research reported here was supported by grants of the Deutsche Forschungsgemeinschaft to the Forschergruppe Konfligierende Regeln (FOR 375), and to the Innovationskolleg Formale Modelle kognitiver Komplexität (INK 12). I want to thank Joanna Błaszczak, Eva Engels, Susann Fischer, Stefan Frisch, Hans-Martin Gärtner, Andreas Haida, Hubert Haider, Gereon Müller, Doug Saddy, Matthias Schlesewsky, Penka Stateva, Arthur Stepanov, Koyka Stoyanova, Ralf Vogel, and the two anonymous referees for helpful comments. 118 Gisbert Fanselow Notes 1. Superiority can be violated in similar contexts in Swedish, too. (ii) was accepted by two of my five informants, two rejected it, one found it questionable. All five informants considered (i) grammatical. (i) Vem Who tror believes att that Johan John (ii) Vad tror vem att Johan gjorde gjorde did vad what 2. Contrast between structures rated as “?” with others rated as “*” may not be too impressive, but examples involving different kinds of A-bar-movement yield clearer contrasts: (i) which violin-1 is this sonata-2 easy to play t-2 on t-1 (ii) *which sonata is this violin easy to play on 3. And the model proposed below does so by linking the acceptability of a crossing constellation to the expression of a non-standard information structure. 4. As suggested by Hubert Haider, p.c. 5. The relevance of such examples has been brought to my attention by Gereon Müller. 6. One may wonder, why scrambling is able to create structures incompatible with a simple MLC. Fanselow (2001) suggests that this problem is part of an argument in favor of the base-generation of scrambling structures. 7. Tibetan shows at least some of the contrasts one is familiar with from English (Seele p.c, Chungda Haller, p.c), in spite of the fact that it is a free constituent order language. I have no explanation for this. (i) a. b. su ga re nyos pa red? who what bought *ga re su nyos pa red 8. One might claim that these languages nevertheless allow scrambling, but only as an intermediate step followed by further movements. It is difficult to assess, however, which data could possibly refute such an account. Its empirical force is thus limited, and we refrain from considering it. The MLC and derivational economy 119 References Aoun, Joseph, Norbert Hornstein, David Lightfoot and Amy Weinberg 1987 Two Types of Locality. Linguistic Inquiry 18: 537–577. Aoun, Joseph, and Audrey Li 1989 Scope and constituency. Linguistic Inquiry 20: 141–172. 1993 The Syntax of Scope. Cambridge, Mass: MIT Press, 2002 Essays on the Representational and Derivational Nature of Grammar. Cambridge, Mass.: MIT-Press. Baker, C. 1970 Notes on the description of English questions: the role of an abstract question morpheme. Foundations of Language 6: 197–219. Beck, Sigrid 1996 Quantified structures as barriers for LF movement. Natural Language Semantics 4, 1–56. Billings, Loren and Catherine Rudin 1996 Optimality and Superiority: A new approach to overt multiple-wh ordering. In; Proceedings of Annual Workshop on Formal Approaches to Slavic Linguistics. The College Park Meeting 1994, Jindrich Toman (ed), 35–60. Michigan Slavic Publications. Ann Arbor. Błaszczak, Joanna and Susann Fischer 2002 Multiple Wh-Konstruktionen im Slavischen. Linguistics in Potsdam 14. Bobaljik, Jonathan and Diane Jonas 1996 Subject Positions and the Role of TP’. Linguistic Inquiry 27: 195–236. Bolinger, Dwight 1978 Asking more than one thing at a time. In; Questions, Henry Hiz (ed.), 107–150. Dordrecht: Reidel. Bonet, Eulalia 1993 Morphology after syntax: Pronominal clitics in Romance. Doctoral dissertation, MIT, Cambridge, Mass. Bo‰koviç, Îeljko 1997 Superiority effects with multiple wh-fronting in Serbo-Croatian. Lingua 102: 1–20. 2002 On multiple wh fronting. Linguistic Inquiry 33: 351–383. Bresnan, Joan 1972 Theory of Complementation in English Syntax. Doctoral dissertation, MIT, Cambridge, Mass. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris. 1993 A Minimalist Program for Linguistic Theory. In; The View from Building 20, Ken Hale and Samuel Keyser (eds), 1–58. Cambridge: MIT Press. 1995 The minimalist program. Cambridge, Mass.: MIT-Press. 120 Gisbert Fanselow Comorovski, I. 1996 Interrogative Phrases and the Syntax-Semantics Interface, Dordrecht: Kluwer Academic Publishers. Dayal, Veneeta 2003 Multiple wh-questions. Case 66, The Syntax Companion. Diesing, Molly 1992 Indefinites. Cambridge, Mass.: MIT-Press. Erteschik-Shir, Nomi 1997 The dynamics of focus structure. Cambridge, Cambridge UP. Fanselow, Gisbert 1998 Minimal Link Effects in German (and Other Languages). Paper, presented at the 1998 MLC conference, Potsdam. 2001 Features, θ -roles, and free constituent order. Linguistic Inquiry 32, 3. Fanselow, Gisbert and Anoop Mahajan 2000 Towards a minimalist theory of wh-expletives, wh-copying, and successive cyclicity. In: Wh-scope marking, Uli Lutz, Gereon Müller and Arnim von Stechow (eds). Amsterdam: Benjamins. Featherston, Sam 2002a Grammaticality and Universals. Wh-constraints in German. to appear in Linguistics. 2002b Magnitude estimation and what it can do for your syntax.: some whconstraints in German. Ms., Tübingen. Fodor, Janet Dean 1978 Parsing strategies and constraints on transformations. Linguistic Inquiry 9: 427–473. Frazier, Lyn and Giovanni Flores d’Arcais 1989 Filler-driven parsing: A study of gap filling in Dutch. Journal of Memory and Language 28: 331–344. Golan, Yael 1993 Node crossing economy, superiority and D-linking. Ms., Tel Aviv University. Grewendorf, Günther 1999 The additional-wh effect and multiple wh-fronting. In: Specifiers, D. Adger, S. Pintzuk, B. Plunkett and G. Tsoulas (eds.), 146–162. Oxford: Oxford University Press, 2001 Multiple wh-movement. Linguistic Inquiry 32: 87–122. Haider, Hubert 1986 Deutsche Syntax – Generativ. Habilitation thesis. Vienna. 1990 Topicalization and other puzzles of German syntax. In: Scrambling and Barriers. Günter Grewendorf and Wolfgang Sternefeld (eds.), 93–112. Amsterdam: Benjamins. 1993 Deutsche Syntax-generativ. Tübingen: Narr. 1997 Economy in syntax is projective economy. In; The Role of Economy Principles in Linguistic Theory. Chris Wilder, Hans-Martin Gärtner and Manfred Bierwisch (eds.), 205–226. Berlin: Akademie Verlag. The MLC and derivational economy 2000 121 Superiority Revisited – Dutch, English, German, Icelandic Contrasts. A representational account. Ms., University of Salzburg. This vol. The Superiority Conspiracy – Four Constraints and a Processing Effect. Halle, Morris 1992 Latvian declension. In Yearbook of Morphology Geert Booij and Jaap van der Marle (eds.), 33–47. Kluwer, Dordrecht. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In The View from Building 20. Ken Hale and S. Jay Keyser (eds.), 111–176. MIT Press, Cambridge. 1994 Some key features of Distributed Morphology. In MITWPL 21: Papers on Phonology and Morphology. Andrew Carnie & Heidi Harley (eds.), 275–288. MIT Press, Cambridge. Hemforth, Barbara 1993 Kognitives Parsing: Repräsentation und Verarbeitung sprachlichen Wissens. Sankt Augustin: Infix. Hendrik, R. & Michael Rochemont 1982 Complementation, multiple wh, and echo questions. Ms., University of North Carolina, Chapel Hill, N.C. and University of California, Irvine, California. Hestvik. Arild 1986 Case Theory and Norwegian Impersonal Constructions: SubjektObject Alternations in Active and Passive Verbs. Nordic Journal of Linguistics 9: 181–197. Hoge, Kerstin 2000 Superiority. Doctoral dissertation. Oxford. Hornstein, Norbert 1995 Logical Form. Cambridge, Mass.: Blackwell. Huang, James T. 1982 Logical relations in Chinese and the theory of grammar. Doctoral dissertation, MIT, Cambridge, Mass. Kayne, Richard 1983 Connectedness. Linguistic Inquiry 14: 223–249. Kitahara, Hisatsugu 1993 Deducing ‘superiority’ effects from the Shortest Chain Requirement. Harvard Working Papers in Linguistics 3: 109–119. Kitahara, Hisatsugu 1994 Target α. Doctoral dissertation, Harvard University. Krems, Josef 1984 Erwartungsgeleitete Sprachverarbeitung. Frankfurt/Main: Lang. Kuno, Susumo and J. Robinson 1972 Multiple wh-questions. Linguistic Inquiry 3: 463–487. Lasnik, Howard and Mamoru Saito 1992 Move a: Conditions on its Application and Output, Cambridge, Mass.: MIT Press. 122 Gisbert Fanselow Lee, Hanjung This vol. Minimality in a Lexicalist OT. Lutz, Uli, Gereon Müller and Arnim von Stechow (eds.) 2000 Wh-scope marking. Amsterdam: Benjamins. Mahajan, Anoop 1990 The A/A-bar distinctin and movement theory. Doctoral dissertation, MIT, Cambridge, Mass. Maling, Joan and Annie Zaenen 1982 A phrase structure account of Scandinavian extraction phenomena. In: The Nature of Syntactic Representations, Pauline Jacobson & Geoffrey Pullum. (eds), 229–282. Dordrecht: Reidel. Mathieu, Eric 2002 The Syntax of Non-Canonical Quantification: A Comparative Study. Doctoral dissertation. London. McCawley, James 1988 The Syntactic Phenomena of English. Chicago: The University of Chicago Press. Meinunger, André 1995 Discourse dependent DP de-)placement. Doctoral dissertation, University of Potsdam. Müller, Gereon 2001 Order preservation, parallel movement, and the emergence of the unmarked. In: Optimality Theoretic Syntax, Geraldine Legendre, Jane Grimshaw, and Sten Vikner (eds), 279–313. Cambridge, Mass.: MITPress. Noonan, Maire 1988 Superiority Effects : How do antecedent government, lexical government and V2 interact. McGill Working Papers in Linguistics 1988: 192–214. Noyer, Robert 1992 Features, Positions and Affixes in Autonomous Morphological Structure. Doctoral dissertation, MIT, Cambridge, Mass. Oka, T. 1993 Shallowness. MIT Working Papers in Linguistics 19: 255–320. Pesetsky, David 1982 Paths and categories. Doctoral dissertation, MIT, Cambridge, Mass. 1987 Wh-in-situ: movement and unselective binding. In: The Representation of (In)definiteness, Eric Reuland and Alice. ter Meulen (eds), 98–129. Cambridge, Mass.: MIT Press. 2000 Phrasal Movement and Its Kin. Cambridge, Mass.: MIT-Press. Reinhart, Tanya 1995 Interface Strategies. OTS working papers in Linguistics. 1998 Wh-in-situ in the Framework of the Minimalist Program. Natural Language Semantics 6: 29–56. The MLC and derivational economy 123 Reis, Marga 1996 Extractions from Verb-Second Clauses in German?. In: On Extraction and Extraposition in German, Uli Lutz and Jürgen Pafel (eds.), 45–88. Amsterdam: Benjamins. 1997 Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In: Sprache im Fokus, Christa Dürscheid, Karl Heinz Ramers, and Monika Schwarz (eds.), 121–144. Tübingen: Niemeyer. Richards, Norvin 2001 Movement in Language. Oxford & New York: Oxford University Press. Rizzi, Luigi 1990 Relativized Minimality. Cambridge, Mass.: MIT Press. Rudin, Catherine 1988 On multiple questions and multiple wh fronting. Natural Language and Linguistic Theory 6: 445–501. Steinitz, Renate 1969 Adverbialsyntax. Berlin: Akademie-Verlag, Stepanov, Arthur 1998 On Wh-Fronting in Russian – Proceedings of NELS 28. Pius N. Tamanji & Kiyomi Kusumoto (eds), 453–467. 2001 Cyclic domains in syntactic theory. Doctoral dissertation, University of Connecticut, Storrs. This vol. Ergativity, Case, and the Minimal Link Condition. Sternefeld, Wolfgang 1997 Comparing Reference Sets. In: The Role of Economy Principles in Linguistic Theory, Chris Wilder, Hans.-Martin Gärtner and Bierwisch (eds.), 81–114. Berlin: Akademie Verlag. Travis, Lisa 1984 Parameters and Effects of Word Order Variation. Doctoral dissertation, MIT, Cambridge, Mass. Vogel, Ralf This vol. Correspondence in OT syntax and minimal link effects. Wiltschko, Martina 1998 Superiority in German. In: Proceedings of the Sixteenth West Coast Conference on Formal Linguistics, E. Curtis, J. Lyle & G. Webster (eds.). 431–445. Stanford, Cal.: CSLI Publications. Zubizaretta, Maria Luisa 1998 Prosody, Focus and Word Order. Cambridge, Mass.: MIT Press.
© Copyright 2026 Paperzz