Czech in Generative Grammar Editors: Mojmír Dočekal Petr Karlík Jana Zmrzlíkova Contents Petr Biskup: Sentence-final sentence adverbs in the phase model 7 Pavel Caha: A Note About A Note About Nothing 17 Markéta Ceplová: Infinitives under 'have'/'be' in Czech 31 Mojmír Dočekal: Only, bound variables and VP ellipsis in Czech 47 Jakub Dotlačil: Clitic omission in Czech as across-the-Board extraction 65 Joseph Emonds: Czech Cases and the Syntacticon: Poznámky k, o, okolo, nad něčím a pro někoho 81 Petr Karlík: Mixed Nominals in Czech 105 Lucie Medová and Tarald Taraldsen: 1, 2, se 119 Radek Šimík: The Czech invariant demonstrative to is a Foe head 139 Hana Skrabalova: Wh-questions with conjoined wh-words 161 Andrea Volencová: Reflexive verbal forms in Czech from the Romance perspective 175 Markéta Ziková: Why Czech case markers sometimes get lost 193 3 Preface This festschrift is the collective work of people who believe that transformational generative grammar is empirically the strongest linguistic theory, as it offers a restrictive and yet sufficiently robust apparatus for the description of any natural language, while offering mechanisms for evaluating each newly arising hypothesis. All the authors are united as well either by their knowledge of the Czech language, or by their use of Czech as a data source for more general understanding. As far as we know, ours is the first work with such an ambition. But more important, all the authors are united in their wish to give pleasure to the First Lady of Czech linguistics, who introduced modern generative grammar to the Czech Republic, and who taught and still teaches generative grammar to many of us directly or indirectly, our collaborator in the past and the present: Lida Veselovska. Without Lida generative grammar would not have arisen in the Czech Republic, just as it would not have existed in the world without Chomsky's book which gave it birth fifty years ago. 5 Sentence-final sentence adverbs in the phase model Petr Biskup biskup@rz.uni-leipzig.de 1 Introduction It has been argued that sentence adverbs cannot occur in the sentence-final position, unless they are separated by a comma intonation. See, for example, Jackendoff (1972) for English (1), Belletti (1990) for Italian (2) and French (3), and Alexiadou (1997) for Greek (4). (1) *Horatio has lost his mind evidently/probably. (Jackendoff 1972, 50 (3.9)) (2) *Gianni partirá probabilmente. 'Gianni will leave probably' (Belletti 1990, 53 (53)) (3) *Jean partira probablement. 'Jean will leave probably.' (Belletti 1990, 53 (53)) (4) *0 Janis tha figi pithanos. The JohnNOM FUTg03SG probably (Alexiadou 1997, 157 (80)) The following examples show that the same holds for Czech; sentence adverbs such as the epistemic možná 'possibly' or pravděpodobně 'probably' are not allowed in the sentence-final position. (5) *Tu knihu pošle Jirkovi Pavel pravděpodobně. the bookAcc sends JirkaDAT PavelNOM probably 'The book, Pavel will probably send to Jirka.' (6) *Tu knihu pošle Jirkovi Pavel možná. the bookAcc sends JirkaDAT PaveliroM possibly 'The book, Pavel will possibly send to Jirka.' This seems to be in accordance with the claim that sentence adverbs occupy a pretty high position in the clausal structure (Cinque 1999, Laenzlinger 2002). There is also an alternative explanation why the adverbs in (l)-(6) are impossible in the sentence-final position. According to Lang (1979), sentence adverbs are focus sensitive operators and they themselves cannot be focalized. Compare also Hajičová, Partee &; Sgall (1998a,b) or Koktova (1987, 1999), who 7 argue that the prototypical position of focus sensitive adverbs is at the boundary between the background and the focus, and Krifka (1992) or Jacobs (1986, 1988) in Czech who argue that focusing adverbs must c-command their focus. However, there are sentence adverbs that can occur in the sentence-final position in Czech and Italian; consider examples (7) and (8). (7) Tu knihu pošle Jirkovi Pavel určitě. the bookAcc sends JirkaDAT PaveliroM certainly 'The book, Pavel will certainly send to Jirka.' (8) Gianni lo merita sicuramente / di sicuro. 'Gianni deserves it surely.' (Cinque 1999, 180 note 80)) These examples pose a problem for the argument that the source of the ungrammaticality of sentence adverbs in the sentence-final position is the height of their structural position and they show that the ungrammaticality of (l)-(6) is not due to the general impossibility of sentence adverbs to appear in the sentence-final position and be accented there. Therefore Cinque (1999, 180 note 80) proposes that adverbs like sicuramente belong to the class of (realis) mood adverbs that can be used as focusing adverbs - they are heads taking their modifees as complements - and allow their complements to move across them. On the other hand, Lang (1979) distinguishes three classes of German sentence adverbs. Class A adverbs are, for example, wahrscheinlich 'probably' or möglicherweise 'possibly'; class B adverbs bedauerlicherweise 'unfortunately' or überraschenderweise 'surprisingly'; and class C adverbs tatsächlich 'really' or in der Tat 'certainly'. According to him, class A adverbs, that is the adverbs in (5) and (6), differ from class C adverbs, that is, the adverbs in (7) or (8), in that they cannot be accented and focalized. However, both proposals have a problem with cases like (9) because the sentence-final adverb možná 'possibly' is epistemic (that is, Lang's class A), is accented and represents the focus itself; it is associated with the focus-sensitive adverbial jenom 'only'. Since non-clausal non-selected adverbials are merged to the left in Czech (see Biskup in prep.), the grammaticality of (9) cannot be accounted for through right adjunction of možná. (9) Tu knihu pošle Jirkovi Pavel jenom možná. the bookAcc sends JirkaDAT PaveliTOM only possibly 'It is only possible that Pavel will send the book to Jirka.' The remainder of this paper is organized as follows. In section 2,1 will discuss veridicality and a downward-monotonicity approach to sentence adverbs and on the basis of empirical arguments I will argue that these approaches are not appropriate. In section 3, I will argue that sentence adverbs generally can be merged in the vP phase and that the (un)grammaticality of certain sentence adverbs in the sentence-final position depends on the (non-)interpretability of a given adverb in the vP position and not on its syntactic position. I will propose an analysis in a model with a correlation between the phase structure, tripartite quantificational structure and information structure and will argue that sentence adverbs can appear in the sentence-final position only if they represent the extreme value with respect to the set of focus alternatives. Conclusions will be drawn in section 4. 8 2 Other proposals: veridicality and downward-monotonicity Let us begin with the difference between the adverbs určitě 'certainly' and možná 'possibly'. It is easy to show that these adverbs have different lexicosemantic properties. The epistemic adverb možná is excluded from contexts licensing negative polarity items. More specifically, it is degraded, for example, under the question operator (10) or the imperative operator (11). (10) Pošle Pavel (*možná) Jirkovi tu knihu (*možná)? sends PavelNOM possibly JirkaoAT the bookAcc possibly 'Will Pavel possibly send the book to Jirka?' (11) Pošli Jirkovi (*možná) tu knihu (*možná)! send JirkaoAT possibly the bookAcc possibly 'Possibly send the book to Jirka.' In contrast, the (realis) mood adverb určitě can appear in these environments, as demonstrated by question (12) and the imperative sentence in (13). (12) Pošle Pavel (určitě) Jirkovi tu knihu (určitě)? sends PavelNOM certainly JirkaoAT the bookAcc certainly 'Will Pavel certainly send the book to Jirka?' (13) Pošli Jirkovi (určitě) tu knihu (určitě)! send JirkaoAT certainly the bookAcc certainly 'Certainly send the book to Jirka.' In the recent literature, there are two interesting approaches to this issue. First, let us look at the veridicality approach. 2.1 Veridicality It has been argued that questions and imperatives are nonveridical environments, see, for example, Giannakidou (1999, 2002). Giannakidou (2002, 5) defines (non)veridicality for prepositional operators as follows: (Non)veridicality for propositional operators (i) A propositional operator F is veridical iff Fp entails p: Fp —► p; otherwise F is nonveridical. (ii) A nonveridical operator F is antiveridical iff Fp entails not p: Fp —► ->p. Given the different behavior of the adverbs in (10)-(13), one might suggest that the reason why (5) and (6) are ungrammatical is that the appropriate sentence adverbs are in the scope of a nonveridical operator. Since there is no other overt operator in the sentence, it could be something like Jacobs's (1988) assertion operator (ASSERT). It is feasible to suggest that ASSERT(p) does not entail p. A potential problem is that one cannot assume that the truth of a sentence is always evaluated with respect to a certain epistemic model because in a speaker's belief model, the asserted proposition can be true (see discussion in Giannakidou 1999). However, there is also an empirical argument against this analysis. When the sentence adverb možná occurs in the background domain of a declarative sentence (as evidenced by the congruent context (14a)), the sentence is grammatical; compare (14b) with example (6). 9 (14) a. Komu pošle Pavel tu knihu? (To whom will Pavel send the book?) b. Tu knihu pošle možná Pavel Jirkovi. the bookAcc sends possibly PavelNOM JirkaDAT 'The book, Pavel will possibly send to Jirka.' Given the grammaticality of (14b) and the fact that the assertion operator has in its scope the whole sentence, one needs to use a tripartite structure for the operator. This is in line with Jacobs (1988), according to whom all sentences have a focus-sensitive element (either overt or covert) and all illocutionary operators can be focus sensitive and can participate in a tripartite structure. Thus, if one wants to keep the idea that it is the nonveridicality of the assertion operator to which the sentence adverb is sensitive, the assertion operator would have to be veridical in its restrictor (the background domain of the sentence) and nonveridical in its nuclear scope (the focus domain). However, there is a reason to think that nonveridicality of the focalized position is not the right issue here. According to Zwart (1995), the dyadic operator or (both exlusive and nonexclusive) is nonveridical in both conjuncts. (Non)veridicality for dyadic operators is defined as follows (Zwart 1995, 288): (Non)veridicality for dyadic operators Let C be a dyadic truth-functional connective. C is said to be veridical with respect to p [q] just in case pCq =>- p [pCq=^ q] is logically valid. If C is not veridical with respect to p [q], then C is nonveridical with respect to p W- This means that the sentence adverb možná should be bad in this environment. This is indeed so, as shown by the following example. (15) ??Buď sousedka možná vaří, nebo soused lakuje auto. Either neighborpEM possibly cooks or neighborMAsc paints car 'Either the neighbor possibly is cooking or her husband is painting his car.' So far so good. However, according to Czech speakers, there is a clear contrast in judgments of (15) with možná in the background position and (16) with možná in the focus position. This unexpected since the focus position of the adverb should also be excluded due to nonveridicality. (16) *Buď sousedka vaří možná, nebo soused lakuje auto. Either neighborFEM cooks possibly or neighborMAsc paints car 'Either the neighbor possibly is cooking or her husband is painting his car.' Since the explanation in terms of nonveridicality of the focalized position does not work, let us turn to the down ward-monotonicity approach. 2.2 Downward-monotonicity Nilsen (2003) argues that sentence adverbs like možná are positive polarity items and as such are excluded from environments licensing negative polarity items. More specifically, he argues that they are excluded from downward-entailing environments. Downward-entailing operators reverse the direction of entailment. This means that in downward-entailing environments truth is preserved when the predicate is replaced by a stronger (subset) predicate. See the definition below, taken from Nilsen (2003, 41). 10 DE function A function / is downward-entailing iff whenever a is semantically stronger than b, it holds that f(b) is semantically stronger than f(a). Thus, for example, (17a) entails (17b); the replacement of knihu with the stronger predicate zajímavou knihu under negation preserves the truth. (17) a. Pavel nepošle Jirkovi knihu. PavelNOM NEcsends JirkaDAT bookAcc 'Pavel will not send a book to Jirka.' b. Pavel nepošle Jirkovi zajímavou knihu. PavelNOM NEGsends JirkaDAT interesting bookAcc 'Pavel will not send an interesting book to Jirka.' Again, given the grammaticality of sentence (14b) with možná in the background domain, the covert assertion operator should be downward-entailing in its nuclear scope (the focus of the sentence) but not in its restrictor (the background domain of the sentence). This does not go through because example (18) - with the appropriate context (18a) - shows that the focus of the assertion operator is upward-entailing; sentence (16b) entails (16c). (18) a. Co pošle Pavel Jirkovi? 'What will Pavel send to Jirka?' b. Pavel pošle Jirkovi zajímavou knihu. PavelNOM sends JirkaDAT interesting bookAcc 'Pavel will send an interesting book to Jirka.' c. Pavel pošle Jirkovi knihu. PavelNOM sends JirkaDAT bookAcc 'Pavel will send a book to Jirka.' Thus, the conclusion drawn from this section is that the explanation in terms of downward monotonicity does not work either. If it is not nonveridicality or downward-monotonicity that makes the sentence adverbs like možná bad in the sentence-final position in declarative sentences, it is necessary to find an alternative analysis. 3 The analysis According to Chomsky (2001), every phase has its own subarray that chooses its elements from the numeration. Therefore it seems natural that one and the same adverb can appear in both the vP subarray and the CP subarray, and can be merged in both the vP phase and the CP phase. In Biskup (2006), I show that phrases moved to the CP phase - that is, scrambled or topicalized - are backgrounded and get a specific interpretation, as illustrated in (19). The specific interpretation can be epistemic, partitive, or generic. I argue that this is driven by the grammar requirement that backgrounded specific elements are to be linearized and interpreted in the CP phase (the left part of sentences) in scrambling languages like Czech. 11 CP Building on Partee (1992), Diesing (1992), and Chomsky (2000, 2001, 2005, 2006), I propose that there is a correlation between the phase structure, tripartite quantificational structure and information structure of the sentence. This is illustrated in (20). This means that at the semantic interface, the vP phase (the elements in the phase) is interpreted as the nuclear scope of the quantificational structure and the information focus. The CP phase is interpreted as the restrictive clause and the domain of background. CP CP phase vP phase\^ restrictive nuclear scope ^^^^ clause, background information focus Then, without going into details, examples (6), (7) and (9) look like (21), (22), and (23). (21) *[Cp [xp Tu knihu pošle [tp Jirkovi Pavel [vp možná ]]]]. the bookAcc sends JirkaDAT PaveliTOM possibly 'The book, Pavel will possibly send to Jirka' (22) [Cp [xp Tu knihu pošle [tp Jirkovi Pavel [vp určitě ]]]]. the bookAcc sends JirkaDAT PavelNOM certainly 'The book, Pavel will certainly send to Jirka.' (23) [cp [xp Tu knihu pošle [tp Jirkovi Pavel [vp jenom možná ]]]]. the bookAcc sends JirkaDAT PaveliTOM only possibly 'It is only possible that the book Pavel will send to Jirka.' I make the standard assumption that every sentence has a focus. Then, given the proposed model, sentence adverbs in the sentence-final position in vP are necessarily interpreted as focalized at the semantic interface and introduce a set of alternatives. To account for the behavior of sentence adverbs in the sentence-final position and the difference in behavior between možná and určité, I will make use of Krifka's (1995) notion of 'extreme value'.1 Krifka argues that polarity items as well as focalized elements introduce alternatives and that the alternatives are ordered according to semantic strength (set relations). And that the set of alternatives can be represented by values on a scale. Here, I propose that a sentence adverb can appear in the sentence-final position in the vP phase and be focalized there only if it represents the extreme value with respect to the set of 1It has been suggested by Piňón (2006) that Krifka's approach (1995) might be used for the analysis of sentence adverbs. By 'extreme value', I mean the extremity with respect to particular focus alternatives, not with respect to all the alternatives together, for details, see Krifka (1995). 1 12 focus alternatives. Epistemic adverbials can be taken to correspond to certain values on an epistemic scale. It is a lexicosemantic property of the (realis) mood adverbial určitě that it corresponds to the highest value on the epistemic scale. Thus, it represents the extreme value with respect to the focus alternatives because all other alternatives are, of course, lower on the epistemic scale. Therefore určitě can occur in the focalized sentence-final position and in this way sentences with (realis) mood adverbials like určitě get the verum-focus interpretation; see (7)=(22) again. In contrast, možná, given its lexicosemantic properties, does not represent an extreme value on the epistemic scale, therefore it cannot serve as the asserted focus alternative. Hence, if možná appears in the sentence-final position in vP, the sentence becomes ungrammatical, as demonstrated by (6)=(21). However, this changes when the exclusive particle jenom 'only' is added. I demonstrated by example (9)=(23) that the sentence adverbial možná can in fact occur in the sentence-final position. It is well-known that the exclusive adverb only is focus sensitive, see, for example, Rooth (1985), Beaver & Clark (2003) or Krifka (2006). In sentence (9)=(23), jenom associates with the focalized sentence adverbial možná and excludes all other alternatives. The sentence then gets the interpretation according to which it is only possible - not, for example, probable or certain, which are possible alternatives to možná here - that Pavel will send the book to Jirka. Thus, jenom makes from možná the lowest alternative on the epistemic scale, which means that this alternative has the required extreme value; but in this case, it is extreme low probability. So, sentence (9)=(23) shows that sentence adverbs other than (realis) mood adverbs can also merge within vP and appear in the sentence-final position, and that the appropriate adverbs do not have to be used as focusing adverbs in that position. Similarly, the epistemic adverbial pravděpodobně can also occur in the sentence-final position if it represents the extreme value in the set of focus alternatives. This is shown in example (24). (24) a. *[cp [xp Pavel přijde [vp pravděpodobně ]]]. PavelNOM comes probably 'Pavel will probably come.' b. [cp [xp Pavel přijde [vp nanejvýš pravděpodobně ]]]. PavelNOM comes most highly probably 'It is at most probable that Pavel will come." 'It is highly probable that Pavel will come.' Sentence (24a) demonstrates that under usual circumstances, pravděpodobně is ungrammatical in the sentence-final position. As shown by (24b), after adding of the adverb nanejvýš 'most highly', the sentence becomes grammatical. This case is especially interesting because (24b) allows the extreme value to appear on either end of the epistemic scale. The first interpretation is the extremely-low-value interpretation, according to which there is no focus alternative lower than the pravděpodobně value on the epistemic scale. In this case, sentence (24b) can be continued, for example, by the following sentence: Určitě ne najisto 'Certainly not beyond doubt'. According to the second interpretation, the probability of Pavel's coming is highest, it is in fact certain that Pavel will come. Hence, in this case, the asserted alternative represents the extremely high value on the epistemic scale. As in the case of sentence (9)=(23), example (24b) demonstrates that although sentence adverbs like pravděpodobně cannot occur in the sentence-final position by themselves, they can occur there and be interpreted if they get help through another element. Hence, the problem does not lie in their low syntactic position, counter to 13 the accounts mentioned in the introduction. In certain contexts, pravdepodobne associated with the focus particle dokonce 'even' can also represent the extreme value. Consider example (25). (25) [cp [xp Pavel pfijde [vp dokonce pravdepodobne ]]]. PavelNOM comes even probably 'It is even probable that Pavel will come.' Imagine a situation where somebody is asking a few friends whether they will come to his party. All guys before Pavel say that they possibly will come. In this context, the value of pravdepodobne is extreme because it is unexpectedly high on the probability scale with respect to the other alternatives. 4 Summary In this paper, I have argued that sentence adverbs can be merged in the vP phase, that they can occur in the sentence-final position, be accented there and focalized. Depending on their lexicosemantic properties and on the properties of the appropriate sentence, either they can be interpreted in the vP position or they cannot. This means that the (un)grammaticality of the appropriate sentence depends on the (non-)interpretability of the adverb in the vP position and not on the syntactic position of the adverb. I have argued that the ungrammaticality of certain sentence adverbs in the sentence-final position in the vP phase is due to the fact that they cannot serve as the asserted alternative in the focus in the appropriate sentence. Sentence adverbs can appear in the sentence-final position only if they represent the extreme value with respect to the set of focus alternatives. References Alexiadou, Artemis. 1997. Adverb placement: A case study in antisymmetric syntax. Amsterdam: John Benjamins. Beaver, David and Brady Clark. 2003. Always and Only: Why not all Focus Sensitive Operators are Alike. Natural Language Semantics 11:323-362. Belletti, Adriana. 1990. Generalized Verb Movement. Aspects of Verb Syntax. Turin: Rosenberg & Sellier. Biskup, Petr. 2006. Scrambling in Czech: syntax, semantics and information structure. Seok Koon, Chin &; Fujimori Atsushi, (eds.). Proceedings of the 21st North West Linguistics Conference. UBC Occasional Papers of Linguistics 1:1-15. —. in prep. The Phase Model and Adverbials. PhD. dissertation. Universität Leipzig. Chomsky, Noam. 2000. Minimalist Inquiries: The Framework. Martin, R., D. Michaels &; J. Uriagereka (eds.). Step by Step: Essays on minimalist syntax in honor of Howard Lasnik: 89-156. Cambridge, Mass.: MIT Press. —. 2001. Derivation by Phase. Kenstowicz, M. (ed.). Ken Hale: A Life in Language: 1-52. Cambridge, Mass.: MIT Press. -. 2005. On Phases. Ms. MIT. -. 2006. Approaching UG from below. Ms. MIT. 14 Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic perspective. Oxford: Oxford University Press. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Giannakidou, Anastasia. 1999. Affective dependencies. Linguistics and Philosophy 22:367-421. —. 2002. Licensing and Sensitivity in Polarity Items: from Downward Entailment to (Non)Veridicality. Chicago Linguistic Society 39:29-53. Hajicovä, Eva, Barbara Partee and Petr Sgall. 1998a. Focus, topic, and semantics. Benedicto, E., M. Romero & S. Tomioka (eds.). UMOP 21: Proceedings of workshop on focus: 101-124. University of Massachusetts, Amherst: GLSA. —. 1998b. Topic-Focus Articulation, tripartite structures, and semantic content. Dordrecht: Kluwer. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. Jacobs, Joachim. 1986. The Syntax of Focus and Adverbials in German. Abraham, W. & S. de Meij (eds.). Topic, Focus and Configurationality: 103-127. Amsterdam: Benjamins. —. 1988. Fokus-Hintergrung-Gliederung und Grammatik. Altmann, H. (ed.). Intonationsforschungen: 89-134. Tübingen: Narr. Koktovä, Eva. 1987. On the scoping properties of negation, focusing particles and sentence adverbials. Theoretical Linguistics 14:173-226. —. 1999. Word-order Based Grammar. Berlin: Mouton de Gruyter. Krifka, Manfred. 1992. A Compositional Semantics for Multiple Focus Construction. Jacobs, J. (ed.). Informationsstruktur und Grammatik: 17-53. Opladen: Westdeutscher Verlag. —. 1995. The Semantics and Pragmatics of Polarity Items. Linguistic Analysis 25:1-49. —. 2006. Association with focus phrases. Molnar, V. & S. Winkler (eds.). The Architecture of Focus: 105-136. Berlin, New York: Mouton de Gruyter. Laenzlinger, Christopher. 2002. A Feature-Based Theory of Adverb Syntax. GG@G 3:67-106. Lang, Ewald. 1979. Zum Status der Satzadverbiale. Slovo a slovesnost 40:200-213. Nilsen, 0ysten. 2003. Eliminating Positions. Ph.D. dissertation, Utrecht University. Partee, Barbara. 1992. Topic, focus, and quantification. Proceeding of the First Conference on Semantic and Linguistic Theory (SALT 1):159-187. Pinon, Christopher. 2006. Modal Adverbs. Gärtner, H.M., S. Beck et al. (eds.). 40-60 puzzles for Manfred Krifka. Berlin, http://www.zas.gwz-berlin.de/40-60-puzzles-for-krifka/ Rooth, Math. 1985. Association with Focus. Ph.D. dissertation. University of Massachusetts, Amherst. Zwart, Frans. 1995. Nonveridical Contexts. Linguistic Analysis 25 (3/4):286-312. 15 A Note About A Note About Nothing Pavel Caha Pavel.Caha@hum.uit.no 1 Introduction 1 start off by reporting some observations made by Lída in her paper A Note About Nothing (Veselovská, 2003) concerning Case-assignment asymmetries in Czech DPs with inanimate pronouns (like nic 'nothing'). These data show that there is a clear asymmetry between Structural (nominative, accusative) and Oblique Cases, which Lída accounts for in terms of S-structure vs. D-structure asymmetry. Since D-structure as an explanatory tool has been abandoned in recent minimalist theorizing (see McCawley, 1968 for an early proposal along these lines), I try to recast the distinction in different terms. Specifically, I look at pairs of Czech constructions that exhibit a systematic alternations between the instrumental (ins, an Oblique Case) and accusative (acc, a Structural Case), and I propose the following, (i) these constructions are transformationally related, (ii) the DPs are base-generated bearing ins, and (hi) they become acc by movement. I derive (ii) and (in) from the assumption that ins contains two case projections, Ki and K2, whereas acc only K2, reconstructing the S-structure vs. D-structure distinction in terms of the amount of functional structure. I further follow Starke (2005) by having the projection of K2 raise from within K^P. I suggest to extend this approach to Case marking in general, and I lay out a first step of the program: I propose a hierarchy of Czech Case. 2 Structural Cases vs. Oblique Cases Veselovská (2001) observes that DPs headed by inanimate pronouns behave differently in Structural and Oblique Case environments. Specifically, adjectives following the pronoun bear genitive when the pronoun is nominative or accusative, and thus their Case differs from that of the pronoun, as demonstrated in (1). However, they are marked instrumental when the pronouns appears in instrumental, thus agreeing with the Case of the pronoun, as can be observed in (2). (1) něco takového velkého nothingNOM/Acc sucIigen bigGEN (Veselovská, 2003:ex. 17a) (2) s ničím takovým velkým with nothingiNs suchiNs bigiNs (Veselovská, 2003:ex.20b) *I am grateful to the following people: Marcel den Dikken, Gillian Ramchand, Björn Lundquist, Lucie Medovä, Marina Pantcheva, Michal Starke, Peter Svenonius, Tarald Taraldsen. 17 The first step of the analysis that Veselovská (2003) puts forth in order to understand the data theoretically, is to ground the distinction between Structural and Oblique Cases in terms of Case Hierarchy, understood (along the lines of Franks (1995)) in terms of different timing of operations. Specifically, Structural Case is assumed to be assigned at S-structure, whereas Oblique Case is assigned at D-structure. (3) Case Hierarchy Lexical [+Oblique] Case is assigned at D-Structure Structural Case is assigned at S-Structure (Veselovská, 2003:ex.22) The problem I address here is what to do with the distinction of Structural vs. Oblique Case when the distinction between D-structure and S-structure is abandoned (a point argued for in McCawley, 1968, or much later in Chomsky, 1995). Apart from general considerations presented in the works cited, there is in fact empirical evidence against the implementation of Case Theory given in (3), which comes from the domain of Case itself. Specifically, the formulation in (3) stems from the conviction that while Structural Cases are purely structural and devoid of any special semantic contribution, Oblique Cases are "Lexical" in the sense that they satisfy both semantic and syntactic selectional requirements of items that they co-occur with. Here, I argue against both of these propositions. On one hand, I present (additional) evidence that Structural Cases are relevant for semantic information (and here I follow much of previous work). On the other hand, I try to show that there is a great deal of specifically syntactic factors that contribute to the distribution of Oblique Case.1 Hence, I try to re-establish the distinction in completely different terms: in terms of syntactic structure.2 I start by looking at Czech instrumental and accusative alternations and see what kind of theory is able to capture their systematic relation. 3 Instrumental and Accusative 3.1 Raising-to-Object Constructions Consider the pair (4) and (5). (4) Petr nechal [ televizi opravit odborníkem ] Peter let television repair^ professional's 'Peter had the TV repaired by a professional' (5) Petr nechal [ odborníka opravit televizi ] Peter let professionaUcc repair^ television 'Peter had the professional repair the TV The DP odborník 'professional' takes INS in (4), but ACC in (5). First thing to note is that the verb nechat 'let/have' is a restructuring verb (for restructuring in Czech see Dotlačil, 2004), and takes a clause headed by the infinitive as its complement. 1 This conclusion in fact does not make sense if syntax and compositional semantics are identical, as suggested in some recent works (e.g. Ramchand, 2007). 2 For proposals along similar lines see e.g. Bayer et. al. (2001) and Asbury (2006), as well as references therein. 18 The infinitival clause is passive in (4), with the internal argument (ia) of the infinitive raised to the object of the matrix verb 'let', and the external argument EA (of the infinitive) expressed by INS (as is the standard for EAs in Czech passives). In (5), the infinitive is active, and it is the infinitive's EA that becomes the object of the matrix verb. That this analysis is on the right track, is confirmed by the (im)possibility of so called long passive (see e.g. Wurmbrand, 2001). Specifically, the reflexive passive of the matrix verb is able to promote the IA of the infinitive in the case of (4), but not in the case of (5), as shown in (6). (6) Televize se nechala opravit odbornik-em / *-a Television refl let repair^ professional's / *acc 'The TV was repaired by a professional' On standard interpretation, the EA blocks the long passive (by relativized minimality) in (5), but not in (4), suggesting that the IA must have moved across the EA in the course of deriving (4). The conclusion is the following: the EA and the IA are both generated in the infinitival clause, but only one of them can become the object of the matrix verb. If the IA manages to do so (as in (4)), the EA is marked INS. If the EA becomes the object (blocking the long passive), it is marked ACC. 3.2 Spray/Load Alternation The Czech spray/load alternation is presented in (7) and (8). (7) Naložil [ vůz senem ] l0aded3.pl truck hay ins 'They loaded the truck with hay.' (8) Naložil [ seno na vůz ] l0aded3.pl hayAcc on truck 'They loaded the hay on the truck.' In (7), the DP 'hay' is marked INS, but it shows ACC in (8). Again, the question arises how to capture the case marking of the DP in question. It is theoretically attractive to understand this pattern in terms of the preceding one. Specifically, the bracketed structure in (7) is to be understood as a passive version of the bracketed structure in (8), on analogy with (4) and (5). The same conclusion is reached by Romanova (2004, 2007) and Ramchand and Svenonius (2004). The works cited propose that the 'hay' and the 'truck' are to be understood as the EA and the IA, respectively, of a spatial relation denoted by the preposition 'on'. In (7), the IA of the preposition moves to the object position of the verb 'load', and the EA is marked by INS. In (8), it is the EA that moves to the object position of the matrix verb, and surfaces as ACC. The conclusion here can thus be very similar to the one arrived at in the first subsection: the EA and the IA of the P are both generated in a non-finite clause, but only one of them can become the object of the matrix verb 'load'. If the IA manages to do so (as in (7)), the EA is marked INS. If the EA becomes the object, it is marked ACC. 19 3.3 EAs of PPs So far, there is no clear indication of how to analyze the relation between INS and ACC, nor how to address the problem of case in these constructions in general. One option is to take INS for an Oblique Case that serves the purpose of marking demoted EAs (of both Vs and Ps), and ACC to be assigned by the matrix verb in a designated structural position, keeping to the version of Case Theory presented in (3). However, a clearer picture (as for which way not to go) emerges, if we take a closer look on the ACC/lNS alternation of the P's EA. (9) Petr bodnul nožem do chleba Peter stuck knifeiNs in bread 'Petr stuck the knife in the bread and pulled it out.' (10) Petr bodnul nůž do chleba Peter stuck knifeAcc in bread 'Petr stuck the knife in the bread and the knife stayed there.' As in the previous section, I consider the 'knife' in both examples to be the EA of the spatial relation denoted by the directional P do. The difference between the INS and ACC here does not have to do with active or passive, since the IA of the P stays the same in both examples and never moves across the EA 'knife'. Rather, the difference here is best characterized in semantic terms: either the event expressed by the V is followed by a state resulting from that event (10), or not (9). We can clearly see here that (contrary to expectations based on the version of Case Theory in (3)), it is a Structural Case that brings in a semantic load. One way how to understand the semantic (result state vs. pure process) and syntactic (ACC vs. ins) difference theoretically, is to have the resulting state (knife in the bread) represented and projected as a phrase in syntax (resP), along the lines of Ramchand (2007). Ramchand further proposes that the head of the resP is a complement of the process (proc) denoted by the verb, which again she takes to be represented in syntax by a projection she calls procP. In these terms, the constructions under consideration differ in whether or not the syntactic structure contains the resP. Specifically, the PP ([ 'knife' [ into 'bread' ]]) is a complement of proc in (9), but a complement of res in (10). (11) [proc [ea [P ia]]] no entailment of a state resulting from the process (12) [proc [res [ea [P ia]]]] entailment of a state resulting from the process If this is a correct account of the semantic difference, it is tempting to understand the case marking pattern in terms of the res head. A straightforward way how to relate it to the structures proposed, is to connect the ACC with the EA raised to Spec,resP, and to see the INS as arising if the EA stays in situ. The generalization, then, is that if the P's EA raises, it surfaces as ACC, whereas if it stays in situ, it bears INS. It is not difficult to see how the solution proposed for these examples generalizes to the case of spray/load alternation. First thing to note is that in the case of 'load', there is always a resulting state implied, hence both versions have the structure in (12). The difference is whether it is the 'hay', or the 'truck' that ends up in the state of being loaded. In terms of the decomposition adopted, the difference is whether the IA or the EA ends up in Spec,resP. If the EA stays in situ, the IA becomes the holder of the resulting state (Resultee), and the EA is marked INS, in line with the observations made for the verb 'stick'. If the EA moves, it becomes the Resultee, and it surfaces as ACC. 20 (13) [proc [res [ea [P ia]]]] ea stays in situ --> ins (14) [proc [res [ea [P ia]]]] ea moves --> acc A similar story can be told for the cases we have observed at the beginning, replacing res in (13) and (14) for the relevant head corresponding to the restructuring verb, and P for the embedded verb.3 An intermediate conclusion is that whatever it is that drives the movement of EA in (14), it is not the need to be assigned case, since instrumental case is available already in Spec,P. This is not as surprising as it might seem, since a similar conclusion can be drawn for pseudo-passive sentences (John was yelled at John), where the DP (John) raises from a Case position (unless there is a way how to connect the passive marking on the verb with the P's ability to assign accusative Case). Put even strongly, the DPs in the examples discussed not only do not move for reasons of Case, they move from one Case position to another. Hence, the reasons for DP movement are rather to be sought in the fact that in each movement step, there is a new predication relation is established between the moved DP and the target of movement. So in the examples at hand, the P's EA does not move for Case reasons, it moves in order to be interpreted as the subject of the result phrase. 3.4 ias of Ps With a partial solution in place, we want to know more on how case marking and case shifting happens. Is INS assigned to every DP that happens to be in Spec,P at the end of the day? Is all ACC marking connected to Spec,Res? Let us see if further patterns help to deepen our understanding and sharpen our conclusions. Let's start off with the complements of locative and directional PPs and case shifting that happens there. (15) nad I pod Ipřed I za I mezi něčím above / under / in front of/ behind / between somethingiNs location (16) nad I pod Ipřed I za I mezi něc-o above / under / in front of/ behind / between somethingAcc direction All the PPs in (15) denote simple locations, whereas those in (16) denote directions ('to above / to under' etc.). The change from locative to directional meaning is accompanied by the change of case on the P's object from INS to ACC. It is immediately clear that INS here cannot be viewed as a case that marks EAs in situ, since here INS is associated with the IA of P. It is also clear that ACC can't be considered the case assigned by V (or res) to the direct object either: all the PPs in (16) can in fact be embedded under certain types of Ns (like 'path' or 'jump') without any effect on the case of the P's object. Still there is a way how to connect the pattern at hand with the one arrived at before. In order to see that, let me first present some basic observations concerning the syntax of locative and directional PPs. A body of literature (van Riemsdijk, 1990, Koopman, 1997, Helmantel, 2002, den Dikken, 2003, Svenonius, 2004 on the syntactic side, and also Jackendoff, 1990, Zwarts and 3 See Collins (2005) for a way how to get the ia across the ea. See Taraldsen (2006) for an alternative. See Hoekstra (2004:ch.l) for an interesting proposal in the same spirit as presented here. 21 Winter, 2000, and Kracht, 2002 on the semantic side) has established that directional PPs are built on top of locative PPs. The basic structures, taken from Svenonius (2004), are given below. (17) [Place [DP]] locative adposition (18) [Path [Place [DP]]] directional adposition Again, we face the problem of how to connect the structures with Case. Making profit of the Path head in (18), it is plausible to view it as the source of ACC in directional PPs. Let me therefore propose that the DP has to raise to Spec,PathP (see (19)), which is followed by a remnant movement of the PlaceP to a yet higher position. Before I turn to some evidence for the movement, let us see where this account brings us wrt Case. (19) [PatiTrPkceTDP]]] acc (20) [Place [DP]] ins In (20), I repeat the structure for locative PPs with the DP in INS. As with the previous constructions, we can see the INS arising as a result of a DP having stayed in situ. In directional PPs, the DP moves to Spec,Path, where it surfaces as ACC. Granted the assumptions I have made, the following unified view is achieved: (21) DPs in situ are ins, DPs in derived position are acc. (22) Case marking correlates with movement, oblique marked DPs are in situ, structural case arises in derived positions. Let me call (22) the Case-Movement Conjecture (CMC), which is meant to be an update of (3). Now back to what justifies the movement in (19). Consider the following examples from Dutch: (23) de weg op de heuvel locative the road up the hill (24) de weg de heuvel op directional the road the hill up (Dutch, den Dikken, 2006) The Czech alternation in Case is replicated in Dutch by alternation in word order, where in directional contexts, the DP precedes the P for exactly the same reason as the Czech DPs are marked ACC, in accordance with CMC.4 3.5 Affected Instrumentals Surface as Accusatives Let me continue with the idea that some DPs take INS, when in situ, and switch to ACC when they move up. The following alternation then falls in place: The reason is movement of the P's ia to Spec,PathP in directional contexts. 22 (25) Petr prosel lesem Peter wentyiA woodiNs 'Peter went through the wood' (26) Petr prošel Peter wentyiA les woodAcc 'Peter walked through the whole wood/ Peter searched the wood through' Let me first present data that shed light on the INS in (25). As shown in (27), bare INS on a DP can be used in Czech to denote 'through'. The 'path' here is unbounded, which means that it can be fully contained in the 'wood.' (27) cesta lesem path woodiNs 'a path through the wood' The sentence in (25) exploits this usage, with the additional factor that the prefix (pro-) introduces a bounded interpretation of the path Peter moves along, so the interpretation is that he enters the wood on one side and leaves it on another side. For examples like (25), I propose that the INS DP, denoting an unbounded path 'through', is a structural complement of the prefix pro- that introduces the bounded interpretation of the 'through' path. The PrefixP (headed by pro-) is further embedded under the proc head, which gives the right interpretation: there is a process of Peter walking that proceeds along the bounded path through the wood. (28) [proc (walk) [pro- [ DP ]]] ins Now the change from INS to ACC brings about a kind of "affectedness" of the DP 'the wood': it has to be affected somehow by Peter having walked through it. This gives rise to a slight shift in meaning, and the sentence roughly means that Peter has searched the wood through. Theoretically speaking, the sentence with ACC brings about a state resulting from the process of walking. Again, this is to be captured by the presence of res in (26). The case marking then complies with the CMC: once the resP is present, the DP moves to its Spec and surfaces with ACC. 3.6 Interim Summary Before we go on, let us see where we are. I have started with the observations concerning active and passive non-finite complements to verbs, concentrating on the case marking of the EA. The initial observation (for passive VPs and PPs) was that INS marks EAs that are demoted. This had to be changed for EAs of Ps embedded under the verb 'stick', where the same alternation was due to semantic factors (having a resulting state or not). The conclusion which emerged was that EAs in situ are marked INS, whereas moved EAs receive ACC. For these reasons, I have suggested that DP movement is to be dissociated from Case licensing per se, and that DPs are free to originate in one Case position and move to another. (29) [proc (walk) [res [ pro- [ DP ]]]] ACC 23 The investigation of case alternations in PPs brought us further to the conclusion that there is no designated projection for either ins or acc. It was concluded that a unified view on the alternations can only be achieved if CMC is adopted: DPs are marked by Oblique Case in situ, and they receive acc once moved to a derived position. The formal mechanism underlying CMC is presented in the next section. 4 Peeling The technical problem I want to find a solution for in this section is how to make one Case (ins, associated with the base position) change into another (acc) in the derived position. There seem to be two easy options: the addition of a feature (e.g. along the lines of Sportiche, 2005, or Koopman, 2005), or the loss of a feature (along the lines of Sportiche, 1988). Let Case morphology be our guide here. Table I: Subset-superset relations of instrumental and accusative in Coll. Czech muž kuře, kost, stavení, dobrý, ,man', pi , chicken', pl ,bone', pl ,building', pl adj.,pl. nominative muž-i kuřat-a myš-i staven-í dobr-ý accusative muž-e kuřat-a myš-i staven-í dobr-ý instrumental muž-e-ma kuřat-a-ma myš-i-ma staven-í-ma dobr-ý-ma genitive muž-ů kuřat-0 myš-í staven-í dobr-ý-ch dative muž-ů-m kuřat-ů-m myš-í-m staven-í-m dobr-ý-m locative muž-í-ch kuřat-e-ch myš-í-ch staven-í-ch dobr-ý-ch It can be easily observed that morphology tells us to choose the latter way and treat the case shifting patterns discussed as a feature loss. The reason is that ins in Czech tends to be morphologically more complex than acc, which is shown in the paradigms above. Here, the ins actually seems to be built on top of the acc, by the addition of -ma. In other words, the change from ins to acc is to be understood as stranding of -ma. 4.1 How to Lose Features? The scenario of feature loss is given below in (30). This tree encodes the way DPs move from one case position to another in the constructions discussed. Starke (2005) generalizes this scenario for all DP movement (with slight qualifications concerning successive cyclic movement). I Starke (2005) calls the W in (30) the Peel, and the movement of XP is called Peeling. The adoption of Starke's Peeling proposal for the cases at hand, leads to the conclusion that ins 24 is structurally more complex than ACC, and that, in fact, accusative is contained inside the ins. The following representations for ins and ACC encode this, with ins containing two Case heads, Kx and K2, as opposed to ACC with only one Case head. (31) [Kj[K2[XP]]] (32) [K2[XP]] ins acc In these representations, ACC is a proper subset of ins, reflecting directly the morphology of Table I. In syntax, DPs are base generated as ins (at least in the positions discussed), and when they move, they move by Peeling.5 This means that when they undergo movement for reasons of extra predication, they strand the highest shell, Kx, in situ, and surface as K2Ps, that is as ACCs. 5 Nominative If (30) is the way Case movement is to be represented in the case of ACC and ins, a natural hypothesis is that (30) holds for all Case driven movement. In particular, pairs of active and passive sentences (illustrated in (33) and (34)) seem to be easily susceptible for such an approach. (33) Vypil becherovku drank becherovkaAcc 'He drank becherovka.' (34) Becherovka byla vypita becherovkaNOM was drunk 'Becherovka was drunk.' In present terms, when the DP 'becherovka' moves from the object position to the subject position, it changes case from ACC to nominative (nom). The way this can be understood under the Peeling theory of Case assignment, is to assume (30) together with the representation of nom and ACC depicted below: On analogy with the previous example, nom is a syntactic subset of ACC. When an ACC DP undergoes passivization, it moves to Spec,TP and since it moves by Peeling (30), it strands the projection of accusative in situ and surfaces as nom in the derived position. The observations above suggest that Cases are neither primitive units, nor structurally identical to one another. In fact, it appears that each Case is a unique collection of functional heads, and that individual Cases stand in structural subset - superset relations, as determined by the collection of heads they are the manifestation of. Specifically, I have argued that if we want 5 See Medová (to appear) for the same approach. (35) [K2[K3P]] (36) [K3P ] nom acc 6 The Case Hierarchy 25 to capture the systematic relationship between ACC and INS in Czech, we can consider ACC to be a structural subset of INS, which is directly reflected by morphology of certain paradigms presented in Table I. In this section, I build on cross-linguistic comparison of Case systems conducted by Blake (1994) and I propose a universal hierarchy of Cases. I show what predictions this hierarchy (understood in terms of syntactic structure) makes with respect to syncretism patterns that are expected, and I demonstrate the validity of the predictions on Czech Case morphology. 6.1 Blake's (1994) Case Sequence (with a Czech Twist) Blake (1994) discusses Case systems of the languages of the world, observing the number and kind of cases each language distinguishes. Among others, he mentions the following types of languages (with languages exemplifying these types in brackets):6 (3 7) NOM - ACC/OBL (Chemehuevi) (3 8) NOM - ERG/OBL (Kabardian) (39) NOM - ACC - GEN/OBL (Modern Greek, Semitic, Nubian) (40) NOM - ACC - GEN - DAT/OBL (Icelandic, German, Nilo-Saharan, Yaqui, Ancient Greek) (41) NOM - ACC - GEN - DAT - LOC/OBL (Latin) (42) NOM - ACC - GEN - DAT - LOC - INSTR/OBL (Slavic) (43) NOM - ACC - GEN - DAT - LOC - ABL/OBL (Turkish) (44) NOM - ACC - GEN - DAT - LOC - INSTR - ABL (Classical Armenian) (45) NOM - ACC - GEN - DAT - LOC - INSTR - ABL - COM (Tamil) What can be observed is that there is a clear hierarchy in which Cases tend to appear in the languages of the world. The hierarchy is given below: (46) NOM > ACC > GEN > DAT > LOC > INSTR > ABL > COM Interpreting this hierarchy in terms of syntactic structure means that not only is accusative formed on top of nominative by the addition of a feature (as already argued), but genitive is formed on top of accusative in a similar fashion (and mutatis mutandis for other Cases). Having rebuilt Blake's hierarchy this way in the syntax, we in fact achieve the result that the hierarchy is derived, granted (in addition) the standard assumption that if in a particular language, the head noun is able to move to the left of, say, dative (so that dative surfaces as a suffix), the DP can also move to the left of all cases lower on the hierarchy (and so they also end up as suffixes). Another consequence of the Blake's hierarchy (given its syntactic implementation) can be brought to light under particular assumptions concerning Case syncretism. Specifically, Bobaljik (2007) argues that only heads that are adjacent in the syntactic structure can be syncretic.7 It follows that if two non-adjacent Cases of the Blake's hierarchy are syncretic, all cases that intervene have to be syncretic with these two Cases. This is what Bobaljik refers to as *A-B-A. This kind of reasoning applied to Czech, however, reveals that a twist in the hierarchy is needed, and that its correct rendering for Czech implies the order LOC > DAT (47) (opposed 6 See also a summary of the Moving Right Along seminar held in Troms0 on on June 13, 2006 by Pantcheva, available at http://www.hum.uit.no/mra/. 7 See Bobaljik's work for one possible reasoning that derives this fact, but see also the implementation in Caha (2007), which derives the same result under different assumptions concerning lexical access. 26 to Blake's DAT > LOC). There are two reasons. First, it is sometimes the case that locative is syncretic with genitive, with dative not participating in the syncretism (see the first two lines below (47)). Second, dative can be syncretic with instrumental, with locative standing aside (a syncretism characteristic of the old dual). Syntactically, the reason that Czech requires the order LOC > DAT is that the Czech Locative never conveys locational meaning on its own, but always requires the help of a preposition. Hence, it does not count as a locative case in a strict sense, and the fact that it is called locative is rather misleading. (47) [ Ins [ Dat [ Loc [ Gen [ Acc [ Nom 'we' ná-m-i ná-m ná-s ná-s ná-s my 'Adj., pi.' A-ý-m-i A-ý-m A-ý-ch A-ý-ch A-ý A-ý 'both' ob-ě-m-a ob-ě-m-a ob-ou ob-ou ob-a ob-a Space limitations prevent me from demonstrating Czech Case morphology in greater detail. However (once certain intervening factors are controlled for), there are only two exponents whose distribution presents a counterexample to the hierarchy (47) coupled with the *A-B-A reasoning. The first problematic exponent is acc.sg. -i in the paradigm růže, which is syncretic with locative and dative, but accusative is (unexpectedly) -e.8 Similarly, -ou is accusative and instrumental of feminine adjectives, but the intervening Cases have a different ending. I suppose that these are examples of homonymy, and not systematic syncretism. (47), then, is the Case hierarchy that seems motivated on both morphological and typological grounds, underlying the syntax of Czech Case. 7 Conclusions The proposal I have put forth here is that the distinction between Structural and Oblique Cases is a a difference in the amount of structure. Specifically, I have proposed that Oblique Cases have a large number of functional projections and they in fact contain the Structural Cases inside them. This proposal has a direct bearing on several questions that have been discussed. First, the fact that Oblique Cases tend to be expressed by more morphology, and that Oblique Cases can be missing from a given language (Blake's hierarchy), are both in direct correlation with the proposed hierarchy of Cases. Second, a mechanism for Case shifting, namely Peeling, has been proposed, following the guidelines of Starke (2005). References Asbury, Anna. 2006. Towards a typology of morphological case. In: Svenonius, P. (ed.). Nord-lyd: Troms0 Working Papers in Linguistics, special issue on Adpositions 33.1:120-148. Bayer, Joseph, Markus Bader, and Michael Meng. 2001. Morphological underspecification meets oblique case: Syntactic and processing effects in German. Lingua 111:465-514. Blake, Barry. 1994. Case. Cambridge: Cambridge University Press. Bobaljik, Jonathan. 2007. ABCs of comparative suppletion. Ms., University of Connecticut. Caha, Pavel. 2007. The Superset Principle. Ms., University of Troms0. Chomsky, Noam. 1995. The minimalist programme. Cambridge, MA: MIT Press. 8 Incidentally, the accusative sg. -i also gives rise to A-B-A pattern of another exponent in the paradigm růže, namely -e, which appears in nominative and genitive. 27 Collins, Chris. 2005. A smuggling approach to the passive in English. Syntax 8:81-120. den Dikken, Marcel. 2006. On the functional structure of locative and directional PPs. Ms., CUNY. Dotlačil, Jakub. 2004. The syntax of infinitives in Czech. Master's thesis, University of Troms0. Fong, Viviane. 1997. The Order of Things: What Directional Locatives Denote. PhD thesis, Stanford University. Franks, Steven. 1995. Parameters of Slavic morphosyntax. Oxford, New York: Oxford University Press. Helmantel, Marijon. 2002. Interactions in the Dutch adpositional domain. Utrecht: LOT. Hoekstra, Teun. 2004. Arguments and Structure. Studies on the Architecture of the Sentence. Sybesma, R. et al. (eds.). Berlin: Mouton de Gruyter. Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press. Koopman, Hilda. 1997. Prepositions, postpositions, circumpositions and particles: The structure of Dutch PPs. Ms. UCLA. http://www.linguistics, ucla.edu/people/ koopman / papers/pp .pdf Koopman, Hilda. 2005. Korean (and Japanese) Morphology from a Syntactic Perspective. Linguistic Inquiry 36:601-633. Kracht, Markus. 2002. On the semantics of locatives. Linguistics and Philosophy 25:157-232. McCawley, John. 1968. Lexical insertion in a grammar without Deep Structure. In: Darden, B. J., C.-J. N. Bailey &; A. Davison (eds.): Papers from the fourth regional meeting of the Chicago Linguistic Society, Chicago. Medová, Lucie. 2007. Reflexive clitics, Movement, and Layered Case. To appear in: Proceedings ofFDSL 6.5. Frankfurt: Peter Lang. Ramchand, Gillian. 2007. Verb Meaning and Lexicon. Ms. Troms0, lingBuzz/000346. Ramchand, Gillian and Peter Svenonius. 2004. Prepositions and External Argument Demotion. Handout of a talk in Oslo. Downloadable at www.hum.uit.no/a/svenonius/papers/RamchandSvenonius04Oslo.pdf Riemsdijk, Henk van. 1990. Functional prepositions. In: Pinkster, H. &; I. Genee (eds.). Unity in diversity: Papers presented to Simon Dik on his 50th birthday: 229-241. Dordrecht: Foris. Romanova, Eugenia. 2004. Transitive, intransitive and passivizing... prefixes. Handout of a talk at the Argument Structure Workshop, November, Troms0. Romanova, Eugenia. 2007. Constructing Perfectivity in Russian. Ms. Troms0. Sportiche, Dominique. 1988. A Theory of Floating Quantifiers and its Corollaries for Constituent Structure. Linguistic Inquiry 19:425-451. Sportiche, Dominique. 2005. Division of Labor between Merge and Move: Strict Locality of Selection and Apparent Reconstruction Paradoxes. Ms. UCLA. lingBuzz/000163 Svenonius, Peter. 2004. Spatial P in English. Ms. University of Troms0. Starke, Michal. 2005. Lecture in Troms0, Fall. Taraldsen, Tarald. 2006. Resumptive pronouns and stranded Case in the Bod0 dialect. Talk at NORMS Workshop, October, Troms0. 28 Veselovská, Ludmila. 2003. A note about nothing. Ms., Palacký University, Olomouc. Wurmbrand, Susanne. 2001. Infinitives: restructuring and clause structure. Berlin, New York: Mouton de Gruyter. Zwarts, Joost and Yoad, Winter. 2000. Vector Space Semantics: A Model-Theoretic Analysis of Locative Prepositions. Journal of Logic, Language, and Information 9:169-211. 29 Infinitives under 'have'/'be' in Czech Markéta Ceplová ceplova@mit.edu 1 Introduction The verb 'have' in Czech can show up in three different versions: 1) possessive 'have', 2) obligational 'have', and 3) wh-existential 'have', as seen in the following example: (1) a. Anna měla kočku. Annanom Sg have^,^.past catacc,sg 'Anna had a cat.' b. Anna měla spát. Annanom Sg ssgf.past sleepjny 'Anna was supposed to sleep.' c. Anna měla kde spát. Annanom 3^ have3sgf,past where sleepin/ 'Anna had a place where she could have slept.' Parallel to the verb 'have', similar options, with the exception of the obligational construction, are also available to the verb 'be': (2) a. Na zahradě byla kočka. in garden he3sgf,past catnom.Sff 'There was a cat in the garden.' b. * Na zahradě bylo spát. in garden be3sgn,past sleepin/ c. Na zahradě bylo kde spát. in garden be3sgn.past where sleepin/ 'There was a place in the garden where we could have slept.' The three options vary significantly in several factors. The first difference is what complement the 'have'/'be' verbs take, a DP/NP ('cat') in the first case, an infinitival verb ('sleep') in the second (this option is not a possible complement for the verb 'be'), or a wh-complement with a non-finite (infinitival or subjunctive) verb in the third ('where to sleep'). Parallel to these three options of complements is the interpretation of the whole clause. While the first case receives a possessive/existential interpretation, the second, as well as the third one, receives a modal interpretation. The difference between the last two options is in part in the specific flavor and force of the modality. A bare infinitival verb, as in (lb)), receives an 31 obligational modality (a standard 'have-to' interpretation, as in e.g. Bhatt, 1998), the modal force in the third version of the verb 'have' receives a modality of 'existential possibility.' The constructions along the ones in (lc) and (2c) are available not only in Czech, but also in other Slavic, Romance, as well as in other languages1. These constructions have been given several different analyses and different names in the literature: irrealis free relatives (Grosu and Landman, 1998), indefinite free relatives (Caponigro, 2001), indefinites in disguise (Rappaport, 1986; Rudin, 1986), infinitival existential sentences (Babby, 2000) or wh-existential constructions (Izvorski, 1998; Pancheva Izvorski, 2000). I will henceforth call them 'wh-existential constructions,' or WHECs for short. In this paper I look mainly at the wh-complements of the wh-existential constructions. But before that, let us look at the constructions as whole. WHECs show several notable properties, that will not be a subject of this paper, but are still worth setting clear up-front and that distinguish them from other cases that could look similar. For one thing, as already mentioned above, the constructions obligatorily receive a modal interpretation - one that is parallel to the English case in (3), namely the "could" interpretation that is in (4a), but not a "should" interpretation as in (4b) that is an alternative interpretation of the English sentence here2: (3) I have something to read. (4) a. I have something that I can read. b. I have something that I have to read. To be somewhat more precise about it, the available modality is claimed by Pancheva Izvorski (2000) to be "associated with a particular modal interpretation - an existential modal restricted by a bouletic accessibility relation" (Pancheva Izvorski, 2000, p. 24), or "the root modal of circumstantial possibility" (p. 36). Another point worth mentioning is that the number of possible matrix verbs that can be involved in WHECs is argued to be strictly limited by Pancheva Izvorski (2000). Besides 'be' and 'have' there are (in some languages) also other ones available in WHECs, namely Pancheva Izvorski (2000) includes 'find', 'look for', 'choose' in the languages she looks at, see her ex. (39) from Russian on page 35. However, neither of these alternative matrix verbs allows properties of the wh-clause parallel to the ones allowed by 'be' and 'have' in Czech (if available at all). (5) On iscet [s kem poexat']. he look-for3Sg.pres [with whom gOjn/] 'He is looking for someone to go with.' For the purposes of this paper, I am going to concentrate on the matrix verbs 'be' and 'have'. The properties of the wh-complement, which is the main focus of this paper, are as follows: I first show that in general and more specifically in Czech 1) the verb needs to be non-finite, namely infinitival, or subjunctive (see section 2); 2) it needs to involve one or more wh-words (see section 3); 3) the size of the complement has to be small enough to enable restructuring, at least optionally (see section 4; 4) the (understood) Subject of the wh-clause is obligatorily coreferential with the Subject of the matrix verb (where available). 1 Pancheva Izvorski (2000) includes in the list also Greek, Hebrew, dialects of Arabic, and Yiddish. 2 For arguments and treatment that exclude the "should" modality reading in WHECs see (Pancheva Izvorski, 2000, p. 27n). 32 2 Non-finiteness restriction on the wh-complement Besides other restrictions, there is a strict restriction on the verb in the wh-complement of the matrix verb. Specifically, the verb has to be either infinitival, or have subjunctive3 tense, no finite verb is allowed: (6) a. infinitive Petr neměl koho pozvat. Petrnom.Sff not-have.?S9.pasí whoacc.Sff invitein/. 'Petr didn't have anyone he could invite.' b. subjunctive Petr neměl koho by pozval. Petrnom Sg not-have^.past whoacc.Sff c. finite verb * Petr neměl koho pozval. Petrnom.Sff not-have.?S5.paSí who„cc.S9 invite^.pasř. In general, the subjunctive form of the Czech verb is a compound form, including an auxiliary (derived from 'be') that behaves as a clitic and agrees in person/number with the Subject, and a past participle of the verb that agrees with the Subject in person/number/gender (marked here as 'pprt.' in the glosses). The subjunctive verb, unlike the infinitival one, can also take a Subject, marked with Nominative Case in most instances, though it can leave it null [pro, or PRO): (7) Petr by přišel rád. Petrnom Sg aux^ comepprt.ssgm glad3Sgm 'Petr would be glad to come.' (8) Přišel by rád. • •oine,,,,,.,..;,,,„. a\ix3sg g\ad3sgm 'He would be glad to come.' (9) Doporučil jí, aby přišla brzo. recommendpasí.ssffm shedaí.ss thats^-.^s comepast,3sgf early 'He recommended to her to come early.' (10) * Petr přijít rád. Petrnom sg comem/ glad.jsgm When a subjunctive form of the verb is used in the wh-complement, it obligatorily agrees with the (matrix) subject, in person, number, and gender, to the extent the agreement has effects on the morphological form of the verb. Since infinitives in Czech do not show any agreement, nothing similar can be observed in those cases: (11) subjunctive 3 This form of the verb is usually described in the Czech literature as 'conjunctive'. To the extent that 'conjunctive' ('kondicionál' in Czech) can be taken as another name for 'subjunctive' it is fair to talk about it this way. Junghanns (1999) also gives what he calls 'subjunctive markers' for Czech, equating 'subjunctive' to the standard Czech 'kondicionál'. The "markers' that he provides are the same ones I have been looking at here. 33 a. Petr měl, co by mu koupil. Petrnom Sg m \\SNQpasi Ssgm wha,taccSj auxsu5j3sg he^^ Sg 'Petr could buy him something.' b. Marie měla, co by mu koupila. Marienom.Sff./ havepast,3sgf whatacc.Sff auxsubj.3sg hedat,sg buypast.sgf 'Marie had something she could buy him.' c. Máme, co bychom mu koupili, havens iv\ whatacc.Sff auxsufej ip\ he^at.sg b\iypast.pi 'We have something we could buy him.' d. Mám, co bych mu koupila. have^ej, isg whatacc-sff &&X-subj.lsg ^^dat.sg b\iypasi sgj 'I have something I could buy him.' (12) infinitive Měl (jsem) / měla (jsem) / měli (jsme) / mám co havepast.3sgm (auxJsff) / havepast.3sgf (auxJsff) / have^.^ (auxJj)Z) / h&vepresAsg what mu koupit. ^dat.sg buyjn/. 'He/she/we/I had/have something he could buy him.' Another difference between the two verbal forms also includes the fact that unlike the infinitival wh-complements that allow restructuring (and thus clitic climbing), wh-complements with a subjunctive verb do not have this option. In the rest of this paper I look primarily at cases with the infinitival version of the verb, noting in passing where relevant or important differences obtain without necessarily providing a detailed account of them. 3 Wh-words in the wh-complement In this section I look at what wh-words can be present in the wh-complement and what determines their exact status and shape. I first go over the ways the form of the wh-words is determined, then specifically looking at those that come with a Nominative Case and restrictions on those, based on that I also look at where the wh-words start and how far they can go. 3.1 Restricting the form of the wh-words I first show that the syntactic restrictions (if any) the wh-words show come from the wh-complement itself (specifically its verb). The restrictions that come from the matrix clause/verb (no which- or how-many-constructions can appear in the wh-complement), are mostly of semantic nature. Wh-words with '-ever' (as the English 'whatever', 'whoever', 'wherever'...) are prohibited in the wh-existential construction: (13) Kalina ma koho(*koli) pozvat. Kalinanom.Sff havepres.3sg whomever) acc.Sff invitein/. 'There is someone Kalina can invite.' 34 This clear impossibility of adding -ever to the wh-words is a strong argument against treating the wh-complement as a Free Relative. Pancheva Izvorski (2000) argues that based on the difference in the (non)availability of -ever with the wh-words, as well as other properties of the wh-clauses, the wh-clauses pattern with interrogative clauses, rather than Free Relatives. The (lack of) availability of wh-pronouns with -ever in English, as in Czech, strongly support that conclusion, as seen below: (14) a. Kdo(koli) přijde, bude považován za pozvaného. who(ever)nom.Sff come3sg.pres be;?S(;./MÍ considerednom.Sffm for invitedacc.ss 'Who(ever) comes will be considered as having been invited.' b. Vím, kdo(*koli) přišel. knowJsff.j,res who(*ever)nom.Sff come3sgm,past 'I know who came.' (15) a. Who(ever) came was sent to the kitchen, b. I wonder who(*ever) came. The wh-existential construction is restricted to clauses that include at least one wh-word in the complement of "be"/"have". Without it, the sentence receives an entirely different interpretation that is not related to the construction under disscussion, as seen below: (16) a. Kalina má někoho pozvat. Kalinanom.S5 have3sg.pres someoneacc.sg invitein/. 'Kalina is supposed to invite someone.' b. Kalina má koho pozvat. Kalinanom.Sfl have%;.pres whoacc.S9 inviteffl/. 'There is someone who Kalina can invite.' The wh-existential constructions allow almost any wh-word in the wh-complement, both nominal as well as adverbial ones, as seen in (17): (17) a. Josef má (určitě) koho pozvat. Josefnom.S9 h.ave3sg.pres (certainly) whoacc,sg invitein/. 'Josef certainly has someone he can invite.' b. Josef má kam Annu pozvat. Josefnom.S(; hme3sg_pres where Anna,,,,.,,, invitein/. 'Josef has a place he can invite Anna to.' c. Josef (opravdu) nemá proč tam chodit. Josef (really) neg-have3sgpres why there goinf. 'Josef does not have a(ny) reason to go there.' The wh-complement allows more than one wh-word: (18) a. (a case in which everyone has to keep introducing people to other people, but Josef refuses to continue and a friend is trying to defend him) Josef už opravdu nemá koho komu představit. Josefnom.S(; already really not-have5sff.pres whoacc.Sff whodaí.Sff introduce^/ 'Josef is done with all introductions.' b. (A statement of an usherer in a completely full space:) 35 Už nemám kam koho posadit, already not-haveiSg.pres where whoacc.Sff seatinf 'No-one else can be seated here.' Yet a further note that is needed here is a note about nominative wh-words. While infinitives in Czech cannot assign a Nominative Case, or take any overt Subject, see e.g. (19), there still are some cases of nominative/Subject wh-words in the WHECs, as e.g. in (10) (repeated here as (20)): (19) * (Petr) přijít rád. Petrnom.ss, comeinf glad^m (20) Neměl kdo přijít. not-have5sffm.pasí whonom comem/. 'There wasn't anyone who could come.' At this point, the only place the Nominative Case could be assigned from is the matrix verb, provided it itself is able to do that. I will return to these cases, as well as to the discussion of the Subject and its interpretation and other cases of nouns with Nominative Case in the wh-complement, in section 5. 3.2 Wh-movement The wh-words also have the option to move higher, out of the wh-complement, e. g. in cases the whole sentence is turned into a wh-question (showing that the wh-complement is not an island4; compare the following cases with a declarative sentence in (18a)): (21) a. Má Josef koho komu představit? have5sff.j,res Josefnom.Sff whoacc.Sff whodai.Sff introduce^/. 'Can Josef still make more introductions?' b. Koho má (ještě) Josef komu představit? whoacc.Sff have5sff.j,res (still) Josefnom.Sff whodai.Sff introduce^/. 'Who could Josef still introduce to someone?' 'Who is Josef (still) supposed to introduce to someone?' c. Komu má (ještě) Josef koho představit? whomdai.Sff have5sff.j,res (still) Josefnom.Sff whoacc.Sff introduce^/ 'Who could Josef (still) introduce someone to him?' 'To whom is Josef (still) supposed to introduce someone?' Three things need to be added to the examples in (21). One is that in this case, either of the two wh-words can move higher, without any clear preference, as seen in (21b) and (21c). 3.2.1 Alternative interpretations The second thing to note is the fact that the sentences with one of the wh-words moved out of the wh-complement (above the verb 'have') allow two possible interpretations, as long as at least one wh-word stays in the wh-complement (see the translations in (21)). One of these two possible interpretations is the one that wh-existential constructions bring (i. e. asserting 4 I would like to thank David Pesetsky for pointing out the islandhood status to me. 36 the existence of the entity that the wh-word stands for, and its availability to the subject of the matrix clause, together with an existential force of a modality; for more on that, see the introduction in section 1). The other one is a "have-to" interpretation that "have" plus an infinitive can bring in Czech independently (in a 'have-to' construction): (22) a. Josef má přijít zítra. Josefnom.Sff have3sg.pres comeinf. tomorrow 'Josef should (is supposed/expected to) come tomorrow.' b. Kdy má Josef přijít? when have5sff.j,res Josefnom.Sff comein/. 'When should Josef (is J. supposed to) come?' However, when all wh-words move higher, only the "have-to" interpretation remains, and the one that is a part of the wh-existential construction is not available at all: (23) Koho komu má Josef představit? whoacc.Sff whodai.Sff have5sff.j,res Josefnom.Sff introduce^/. 'Who is Josef supposed to introduce to someone?' Another take on this argument is also available from cases that do not show any alternatives of interpretation when the wh-words are moved out of the wh-complement (as in sentences with subjunctive verbs in the wh-complements): (24) * Koho komu má Josef by představil? whoacc.Sff wh.Odat.sg ^^Ssg.pres Josefnom.Sff auxsufej_c2 introduce,^.,jSffm In cases with subjunctive verbs in the wh-complement, when all the wh-words are moved out of the wh-complement, the sentences then become ungrammatical, as expected. 4 Size of the wh-complement The size of the wh-complement has standardly been taken to be a CP, see e. g. Babby (2000); Pancheva Izvorski (2000). However, Czech WHECs have some properties that argue against this treatment. Namely clitic climbing that is completely grammatical in WHECs, while it has been argued to be impossible out of all finite, as well as some infinitival clauses, e. g. non-finite wh-complements of interrogative verbs (such as 'know' or 'ask'). The clitic climbing itself then strongly suggests the size of the wh-complements of WHECs to be smaller than CP. (25) a. Petr má kam ho pozvat. Petrnom.Sff have3sg.pres where heacc.Sff_d. invitein/. 'Petr has a place where he could invite him.' b. Petr ho má kam pozvat. Petrnom.Sff heacc.Sff_cz. have3sg.pres where invitein/. It has been argued in the literature (Rezac, ms.; Dotlačil, in progress) that clitics in Czech cannot climb out of a CP, not even out of a non-finite one, as seen in the following examples: (26) a. * Ale nevím hoj opravdu jak zapisovat tj. but not-know him really how record^/. 'But I really do not know how to record it.' [Lenertová (2004), taken from Dotlačil (in progress)] 37 b. * Petr mu, nevěděl koho doporučit t, Petrnom.Sff hedaLsg_cL not-know^.^ whoacc.Sff recommendin/. 'Petr did not know who to recommend to him.' Furthermore, Rezac (ms.) argues that clitics can only climb out of restructuring infinitives (based on looking at the restructuring infinitives within the framework of Aissen and Perlmutter, 1983). Based on some tests, and adopting conclusions from Wurmbrand (1998), Rezac concludes that the infinitive that allows clitic climbing has to be a restructuring one, and more specifically of the size of a bare VP (while a non-restructuring infinitive as a complement is larger). For Dotlačil, the restructuring infinitive is (or at least can be) slightly bigger than just a bare VP. Based on Lenertová (Lenertová, 2004)5, "clitic climbing is not degraded if the clause hosts a temporal adverb." Thus "the infinitival clause must be realizable with T node but without PRO" which Dotlačil argues is the same as that for long-distance agreement in Czech. Based on the previous section, the possibilities of the size of the wh-complement in WHECs are either bare VP, or vP. I suggest treating the wh-complements of WHECs as an vP based on the wh-nature of the complement. 5 Subject of the wh-complement: Control or Raising? The embedded non-finite (especially the infinitival) wh-clause does not allow (with some exceptions to which I return in section 5.2) an overt subject to be present. It has been commonly understood that the understood Subject of the infinitive has to be coreferential with the Subject of the matrix verb. Thus the following sentence in (27a) has only the Subject interpretation that is provided, with no other Subject interpretation, as e.g. in (27b), possible. (27) a. Josef má koho Marii představit. Josefnom.Sff have3sg.pres whoacc.Sff Mariedai.Sff introduce^/. 'Josefi has someone who he, can introduce to Marie.' b. 'Josef, has someone who he^ can introduce to Marie.' There are several options for capturing this strong coreference. One option, that looks very tempting, would be to consider this clS db CclSG of Raising. Another option, considered and argued for in (Pancheva Izvorski, 2000), treats the WHECs, and the coreference between the two Subjects, clS db CclSG of Obligatory (Exhaustive) Control. In this section I argue that the Subject of the embedded verb (in the wh-complement) is best treated clS db CclSG of Raising, rather than Control, contra Pancheva Izvorski (2000)'s arguments. I also provide reasons that the Raising treatment for the Subject provides the right treatment for arguments with Nominative Case in the wh-complement, be it overt Subjects, or nominative forms of 'who'. 5.1 Pancheva Izvorski's arguments against Raising Here I briefly look at Pancheva Izvorski's arguments against considering Raising in WHECs. It is based on several parts: 1) arguing that "A-movement out of the wh-c\auses is theoretically problematic", 2) trying to account for the "lack of agreement between Mary and be", and 3) the 5 This quotation is taken from Dotlačil (2005). 38 fact that "Mary cannot appear in the nominative when the matrix predicate is be" (p. 65)6 To these three arguments, she then adds the lack of availability of expletives in WHECs (standardly associated with cases of Raising), and the lack of truth-functional equivalence between active and passive sentences. I first look at Pancheva's argument from the lack of Nominative Case and agreement, then I look at the lack of expletives and truth-functional equivalence between active and passive sentences, and then at the argument about the "theoretically problematic" A-movement out of the wh-complements. In the end, those arguments, as they stand, turn out not to be tenable, or relevant, for the cases of WHECs in Czech that are considered in this paper. I provide the way to treat (overt) Subjects in WHECs in section 5.2. 5.1.1 'Be' taking more than one (nominal) argument? I look first at the arguments 2) and 3), i.e. the (im)possibility of 'be' assigning Nominative Case together with showing agreement (or, taken together, taking an overt Subject in addition to the wh-constituent) in WHECs. I show that the cases that Pancheva Izvorski (2000) considers are not surprising at all, given the properties of the matrix verb 'be' and thus cannot be used as good arguments against Raising. It is worth noting that the verb 'be', unlike the verb 'have', can only take one (nominal) argument in general7, as well as in WHECs: (28) a. Byla tam hruška. ^-*e3sgf .past there pearnom Sg 'There was a pear there.' b. * Josef byl/-a tam hruška/hrušku. Josefnom Sg ~^o&3sgm/f .past there pearnom Sg/pearacc Sg c. Josef měl hrušku. Josefnom Sg have^gnj pas£ pearacc Sg 'Josef had a pear.' (29) a. Bylo tam co jíst. be3sgn.past there what eatinf 'There was something that one can eat/that can be eaten.' b. * Josef byl/-o (tam) co jíst. Josefnom Sg ^o^ssgm/n.past what eatjny c. Josef měl co jíst. Josefnom Sg have^ggnj pas£ what eatjnj 'Josef had something that he could eat.' Based on that property of the matrix verb 'be', the absence of Nominative Case on Mary is accounted for because 'be' can only assign Case to one argument and in WHECs it, by definition, takes the WH-constituent as its argument. To the extent that the wh-constituent as an argument needs a Case, there isn't any other Case that 'be' could assign (and, as shown in (10), infinitives in Czech do not assign Nominative Case either). 6 As much as one would like to see an example of either one, or both, 2) and 3), there is none, not even ones showing their non-existence or ungrammaticality, provided in Pancheva Izvorski (2000). 7 The existential verb 'be' in Czech, as well as in e. g. English, takes a locative (adverbial) argument as well. For the purposes of this paper I am going to put that issue aside. 39 Parallel to the inability of 'be' to assign Nominative Case to a 'Subject', the lack of agreement is not surprising since assigning Case is standardly taken to go hand-in-hand with Agreement (both happening through the checking of features through the operation Agree for Chomsky, 2000). Thus neither the lack of a Nominative Case, nor the lack of agreement with the matrix verb 'be' argues against Raising in WHECs. 5.1.2 Expletives Pancheva Izvorski (2000) provides the prohibition of weather-subjects, as well as expletives as a support for her argument in favor of Control, rather than Raising as the basis for the coreference of the matrix and embedded Subjects. The cases of impossible "weather-subjects" for Pancheva Izvorski include the following ones from Russian (Pancheva Izvorski, 2000, p. 65, ex. 91): (30) * Est' kogda idti dožd / doždju'. he3sg when goinf rainnom / raindai 'There is a time such that it can rain then.' This case is bad in Czech too, but there still are other potential cases to warrant the other conclusion (arguing for Raising rather than Control): (31) a. * Je / není kdy pršet. be3sg/pres / not-be3sg.pres when raminf b. ? Má / nemá už kam pršet. have3sg/pres I not-have^g.pj-es already where-to rainjn/ 'There is/isn't already place that could take rain.' c. Je / není co jíst. be3sg.pres I not-be3Sff .pres what eating 'There isn't anything one could eat.' What is even more important for the case provided here in (31c) is that a version of that is also possible with an expletive8: (32) (V)ono nebylo co jíst. itnom.sg— expl. ^-Ot-\)(i3sgnpasi what eatjnj 'There wasn't anything one could eat.' Thus the argument from the lack of possibility of expletives, as well as no "weather-subjects" is not as clear-cut as one could hope for. Another comment worth making about the possibility of Raising from the (infinitival) wh-complement is that the matrix verb does play a role here. Given that Raising is a name for movement for Case, it can only happen with verbs that can possibly assign a Nominative Case. That is exactly where 'be' and 'have' as matrix verbs differ in Czech, as already argued above in section 5.1.1. Thus, Raising is only a matter to be considered with the matrix verb 'have', not 'be'.9 8 For more on the expletive "(v)ono" in Czech see Rezac (2004). 9 A rather interesting comment on this topic was made to me by Norvin Richards (p.c), on the note that 'have' could be treated as 'be' plus Raising. I leave that comment, while interesting, aside for now. 40 5.1.3 No A-movement out of wh-clauses? Here I will look at what "theoretical" evidence against Raising out of wh-c\auses there is that warrants Pancheva Izvorski's conclusion and give evidence that A-movement out of the wh-constituents in WHECs is possible (parallel to evidence of such movement in Russian provided by Babby (2000)). To start with, it has been generally accepted that A-movement out of CPs is theoretically problematic. However, I have already argued (and shown) in section 4 that the wh-constituent (from which the Raising would need to occur here) does need to be smaller than a CP since it independently shows other properties that are restricted to infinitival complements smaller than a CP (e.g., clitic climbing). If the wh-constituent is indeed smaller than a CP, it is worth looking at the possibility of Raising in WHECs again. On the same note are arguments by Babby (2000), that show that the "understood" Subject of the infinitive is in fact a true Subject of the infinitive that raises higher and ends up being in a position where one might expect the Subject of the matrix verb 'be'. The cases that Babby (2000) is looking at are Russian WHEC cases, called by Babby Infinitival Existential Sentences, or IES; examples are given in (33), (Babby, 2000, page 1, ex. (I))- (33) a. Nam est' gde spať. us^aí there-is where to-sleep 'There is somewhere for us to sleep.' b. Nam negde spať. us^aí nowhere to-sleep 'There is nowhere for us to sleep.' Babby's arguments go along the following lines. First of all, Babby adopts the statement that "infinitive clauses in Russian all have a dative subject, overt or null" (Babby, 2000, p.3). If the Subject "is overt, it is normally fronted, occupying the Spec position on the sentence's highest functional projection" (ibid). Based on that Babby (ibid) argues that the structure in (34b) is preferable over the structure in (34a): (34) a. Nam, [IP [VP [v est'] [Cp gde^ [IP U spať t,-]]]] b. [if [vp [v est'] [cp gde^ [IP PRO, spať t,-]]]] Further "reasons for assuming that the dative in (1) [(33) here] is in fact the subject of the infinitive" {ibid). First reason is that "pronouns other than the dative pronouns systematically prepose out of the infinitival clause" - as e.g. in (35), (Babby, 2000, p. 4, ex. (6), (7)), which would result in having to claim that the verb 'be' in Russian can take Subjects marked with different Cases (at least accusative and dative ones). (35) a. Menja, nekomu bylo [qp vstrečať t,] meacc no-onedai was3sgn to-meet 'There was no one to meet me.' b. Ne dumajú čtoby teper' nás, býlo [cP za čto upreknuť ti]. 'I don't think that there is anything to rebuke us for now.' Another argument for Babby is that when the wh-word (K-word in Babby (2000)) is itself the dative subject of the infinitive, no other (non-wh) dative argument of est' can occur. The 41 last argument Babby (ibid) gives concerns the properties of the matrix verb 'be' in Russian negative WHECs where it occurs in the copula version which cannot assign theta roles. Thus the only verb that could assign a theta role to the dative arguments is the infinitival verb here. Thus for Russian, Babby concludes, the most reasonable way of treating the dative subjects is treating them as subjects of the infinitival verb in the complement of the existential verb, and systematically preposing them outside of the wh-complement. 5.1.4 Conclusion While Pancheva Izvorski (2000) presents a fairly convincing argument for Control in WHECs for Bulgarian, the conclusion is still somewhat half-way for Czech based on the sections above. In the next section, I provide some more arguments for Raising (parallel to some extent to the Russian cases provided by Babby, 2000) in Czech WHECs and a study of arguments with a Nominative Case in the wh-complements in section 5.2. 5.2 Nominative arguments/wh-words in the (non-finite) wh-complement Even though the infinitival verb does not allow an overt Subject in Czech (as noted in section 2), the wh-existential construction, in certain and restricted cases, does allow an overt Subject in the wh-complement. Both a nominative wh-word (whonom, which is clearly morphologically marked as nominative), as well as a name, or other nominals, are available. (36) a. Nemá kdo přijít včas. not-have^g.pres whonom come^/ on-time 'There isn't anyone to/who would come on time.' b. (Budou hrát,) dokud (se) bude mít s kým (se) hefuL3pi playin/. until (reflacc_d.) hefuL3sg havein/. with whoinsLsg (reflacc_d.) Petr utkat. fightin/. '(They will play) as long as Petr will have someone to play with.' While it seems that both 'be' and 'have' are available and can occur with a Nominative argument in the wh-complement, the situation is not that simple. While cases with 'be' as the matrix verb and an infinitival form of the verb in the wh-complement are fine (with either a nominal or adjectival wh-word), they do not allow a Nominative Case to be assigned to an argument (compare the ungrammatical cases in (37) with parallel ones with the matrix verb 'have' in (38)): (37) a. Tady (si) není kam (si) sednout. here refidai_d. not-be^.^ where refldai_d. sit-downin/. 'There is no place to sit down here.' b. Sem není koho pozvat, here not-bepres.5sff whoacc invitein/. 'There is nobody to/we could invite here.' c. * Není kdo přijít brzo. not-be5sff.j,res whonom comein/. early d. ?? Tady (to) není kdo (to) uklidit. here (itacc) not-be5sff.pres whonom (itacc) cleanin/. 42 e. (Budou hrát) dokud bude s kým se (*Petr) utkat. hzfut.3Pi play inf. until befuL3sg with whoinsi. refl acc—cl P^trnom Sg fightin/. 'They will play as long as there will be someone one could fight with.' (38) a. ? Nemá kdo přijít brzo. not-have3sg.pres whonom,sg comeinf early 'There isn't anyone who could come early.' b. Tady (to) nemá (to) kdo uklidit, here (itacc) not-have5sff.j,res (itacc) whonom cleanin/ 'There isn't anyone who could clean it here.' c. (Budou hrát) dokud (se) bude mít s kým (se) hefuL3pi playin/. until (reflacc_d) hefuL3sg havein/ with whoinsi. (reflacc_d) Petr utkat. Petrnom sg íightjnj 'They will play as long as Petr will have someone he could fight with.' The question that this situation raises then is what assigns the Nominative Case, since it has already been shown that the infinitival verb itself cannot do that. The options are basically two: either it is provided from the matrix verb, or some other null verb (e.g., a modal which needs to be present in the construction anyway, see the initial overview of WHECs in section !)• One way of checking these two options is trying to check whether the tense of the matrix verb has any effect on the status of the sentence, since turning the matrix verb into an infinitival form should strip it of the ability of assigning Nominative Case.10 As the next case shows, if we try to put an infinitival matrix 'have' in the construction, the Nominative Case is no longer available, while other Cases do stay available11: (39) a. Ze by měl kdo přijít (se mi zdálo that auxsubj.3sg havepprt.3sgm whonom come^/. refiacc_C2. l^at — cl. S&&'ÍIlpast3sgn nepravděpodobné). not-likely 'It didn't seem to me to be likely for anybody to come.' b. * Mít kdo přijít se mi zdálo nepravděpodobné. havein/. whonom comein/. reflacc_cZ. ldat-d. seempast.3sgn not-likely c. Mít s kým si povídat je jedna z nejdůležitějších haveinf with whojnsí reňdat-ci talk i>e3sg,pres one from most-importantffen věcí v životě. thing^.^ in life 'To have someone one can talk to is one of the most important things in life.' Another possible argument for claiming that it is the matrix verb 'have' that assigns the Nominative Case would be looking at the possibility of another nominative argument/Subject of a finite 'have'. Parallel to the cases of Russian dative subjects, mentioned in section 5.1.3, a nominative form of who in the wh-complement blocks any other nominative Subject in WHECs: 10 I would like to thank David Pesetsky (p.c.) for suggesting this to me. 11 The point of the other Cases is to show that the infinitive form of the matrix verb 'have' does not interfere with other properties of the construction. 43 (40) * Josef Josef, neměl not-have kdo who. přijít. comein/ nom.sg ' 3sgm.past nom Even though (or because of the case that) the sentences are grammatical, provided the matrix verb ('have') is finite and there isn't any other argument requiring a Nominative Case, it still needs to be specified how the Nominative Case can be assigned into the wh-complement (where the Subject is located in the examples given above). The fact that the nominative arguments can be found in the wh-complement further strengthens the case for Raising in Czech, given that if the Subject of the infinitival verb was just a PRO, there would be no reasonable way of lowering the argument/wh-word to the place where it is found. I argue that the Case in these examples is assigned in situ, without the noun/wh-word raising higher to be assigned the Case. 6 Conclusion In this paper I have shown that the wh-existential constructions in Czech have two options of the non-finite form of the verb in the wh-complements of matrix verbs 'be' or 'have', namely they can appear in subjunctive or infinitival forms. I have argued that the wh-words that have to be present in the wh-existential constructions are determined in their form by the verb in the wh-complement; the wh-word are also allowed to long-distance wh-move out of the wh-complement, provided that at least one wh-word is left inside the wh-complement; when one or more wh-words moves out of the wh-complement with an infinitival form of the verb, the sentence gains another alternative of interpretation - one of a have-to construction. I have further argued that the wh-complement in wh-existential constructions has to have a structure smaller than CP, since it allows restructuring that is the prerequisite of clitic climbing which is freely available in wh-existential constructions. Based on the previous conclusions I have also shown that the Subjects of the matrix and the embedded (non-finite) verb are related through Raising, rather than Control. References Aissen, Judith L. and David M. Perlmutter. 1983. Clause reduction in Spanish. Perlmutter, D. M. (ed.). Studies in Relational Grammar 1:360-403. Chicago: University of Chicago Press. Babby, Leonard. 2000. Infinitival existential sentences in Russian: A case of syntactic supple-tion. King, T. H. and I. A. Sekerina (eds.). Formal Approaches to Slavic Linguistics: The Philadelphia Meeting 1999:1-23. Ann Arbor: Michigan Slavic Publications. Bhatt, Rajesh. 1998. Obligation and possession. Harley, H. (ed.). Papers from the UPenn/MIT Roundtable on Argument Structure and Aspect. MIT Working Papers in Linguistics 32:21-40. Cambridge, Mass: MITWPL. Caponigro, Ivano. 2001. On the semantics of indefinite free relatives, van Koppen, M., J. Sio & M. de Vos (eds.). Proceedings of ConSOLE X:49-62. Leiden: SOLE. Chomsky, Noam. 2000. Minimalist inquiries: The framework. Martin, R., D. Michaels & J. Uriagereka (eds.). Step by step: Essays on minimalist syntax in honor of Howard Lasnik: 89-155. Cambridge, MA: The MIT Press. 44 Dotlačil, Jakub. 2005. Non-local binding in Slavic languages and restructuring. Blaho, S., L. Vicente & E. Schoorlemmer (eds.). Proceedings of ConSOLE XIII:1-16. Leiden, Netherlands: SOLE. —. In progress. Why clitics cannot climb out of CP. Grosu, Alexander and Fred Landman. 1998. Strange relatives of the third kind. Natural Language Semantics 6:125-170. Izvorski, Roumyana. 1998. Non-indicative wh-complements of possessive and existential predicates. Tamanji, P. N. & K. Kusumoto (eds.). Proceedings of NELS 28:159-173. Amherst, MA: GLSA. Junghanns, Uwe. 1999. Generative Beschreibung periphrastische Konstruktionen des Tschechischen. Anstatt, T., R. Meyer & E. Seitz (eds.). Linguistische Beiträge zur Slavistik aus Deutschland und Osterreich. VII. Jungslavistlnnen-Treffen. Tübingen/Blaubeuren 1998: 133-165. München: Sagner. Lenertová, Denisa. 2004. Czech pronominal clitics. Journal of Slavic Linguistics 12: 135-171. Pancheva Izvorski, Roumyana. 2000. Free relatives and related matters. Ph.D. thesis. University of Pensylvania. Rappaport, Gilbert. 1986. On a persistent problem of Russian syntax: Sentences of the type Mne Negde Spat7. Russian Linguistics 10:1-31. Rezac, Milan. 2004. Elements of cyclic syntax: Agree and Merge. Ph.D. thesis. University of Toronto, Toronto. —. The syntax of clitic climbing in Czech, http://minimalism.linguistics.arizona.edu/AMSA/PDF/AMSA-202-0602.pdf. Rudin, Catherine. 1986. Aspects of Bulgarian Syntax: Complementizers and Wh-Constructions. Columbus, OH: Slavica. Wurmbrand, Susi. 1998. Infinitives. Ph.D. thesis, MIT. 45 Only, bound variables and VP ellipsis in Czech* Mojmír Dočekal docekal@phil.muni.cz 1 Introduction This paper presents evidence that a possessive anaphor svuj in Czech can be interpreted either as a bound variable (which is unproblematic) or as a covaluated pronoun.1 The distinction between bound variables and covaluated expressions surfaces in Czech sentences with the focus sensitive particle jen/pouze (which corresponds to the famous only in English), but disappears in the VP ellipsis contexts. This is a bit puzzling, because VP ellipsis and only are both contexts where the distinction between bound variables and covaluated pronouns can be detected, at least in English. In the first sections I will examine the syntax and semantics of VP ellipsis and then I will investigate the behavior of Czech pronouns and anaphors in the context of VP ellipsis and the jen/pouze expressions. Let us introduce basic data. Czech does not have the equivalent of VP-ellipsis (VPE) in English. VPE with past tense auxiliaries and conditional auxiliaries is ungrammatical in Czech: Mary is leaving and John is [yp e] too. Mary hasn't left, and John has [yp e]. Past tense auxiliaries *Ja jsem odesel a ty jsi taky. I AUX leave and you AUX too 'I left and you did too.' Conditional auxiliaries *Ja bych odesel a ty bys taky. I COND leave and you COND too 'I would leave and you would too.' VP-ellipsis in Czech is grammatical only with lexical or semi-lexical verbs, like a future auxiliary, or a copula: (3) a. Lexical verb: *The present study is a part of the project GACR 405/07/P252, the goal of which is an analysis of the Czech negation and negative polarity items. 1See Reinhart (2000) for the distinction between bound variables and covaluated pronouns. (1) (2) a. b. 47 Karel umí odejít, ale Petr neumí. Karel knows leave but Petr neg-knows 'Karel know how to leave, but Petr does not' b. Future auxiliary: Karel bude odcházet, ale Petr nebude. Karel will leave but Petr neg-will 'Karel will leave, but Petr will not.' c. Copula: Karel je učitel, ale Petr není. Karel is teacher but Petr neg-is 'Karel is a teacher, but Petr is not.' This is exactly the difference between high verbal forms (conditional auxiliaries, past tense auxiliaries) and low verbal forms (main finite verbs, future auxiliaries), as proposed in Veselovská (2004). So Czech is somewhere between English, with full VPE, and Romance languages, where VPE is not possible with any auxiliary (see Zagona (1982) and Busquets 2005). 2 Syntax of VPE: stripping and VPE According to Hankamer and Sag (1976: 409) stripping is a rule that deletes everything in a clause under identity with corresponding parts of the preceding clause (the correlate), except for one constituent (the remnant). The distinction between VPE and stripping is that VPE is the ellipsis of the whole verb phrase, which includes the verb and any objects or modifiers it might take. VP ellipsis is licensed by the by the immediately preceding auxiliary (in English) or by the polarity particles (ne, taky, ano in Czech). (4) shows examples of stripping and (5) shows examples of VPE. (4) a. Petr dal knihu Karlovi, ale taky Marii. Petr gave book Karel but too Mary 'Petr gave a book to Karel, but also to Mary.' b. Petr políbil Marii, ale ne Karla. Petr kissed Mary but not Karel 'Petr kissed Mary, but not Karel.' (5) a. Petr dal knihu Karlovi, ale Marie ne. Petr gave book Karel but Mary not 'Petr gave a book to Karel but Mary did not.' b. Petr políbil Marii a Karel taky. Petr kissed Mary and Karel too 'Petr kissed Mary and Karel did too.' The fundamental properties of stripping in Czech are the following (see McShane 2000 and Busquets 2005): 1. Unlike VPE, stripping is not allowed in subordinate clauses: (6) Stripping: a. *Petr dal knihu Karlovi, protože ne Kláře. Petr gave book Karel because not Klára 48 'Petr gave the book to Karel because not to Klara.' b. *Petr políbil Marii, protože ne Karla. Petr kissed Mary because not Karel 'Petr kissed Mary because not Karel.' (7) VPE: a. Petr dal knihu Karlovi, protože Klára taky. Petr gave book Karel because Klara too 'Petr gave the book to Karel because Klara did too.' b. Petr políbil Marii, protože Karel taky. Petr kissed Mary because Karel too 'Petr kissed Mary because Karel did too.' 2. VPE is a constituent operation, stripping is not. 3. Unlike VPE, stripping does not appear to conform to the backward anaphora constraint (i. e. the remnant cannot precede the correlate): (8) a. VPE: Přestože Karel neumí, Petr umí hrát na piáno. Although Karel neg-knows Petr knows play on piano 'Although Karel does not know, Petr does know how to play the piano.' b. Stripping: *Ale ne Kláře, Petr dal knihu Karlovi. But not Klara, Petr gave book Karel 'Although not to Klara, Petr gave the book to Karel.' 4. According to Busqutes (2005), stripping is constrained by its interaction with focus, but this is not the case for VPE: (9) a. Stripping: Petr dal květiny [F Kláře], ale ne Karlovi. Petr gave flowers Klara but not Karel 'Petr gave the flowers to Klara but not to Karel.' b. #Petr dal květiny [F Kláře], ale ne knihu. Petr gave flowers Klara but not book 'Petr gave flowers to Klara, but not the book.' (10) VPE: Petr dal květiny [F Kláře], ale Karel ne. Petr gave flowers Klara but Karel not 'Petr gave the flowers to Klara, but Karel did not.' This is not exactly true in Czech, as we will see later. But the restriction, which was formulated in Busquets (2005: 14) for Catalan, holds also in Czech: "The focused expression constructs a set of alternatives C, if the remnant does not belong to C, then the sentence is infelicitous. By contrast, the position of focus in TP-Ellipsis is not a necessary condition in order to recover the missing VP. " In Czech, as in Romance languages, the VPE can be licensed by polarity particles. For Catalan Busquets 1997 points to following examples: 49 (11) En Pere anirá a Paris, i la Marie també. 'Pere will go to Paris and Maria (will) too.' (12) El Rafel va convidar el Ricard al teátre, pero al cinema no 'Rafel invited Ricard to the theatre, but (he didn't) to the movies not.' In Czech polarity particles similar to Catalan també, no can be found. Their role is analogous to English auxiliaries in VPE: (13) a. Petr přisel, ale Karel ne. Petr came but Karel not 'Petr came, but Karel did not.' b. Petr přisel a Karel taky. Petr came and Karel too 'Petr came and Karel did too.' There is sometimes proposed a dedicated functional projection for the polarity particles (see Laka 1990) which is also used for focus licensing. Laka (1990) postulates two different Polarity phrases (SP), one between VP and TP and one between TP and CP. If we assume that Czech polarity particles in VPE are located in the lower SP, then we predict the difference between high verbal forms and low verbal forms in their ability to occur in VPE. High verbal forms are base generated in T and low verbal forms are base generated somewhere lower, e.g.. in v (future auxiliary) and in V (main verbs). The exact location of low verbal forms is irrelevant for the purposes of this paper; it is sufficient to assume that low verbal forms are base generated below SP. This is of course the same phenomena as the negation test used in Veselovská (2004). The negation test distinguishes between low verbal forms and high verbal forms - the later can bear negation, the former cannot. (14) a. *Já nejsem přišel. I neg-AUX came T did not come.' b. Já jsem nepřišel. I did neg-came T did not come.' (15) a. *Já nebych přišel. I neg-AUX-COND came T did not come.' b. Já bych nepřišel. I AUX-COND neg-came T would not come.' (16) a. Já nebudu chodit. I neg-AUX come I will not come, b. *Já budu nechodit. I AUX neg-come T will not come.'2 2 This is ungrammatical in the sentential negation reading. 50 (17) a. Já nejsem student. I neg-am student T am not a student.' b. * Já jsem nestudent. I am neg-student T am not a student.' If we assume head movement through SP for negated verbs, then high verbal forms cannot be negated, because the T head is above S head thoughout a derivation. 3 VPE Czech pronouns displays a strict/sloppy identity ambiguity in VPE. Compare the standard case shown in (18), which can be interpreted in one of two ways: either John's mother went to Peter's house (strict) or she went to John's house (sloppy). On the other hand, Czech anaphors (possessive or not) usually involve only sloppy identity. Compare the data in (19). (18) Petrova matka šla do jeho domu a Honzova taky. Peter's mother went to his house and John's too 'Peter's mother went to his house and John's did too.' (19) Petrova matka šla do svého domu a Honzova taky. Peter's mother^ went to her^ house and John's too 'Peter's mother^ went to her^ house, and John's did too.' [sloppy/*strict identity] This is quite unproblematic if we assume two distinct procedures for pronoun resolution: binding and covaluation (see Reinhart 2000). In the first, we close the property: a common technical implementation is that the variable gets bound by the A-operator. In the second, the free variable is assigned a value, say, from the discourse storage. Bound variables in natural language must be c-commanded by their antecedents, but covaluation is a pragmatic phenomena, so it is constrained by saliency and other pragmatic notions. From this it follows that a intersentential pronoun dependency must always be covaluation, but in one sentence there are two possibilities: either binding or covaluation. (20) is ungrammati-cal, because quantifiers cannot refer, so covaluation is forbidden, but binding is forbidden also, because there is a sentence boundary between the quantifier and the pronoun. In contrast, (21) is grammatical, because the quantifier c-commands the pronoun. (20) [Každý člověk]i je smrtelný. *Oni je také racionální. '[Every man]i is mortal. *Hei is also rational.' (21) [Každý člověk]x doufá, že hoi ostatní chápou. '[Every man]i hopes that the others understand himi.' The pronoun ho in (21) must be interpreted as a bound variable, because the quantifier cannot refer, so the covaluation option is impossible. In contrast, the pronoun on in (22) must be interpreted as a covaluation for the exactly opposite reasons. (22) [Ten člověk] x je smrtelný. Onx je také racionální. '[The man]i is mortal. Hei is also rational.' 51 4 Possessives A well known disambiguator for the contrast between binding and covaluation is the focus particle only as in (23), where distinct truth conditions result from the two interpretations of pronouns. Intuitively (23-a) means that Karel is the only person in the universe of discourse who thinks that his dog is amazing. In contrast, (23-b) means that Karel is the only person in the universe of discourse who thinks that Karel's dog is amazing. (23) Jen Karel si myslí, že jeho pes je úžasný. Only Karel thinks that his dog is amazing a. binding: Only Karel (Ax ( x think, that x's dog is amazing)); b. covaluation: Only Karel ( Ax ( x think, that y's dog is amazing) & y = Karel) In English the pronoun his can also have the mentioned interpretations. According to Heim & Kratzer (24) has three following structures. (25) is a tree which corresponds the binding semantics for his: John after quantifier raising yields the index 1, which is the H&K system translatable as the A abstraction with the variable corresponding to the index 1 and because he has the same index, the semantics is straightforward: John (Ax ( x loves x's father)). For the covaluation option there are two possibilities: (i) either John does not undergo quantifier raising (QR), so there is no A abstraction and the pronoun remains free - this is (26), or (ii) John undergoes QR, but the indexes are different - as in the tree (27). Both trees correspond to the covaluation reading: John ( Ax ( x loves y's father) & y = John), where the "& y = John" part of the formula is supplied from the pragmatics, but not from the semantics. (24) John loves his father. (25) Heim & Kratzer (1998: 246) DP S John 1 S DP VP ti V DP loves the NP DP N hei father 52 (26) DP VP John V DP loves the NP DP N hei father (27) DP S John 2 S DP VP t2 V DP loves the NP DP N hei father Now the hypothesis is that for Czech possessive anaphors, only the tree (28) is viable, because the binding theory demands it - after all svůj is an anaphor, so there is nothing really unexpected. English his is a pronoun, thus can take its antecedent from the discourse storage, which is not an option for the anaphor. 53 (28) DP S Petr 1 S DP VP ti V DP miluje D NP 0 DP N svehc-i otce Now let us return back to VPE. We can account for the data in (18) and (19), if we assume that Czech pronouns can be interpreted either as bound variables or as covaluated free variables; but Czech anaphors can be interpreted only as bound variables. (18) can have three interpretations: 1. strict identity to Peter (covaluation); 2. strict identity to John (covaluation); 3. sloppy identity (binding). The fourth interpretation of (18) (Peter's mother went to John's house and John's mother went to Peter's house) is ungrammatical. This follows from any theory of VP ellipsis which states that the ellipsis is dependent on identity in LF, where identity means either identity of binding or covaluation (if the antecedent is interpreted as bound, then the elided constituent is too and the same for the covaluation) - see Fiengo & May (1994) and Busquets (1997). The LF identity of elided material with its antecedent can be seen independently from (29), where the non-elided VP has two possible interpretations: (29-a) or (29-b), but if we chose (29-a), then we must chose the wide scope interpretation of DP nějaký dárek in the elided VP also. And the same holds for (29-b). (29) Klára dala nějaký dárek každému návštěvníkovi a Bára taky. 'Klára gave a present to every visitor and Bara did too.' a. 3x(present'(x) DVy(visitor'(y) D give'(Klara',x,y))) b. Vy(visitor'(y) D 3x(gift'(x) D give'(Klara',x,y))) The account of the LF identity of the elided VP with its antecedent holds also for the bound/covaluation distinction. This is confirmed by the following data. I have asked 20 Czech native speakers for their grammaticality judgments for (30) and the results are the following: 54 1. Strict reading, denotation of both pronouns is Karel: 50% of participants. 2. Strict reading, denotation of both pronouns is Petr: everybody agreed that it is a possible interpretation, but for nobody was it the first reading. 3. Sloppy reading: 50% of participants. 4. Strict reading, denotation of the non-elided pronoun as Petr and denotation of the elided pronoun as Karel: impossible reading for everybody. (30) Karlova matka uklízí jeho knihovnu, ale Petrova ne. 'Karel's mother cleans his library, but Peter's does not.' The explanation is straightforward, if ellipsis must preserve the binding/covaluation distinction: 1. For the covaluation reading the referential indexes must be identical (reading 1 and 2); this is translated as the identity of variables: a) reading 1: Karel ( Ay (y's mother (Ax (x cleans z's library) &; z = Karel))) &; Petr ( Ay (y's mother (Ax (x -cleans z's library) &; z = Karel))) b) reading 2: Karel ( Ay (y's mother (Ax (x cleans z's library) &; z = Petr))) &; Petr ( Ay (y's mother (Ax (x -cleans z's library) &; z = Petr))) 2. For the binding reading (reading 3) the indexes can be different, but the A abstraction of the elided VP must be the alphabetical variant of its antecedent: Karel (Ay (y's mother (Ax ( x cleans y's library)))) &; Petr (Aw (w's mother (Az ( z -cleans w's library)))) 3. The fourth reading is forbidden, because the covaluation is not possible and binding is not possible either, since the DP Petrova matka does not c-command the first possessive pronoun. Discourse Representation Theory (DRT, see Kamp &; Reyle 1993) offers quite sensible formalization for these facts. VPE is naturally treated here as a subtype of anaphora (see Klein 1987), where VP as a predicate is represented by a separate subDRS, where a distinguished discourse referent represents A-bound variable. This anaphoric treatment of VPE is quite natural, because it predicts some of the properties of VPE: 1. Impossibility of VPE in subordinate sentences: this follows from the anaphoricity of VPE, because this would lead to inappropriate self-embedding (the antecedent would contain the elided constituent). 2. Pragmatic summation of two or more VP into antecedent for elided VP: Petr became ill. He was taken to hospital. Karel did too. In this DRT theory of VPE, for example (30), can be represented as (31), where fx4J is a distinguished discourse referent, P is the subDRS, where fx4J is the distinguished discourse referent. Conditions (i) and (ii) represent excluding alternatives - only one of them can be chosen, if (i) is chosen, then the sloppy reading results, if (ii) is chosen, then the strict reading emerges. 55 (31) x xi y P Q karel(x) mother(y) r(x,y) P(x) [x4] x3 x2 library (x3) P- r(x3,x2) i . clean (x4,x3) (i) x2=x4 (ii) x2 = x petr(xi) do(xi,Q) Q=P Realizations for each individual case: 1. The first reading (strict denotation to Karel): (32) xxi y P Q karel(x) mother(y) r(x,y) P(x) [xA] x3 x2 library (x3) P: r(x3,x2) clean (x4,x3) (ii) x2 = x petr(xi) do(xi,Q) Q=P 2. The second reading (strict denotation to Petr): 56 (33) x xi y P Q karel(x) mother(y) r(x,y) P(x) [x4] x3 x2 library (x3) P: r(x3,x2) clean (x4,x3) (ii) X2 = Xi petr(xi) do(xi,Q) Q=P 3. The third, sloppy reading: (34) xxi y P Q karel(x) mother(y) r(x,y) P(x) [xA] x3 x2 library (x3) P: r(x3,x2) clean (x4,x3) (i) x2=x4 petr(xi) do(xi,Q) Q=P Interestingly sometimes even Czech possessive anaphors can behave as covaluated, as can be seen from (35) and its two possible continuations. Only in Czech also seems to change the argument structure: -able adjectives are usually taken as agent absorbing, but with only the agent is again possible - see (36-b). 57 (35) Jen prof. Novák nedoporučuje své knihy, ale ostatní profesoři své/Novákovy Only prof. Novak neg-recommends his books, but other professors their/Novak's knihy doporučují. books recommend 'Only prof. Novak does not recommend his books, but other professors do.' (36) a. *Ten brouk je pozorovatelný Karlem. The beetle is observable by Karel b. Ten brouk je pozorovatelný jen Karlem. The beetle is observable only by Karel The central idea is that only in Czech triggers a background-presupposition rule which states that whenever focusing gives rise to a background Ax.0(x), there is a presupposition to the effect that Ax.0(x) holds of some individual. This can explain (36-b) (if we take only A B as ->3x[^Ax A Bx], where B is presupposed). But for the purposes of this article the most important point is that (35) shows that svůj is interpretable either as a bound variable or as a covaluated expression. This is quite unexpected, because svůj is an anaphor. I assume that the covaluation reading of the Czech possessive anaphor stems from its smaller ^-deficiency (in the sense of Reuland &; Reinhart 1995) in comparison with the 0-deiiciency of other Czech anaphors. Other Czech anaphors cannot be interpreted as covaluated expressions - anaphors sobě/se in (37) and (38) can be interpreted only as bound variables. (37) Jen prof. Novák vykládá pořád o sobě. 'Only prof. Novak always talks about himself.' a. [... ] ale ostatní profesoři vykládají i o lingvistice. '[...] but other professors talk also about linguistics.' b. ..] ale ostatní profesoři nevykládají o prof. Novákovi. '[...] but other professors do not talk about prof. Novak.' (38) Jen prof. Novák se pořád chválí. 'Only prof. Novak praises himself.' a. [... ] ale ostatní profesoři chválí i jiné lingvisty. '[...] but other professors praise also other linguists.' b. .. ] ale ostatní profesoři nechválí prof. Nováka. '[...] but other professors do not praise prof. Novak.' 5 Only I suppose that the distinguishing property of only can be traced back to its semantics. This section will be dedicated to an incomplete but sufficient (for our purposes) story of only. I assume the familiar alternative semantics framework (Rooth 1992 and many others). Only is the focus sensitive particle which can (as other focus sensitive particles) alter the truth conditions of a sentence with the presentational focus. Compare (39-a) and (39-b) - only the later sentence has other truth conditions than its non-focused counterpart. In the (39-a) focus evokes the set of alternatives to Mary, but its truth-conditions are the same as its non-focused counterpart. But (39-b) means that Mary was the only girl (from the set of alternatives), which was kissed by Peter. 58 (39) Petr políbil Marii. 'Petr kissed Mary.' a. Petr políbil [F Marii]. 'Petr kissed Mary.' b. Petr políbil pouze [p Marii]. 'Petr kissed only Mary.' I do not discuss contrastive focus like (40), where it is sometimes claimed that similar truth-condition effects which in (39) stem from the focus sensitive particle are caused solely by the contrastive focus. (40) [F Marii] políbil Petr. 'Mary, kissed Petr.' The classical semantics for the only particle is the reverse of implication or inclusion. (41) Karel will come, if it will rain. p ... it will rain, q ... karel will come p ->■ q (42) Karel will come, only if it will rain. q ->■ P (43) Teachers read poetry. Ax ... teachers, Bx ... poetry readers Vx[Ax Bx] (44) Only teachers read poetry. Vx[Bx Ax] But this is not the whole story. Van der Sandt & Geurts (2004) claim that only should be analyzed as a weak quantifier along the lines of (45). (45) only A are B = -n3x[->Ax A Bx] This is of course truth-conditionally equivalent to the strong-quantifier analysis, but it makes some interesting predictions. The first is the scope of negation: in the weak quantifier analysis the negation has scope over the whole formula, so it should license the NPIs in the whole sentence, unlike in the analysis with the strong quantifier, where the downward entailing ability of the strong quantifier is limited to its first argument.3 This is corroborated by the Czech data in (46)- the NPIs vůbec and kdy are in boldface. (46) a. Všichni, kdo vůbec kdy navštívili Brno, ho milují. Everybody who at_all ever visited Brno it love 'Everybody who ever visited Brno loves it.' b. *Všichni, kdo milují Brno, navštívili vůbec kdy jiné město. Everybody who love Brno visited at_all ever another town 'Everybody, who loves Brno, ever visited another town.' (47) a. Pouze ti, kdo vůbec kdy navštívili Brno, ho milují. Only the_people who at_all ever visited Brno it love 3V is downward entailing (entailing goes from set to its subsets) on its first argument and upward entailing (entailing goes from set to its supersets) on its second argument - see (46). 59 'Only the people who ever visited Brno love it.' b. Pouze ti, kdo milují Brno, navštívili vůbec kdy Horní Heršpice. Only the_people who love Brno visited at_all ever Horni Herspice 'Only the people who love Brno ever visited Horni Herspice.' However, if the weak quantifier analysis is on the right track, then there is a problem with the presupposition of sentences like (48). (48) presupposes that there was somebody else than Petr, who visited Brno. From the strong-quantifier analysis of only this presupposition follows straightforwardly - if the strong quantifier presupposes (as is commonly assumed) existential quantification of its first argument (see (49)). (48-a) is the most salient reading of (48), where the negation scopes over the focused element, and in the strong-quantifier analysis it correctly predicts the existential presupposition of the set of Brno visitors. But (48-b), where the first negation before Petr(x) comes from the negation and the second from the rule for only, does not presuppose the non-emptiness of the set of Brno visitors. (48) Brno did not visit only [F Petr]. a. \/x[BrnoVisitor(x) —► ->Petr(x)] b. ->3x[-i-iPetr(x) A BrnoVisitor(x)] (49) All students came. For these purposes we can use the Background/Presupposition Rule from Geurts & van der Sandt 2004 as can be seen in (50). (50) The Background/Presupposition Rule (BPR) Whenever focusing gives rise to a background \x.visit(x), and then (50) predicts that the existential presupposition will arise. (51) Only Petr did not visit [p Brno]. (52) X Petr(x) v u v visited Brno u^x u visited Brno After the global accommodation of the presupposition we have the following DRS, which is equivalent to the predicate logic formula in (54). For details of presupposition projections see van der Sandt & Geurts (2004). 60 (53) X v Petr(x) v visited Brno u u^x u visited Brno (54) 3xv[Karel(x) A VisitedBrno(v) A ->3u[u ^ x A VisitedBrno(u)]] Now let us return to (35), repeated below as (55). For this example the presupposition analysis means that there are two options for the accommodation of the presupposition: 1. Presupposition Ax(x recommends Novak's books), holding of some individual, is accommodated globally - this presupposition is then picked up in the second sentence in its reading where other professors recommend Novak's books. 2. Presupposition Ax(x recommends Novak's books), holding of some individual, is accommodated locally - in the scope of negation, so there is no presupposition that somebody recommends Novak's books, so therefore the sloppy reading in the second sentence arises. (55) Jen prof. Novák nedoporučuje své knihy, ale ostatní profesoři své/Novákovy Only prof. Novak not-recommends his books, but other professors their/Novak's knihy doporučují. books recommend 6 Solution In this section I will propose the solution for the puzzle of the interpretation of bound variables in Czech VPE. I will summarize the facts: the Czech anaphor svůj can be interpreted either as a bound variable or as a covaluated expression; this was shown with the help of the focus sensitive particle jen ('only'). The covaluation reading disappears in the VPE context, but shows up again if the VPE contains jen. The solution can be found in the alternative semantics framework (Rooth 1992). The idea of my proposal is to compare the entailing capability of the discussed sentences. (56-a) is stronger than (56-b), which means that (56-b) implies (56-a). In other words: the set of those people who love somebody's library is the super-set of those people who love their own library: if you love your own library, then you love somebody's library, but not the other way around. This pattern of reasoning is an example of Gricean quantity implicature and can be generalized using the notion of a scale of alternative assertions. Since my loving my library implies my loving somebody's library, but not conversely, we can set up a partially ordered set of two open propositions ordered by entailment: {love(x,x's library), love(x,y's library)}, the ordering relation > is entailment. (56) Petr miluje svou knihovnu. Petr loves his library a. Petr (Ax ( x loves x's library)) b. Petr (Ax ( x loves y's library) &; y = ???) 61 If we look at the focusing pattern in VPE, we see that there is focus on the polarity particle, which is sometimes dubbed polarity focus. The focus of this type is motivated by the suggested contrast, for instance between Karel and Petr. I follow Rooth (1992) with his rule applying to contrastive focus - see (57). (57) Contrasting Phrases. Construe a phrase a as contrasting with a phrase (3, if \\/3\\° G IMI7- 1|° denotes ordinary semantic value of (3 (if (3 is Petr, then \\/3\\° is Petr, if (3 is love then \\/3\\° is the appropriate relation, ...), \\a\\* denotes a focus semantic value for the a, which in the case of a sentence is a set of propositions. For instance, the focus semantic value for the sentence [Karel]f loves Petr is the set of propositions of the form 'a; loves Petr', while the focus semantic value for Karel loves [PetrJF is the set of propositions of the form 'Karel loves x\ For (58) the Contrasting Phrases Rule, (3 is the elided VP, a is its antecedent VP and the Contrasting Phrases Rule states that \\Karel(\x(xlovesx'slibrary)\\° G \\\x(x loves x's library\\*, which is of course fulfilled. (58) a. [Petr]F miluje svou knihovnu a Karel [taky]F. Petr loves his library and Karel too b. [Petr]i? miluje svou knihovnu, ale Karel [ne]i?-Petr loves his library but Karel not This rule has the right predictions, because in (59) we do not have a sloppy reading in the second conjunct - this sentence must mean that Karel does not love Mary's library. This follows from the (57), because \\Karel(\x(x loves x's library)\\° ^ \\\x(x loves Mary's library\\*, but ||Karel{\x[x loves Mary's library) ||° G ||\x(x loves Mary's library]]*'. (59) [Petr]F miluje Mariinu knihovnu, ale Karel [ne]F. Petr loves Mary's library but Karel not But (57) cannot by itself predict why (60-a) has only the sloppy reading. For this we need something stronger. The basic idea is that the listener selects the lowest element from the scale of the open propositions. (60) a. Petr miluje svou knihovnu a Karel taky. Petr loves his library and Karel too Petr (Ax ( x loves x's library)) & Karel (Ay (y loves y's library)) b. Petr (Ax ( x loves x's library)) & Karel (Ay (y loves y's library)) In discussing example (61), Rooth (1992) says that asserting an element 0 of C implicates the negation of any higher element in the scale, that is any ip such that ip >c 4> and 0 ^ il>- So in this case, if the underlying set C is partially ordered as {ace(m),pass(m)}, then asserting 'Mats passed' implicates the negation of 'Mats aced'. (61) Well, I [passed]F. In the (60-a) the scale is {love(x, x's library), love(x, y's library)} and the process goes the other way round than in Rooth's example. Asserting polarity of VP implicates the negation of any lower element of the scale, which in this example is love(x,y's library). On the other hand svuj in (62) can have both covaluated and bound interpretations. This follows from the possible interpretations for the first sentence - see the last section - and either 62 of the two can be picked up by the following VPE. This is also circumstantial evidence for the pragmatic restrictions on VPE. (62) Pouze Karel miluje svou knihovnu, ale Petr ne. Only Karel loves his library but Peter not Summary: the Czech anaphor svůj can be interpreted either as a bound variable or as a covaluated expression. But the covaluation reading disappears in the VPE context. This follows from the Contrasting Phrases Rule and from the scale of the open propositions, which emerges as a consequence of focusing the antecedent VP for VPE. From this scale the listener chooses the highest element, which in this case is the bound variable reading. References Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse. Dordrecht: Kluwer. Busquets, Joan. 1997. A Discourse-Based Approach to VPE : A Survey. Research report RI IRIT/97-48-R, 1997, IRIT Universitě Paul Sabatier, Toulouse. Busquets, Joan. 2005. Stripping vs VP-Ellipsis in Catalan. Research report RR5616, 2005, INRIA, Universitě Bordeaux-1, LaBri. Fiengo, Robert and Robert May 1994. Indices and Identity. Cambridge, MA: MIT Press. Geurts, Bart. 1999. Presuppositions and Pronouns. Oxford: Elsevier. Geurts, Bart and Rob van der Sandt. 2004. Interpreting Focus. Theoretical Linguistics 30:1-44. Hankamer, Jorge and Ivan Sag. 1976. Deep and Surface Anaphora. Linguistic Inquiry 7(3):391-426. Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer. Klein, Ewan. 1987. VP ellipsis in DR theory. Groenendijk, J. & D. de Jongh & M. Stokhof (eds). Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers: 161-187. Dordrecht: Foris Publications. Kučerová, Ivona. 2005. The T-Extension Condition. J. Alderete et. al. (eds.): Proceedings of the 24th West Coast Conference on Formal Linguistics: 227-235. Somerville, MA: Cascadilla Proceedings Project. McShane, Marjorie J. 2000. Verbal ellipsis in Russian, Polish, and Czech. Slavic and East European Journal 44(2): 195-233. Reuland, Eric and Tanya Reinhart. 1995. Pronouns, Anaphors and Case. Haider, H., S. Olsen &; S. Vikner (eds.): Studies in Comparative Germanic Syntax: 241-268. Dordrecht: Kluwer. Reinhart, Tanya. 1983: Anaphora and Semantic Interpretation. London: Croom Helm. Reinhart, Tanya. 2000. Strategies of Anaphora Resolution. Bennis, H., M. Everaert &; E. Reuland (eds.): Interface Strategies: 295-324. Amsterdam: North Holland Amsterdam. Veselovská, Ludmila. 2004. Rozšířená verbální projekce v češtině: tři druhy slovesa být. Hladká, Z. &; P. Karlík (eds.): Čeština - univerzália a specifika 5:203-213. Praha: Nakladatelství Lidové noviny. Zagona, Karen. 1982. Government and Proper Government of Verbal Projections. Doctoral dissertation, University of Washington. 63 Clitic omission in Czech as across-the-Board extraction Jakub Dotlačil Jakub.Dotlacil@let.uu.nl 1 Introduction In the tradition of Slavic generative linguistics, one of the main discussions concentrate on the second position clitics and a question of how to account for their position. Should we blame syntax, phonology, or a mixture of the two? The battlefield of various approaches to the second position of clitics was mostly Serbian or Croatian (see, for example, Schutze 1994 for a phonological approach, Progovac 1996 for a syntactic approach, or Boskovic 2001 for a mixture of the two). On the other hand, the literature on similar issues in Czech was surprisingly unequivocal. As far as I know, linguists applied a purely syntactic approach to the second position in Czech (see, for example, Rivero 1991, Veselovská 1995, Lenertová 2004). In this paper, I am going to discuss an argument from clitic omission in conjunction which goes against a purely syntactic account of clitic placement in Czech and supports Boskovic's analysis of clitic positioning: namely, their surfacing in the second position is a result of an interplay of both, syntax (which takes care of clitics' movement) and phonology (which takes care of pronunciation of the right copy) (see Boskovic 2001). To have a taste of the argument, take a look at the following examples. In coordination, Czech allows for clitic omission in one of the conjuncts, as shown on two examples: (1-a), in which the clitics jsem and ho are omitted in the second conjunct, and (1-b), in which the clitic se is omitted in the second conjunct.1 (1) a. Já jsem ho zavolala představil známým. I past-aux^Sg himacc called and introduced friends 'I called him and introduced him to friends.' b. Petr sej umyl a oholil. Petr refl washed and shaved 'Petr washed and shaved.' As I am going to argue, cases like (1-a) and (1-b) should be analyzed as cases of Across-the-Board (ATB) extraction. ATB extraction is, descriptively speaking, movement of like-phrases out of the coordination. (2) is an example of ATB wh-movement. 1AU clitics are boldfaced. All glossed examples are from Czech. 65 (2) What did you buy and read? The wh-word what starts as an argument of both buy and read. Later on, it moves out of the VP coordination (i.e., undergoes ATB-extraction) into its specifier position (Spec, CP). The final structure with traces after movement is sketched in (3): (3) CP whatj TP did DP you ConjP VP Conj VP V~DP and V^DP buy tj read t; Similarly to the wh-word in example (2), I am going to argue that in examples (1-a) and (1-b), clitics originate in both conjuncts. Later on, they move out of the coordination (i.e., undergoing ATB extraction) into their specifier position. Somewhat surprisingly, I will show that we can find cases similar to (1-a) and (1-b) (i.e., cases of ATB extraction of clitics), in which, crucially, clitics are pronounced not outside of the coordination but inside the first conjunct. In other words, we will deal with a special case of ATB extraction: one in which ATB extracted constituents do not surface outside of the coordination but inside the first conjunct. This, I am going to argue, provides an argument for a particular view on the second position clitic placement. In particular, it is compatible with Boskovic's analysis of clitic placement but problematic for other, purely syntactic, accounts. The paper is organized in the following way. Section 2 discusses arguments for ATB extraction analysis of (1-a) and (1-b). Section 3 discusses cases of ATB extraction of clitics in which clitics surface inside the first conjunct. Section 4 focuses on some problems with the presented analysis of clitic placement in Czech. Section 5 concludes. 2 Clitic omission as ATB movement In Czech, clitics can be omitted in one of the conjuncts, as shown in (4), repeated from above: 66 Já jsem ho zavolal a představil známým. I past-aux^Sg hiniacc called and introduced friends T called him and introduced him to friends.' The clitic omission in the second conjunct, one might suggest, is a result of an ellipsis that targets clitic clusters. However, an ellipsis analysis runs into immediate problems. First, notice that clitic omission is possible only in coordinations. It is ungrammatical in a non-coordinated sequence of clauses (5-a). Furthermore, clitic omission is impossible in the matrix clause if the anteceding clitics appear in a subordinate clause (5-b) or if the anteceding clitics appear in the matrix clause and the clitic cluster is deleted in a subordinate clause (5-c). (5) a. * Já jsem ho zavolal. Potom představil známým. I past-aux^Sg himacc called. Then introduced friends 'I called him. Then I introduced him to friends.' b. *Poté co jsem ho zavolal představil známým. After what past-aux^gg himacc called introduced friends 'After I had called him I introduced him to friends.' c. *Představil jsem ho známým poté co zavolal. introduced past-aux^gg himacc friends after what called 'I introduced him to friends after I had called him.' The fact that clitic omission is licensed only in coordination would seem arbitrary and surprising if we assumed that clitic omission is a result of an ellipsis of clitics.2 However, if we assumed that clitic omission is a result of ATB extraction of clitics the difference between (4) and (5-a)-(5-c) would fall out automatically. This is so because ATB extraction (a label for the phenomenon in which a constituent that is understood to originate in all the conjuncts surfaces outside of the conjunction (see (2))) occurs only in coordinations. Thus, based on the difference between (4) and (5-a)-(5-c) I suggest that (4) should be analyzed as (6): (6) I past-auxj himj ... [conjPbcP caUed tj tj ] and [^p introduced tj tj friends ]] The next three subsections are going to present an additional support for the analysis of clitic omission as ATB extraction. 2VP ellipsis, sluicing, or NP ellipsis can normally be licensed in contexts similar to (5-a)-(5-c). Gapping (verb deletion) is a surprising case: it can be licensed only in coordinations (cf. (i) and (ii)): (i) Marie políbila Petra a Táňa Tomáše. Marie kissed Petr and Táňa Tomáš 'Marie kissed Petr and Tanya - Tom.' (ii) ??Marie políbila Petra předtím než Táňa Tomáše. Marie kissed Petr before than Táňa Tomáš '*Marie kissed Petr before Tanya - Tom.' In fact, this has been used as an argument to treat gapping not as an ellipsis (contra Ross 1967) but as ATB extraction (Johnson, 2006). 67 2.1 Case matching It has been noted (see Dyla 1984, Franks 1993) that the constituent that undergoes ATB extraction must be assigned cases in conjuncts which have the same overt manifestation. To see the validity of this generalization, let us turn for a moment to ATB wh-movement. The following table summarizes nominative, genitive and accusative forms for Czech kdo 'who' and co 'what'. Notice that 'who' has a syncretic form in genitive and accusative, whereas 'what' has syncretic forms in nominative and accusative: Table 1: Case o: Case (7) nominative genitive accusative who who and what what kdo koho koho co čeho co Syncretism in forms corresponds to acceptability of ATB extraction. In case the first conjunct assigns nominative and the second conjunct assigns accusative, only 'what' is (marginally) acceptable (8-a). Similar examples with 'who' are ungrammatical ((8-b)-(8-c)). (8) 1st conjunct - nominative, 2nd conjunct - accusative a. ?Co tam leželo a ty jsi sebral? whatgen-acc there lay and you past-aux picked 'What lay there and you picked it?' b. *Kdo tam ležel a ty jsi sebral? whonom there lay and you past-aux picked 'Who lay there and you picked him?' c. *Koho tam ležel a ty jsi sebral? whoacc there lay and you past-aux picked 'Who lay there and you picked him?' This exemplifies the descriptive generalization, mentioned above: only 'what' can undergo ATB extraction in (8-a) because it has one and the same form for nominative and accusative, 'who' cannot undergo ATB extraction in this case because its forms for nominative and accusative differ. This descriptive generalization can be made sense of if we follow approaches to the architecture of the language which assume that syntax feeds lexical insertion.3 In the examples (8-a) to (8-c), the wh-words are assigned two cases in syntax (either by agreement projections inside each conjunct, or by the verb itself, or by combination of the two). When lexical insertion applies, the lexicon is searched for an item that matches the requirement of being an exponent of both cases (nominative and accusative). The paradigm of 'what' has such a lexical item (namely, co), unlike the paradigm of 'who'. Thus, in case of 'who', whatever lexical item is inserted it cannot satisfy morphological requirement and the ungrammaticality of (8-b) and (8-c) follows.4 3 Generative semantics was one such an approach. Recently, the same idea has been revived in Distributed Morphology. 4This reasoning only works if we do not assume that morphological forms in paradigms are always underspec-ified. 68 Notice that there is no reason to expect a similar requirement for ellipses. This is so because whereas in cases of ATB movement one and the same object originates in both conjuncts and is assigned two cases in each of the conjuncts, in cases of ellipses, we deal with two different objects: the antecedent and the ellided constituent. And, in fact, the head that assigns case to the antecedent and the head that assigns case to the ellided constituent do not need to assign case with the same overt manifestation. For example, a subject pro-drop can differ from its antecedent (9-a). The same holds for noun ellipses (9-b).5 (9) the antecedent - accusative, the ellipsis - nominative a. Petr měl rád Mariij. Hlavně když _j mu vařila. Petr had like Marieacc- especially when _ him cooked 'Petr liked Marie. Especially when she cooked for him.' b. Marie měla ráda černovlasé klukyj. Táně se líbili blonďatí _j Marie had like black-hair guysacc- Tanya refl appealed blondnom _ 'Marie liked blackhair guys. The blond ones appealed to Tanya.' Crucially, clitic omission patterns with ATB-wh-movement and unlike ellipsis. The following table shows that hecj has a syncretic form for genitive and accusative, unlike theycl. Table 2: Case of heci and theyci (10) Case he they genitive ho jich accusative ho je As predicted, in case one conjunct assigns genitive and the other conjunct assigns accusative, only 'hecj' is acceptable: (11) 1st conjunct genitive, 2nd conjunct accusative a. ?Já se ho bojím a nenávidím. I refl hinigen-acc be-afraid and hate 'I am afraid of him and loathe him.' b. *Já se jich bojím a nenávidím. I refl themgen be-afraid and hate T am afraid of them and loathe them.' c. *Já se je bojím a nenávidím. I refl themacc be-afraid and hate T am afraid of them and loathe them.' 2.2 Coordination with more than two conjuncts As discovered by Ross (1967), conjuncts are islands (i.e., structures which block movement out of them). This descriptive generalization is known as the Coordinate Structure Constraint: (12) Coordinate Structure Constraint (CSC): In a coordinate structure, no element contained in a conjunct may be moved out of that conjunct. 5Ellipses in these two examples are marked by the underline _ coindexed with the ellipsis' antecedent. 69 There is a well-known exception to that generalization, an exception that is a topic of this paper: ATB extraction. This has also been noticed by Ross in his seminal thesis: (13) There is an important class of rules to which (12) does not apply. These are rule schemata which move a constituent out of all conjuncts in a coordinate structure. (Ross, 1967, page 107) Thus, whereas (14-a) is ungrammatical (violation of the CSC), (14-b) is fine (ATB extraction). (14) a. *Co jsi koupil a odešel? what past-aux bought and left '*What did you buy and left?' b. Co jsi koupil a prodal? what past-aux bought and sold 'What did you buy and sell?' So far, we have seen examples of clitic omission in which clitics originated in both conjuncts and surfaced outside of the coordination. This confirms the description as given in (13). If clitics originated only in some of the conjuncts we should expect ungrammaticality because clitic omission is a case of ATB movement which is movement out of all conjuncts. If clitics moved from some, but not all, conjuncts, the Coordinate Structure Constraint should be violated, which should cause ungrammaticality. This prediction is borne out. (15) shows a case of coordination of two conjuncts. Clitics undergo ATB extraction out of both conjuncts: (15) Petr se umyl a oholil. Petr refl washed and shaved 'Petr washed and shaved.' Coordination of more than two conjuncts is also possible if clitics undergo ATB extraction out of all conjuncts: (16) Petr se myl, česal a holil. Petr refl washed, combed and shaved 'Petr washed, combed and shaved.' However, it is crucial that clitics end up outside of the coordination and are omitted in all conjuncts: (17) *Petr se myl, čistil si zuby a holil. Petr refl washed, brushed refl teeth and shaved 'Petr washed, brushed his teeth and shaved.' As said above, the ungrammaticality of (17) follows since the CSC is violated in this case. On the other hand, if clitic omission was a case of ellipsis, the ungrammaticality of (17) would come as a surprise. Simply put, there is no reason why ellipses should be all-or-nothing phenomenon. In other words, there is no reason to expect that ellipses can either appear in all conjuncts (apart from the first one), or in none of them. 70 2.3 When ATB extraction is blocked The final argument that I believe supports my analysis of clitic omission as ATB extraction comes from a surprising piece of data. As pointed out to me by an anonymous reviewer of Formal Approaches to Slavic Linguistics 2007, not all coordinations allow for clitic omission within their conjuncts. According to the reviewer (18), in which the clitic mu 'him^^' is omitted, is ungrammatical (I myself do not find it completely ungrammatical, but agree that the sentence is degraded): (18) *Uz se mu ulevilo a je lip already refl himnjm relieve and is better 'He was relieved of pain and feels better.' The ungrammaticality of (18) surely is surprising. The example represents a coordinated structure and therefore ATB extraction of clitics should be licensed. Fortunately, independent reasons have been found which suggest that (18) might not be a coordinated structure after all. As discussed at length by Postal (1998), coordinations in which the first conjunct serves as a cause of the second conjunct show different behavior than standard coordinations. For example, they allow for violation of the Coordinate Structure Constraint (19) , as noticed by Lakoff (1986): (19) The stuff which} the guys in the Caucasus drink tj and live to be 100. Postal (1998) suggests that these constructions should not be analyzed as coordinate structures but as subordinations. If his analysis is on the right track, we might have an explanation for the ungrammaticality of (18). Notice first that (18) is also a case in which the first conjunct is a cause of the second conjunct. Now, if, following Postal (1998), we do not analyze (18) as a coordinate structure, we actually expect ATB extraction to be ungrammatical. Notice that ATB extraction of a wh-word is also impossible in the same coordination which supports our conclusion that (18) is ungrammatical because ATB extraction is impossible in this case: (20) *Komuj se uz ulevilo a je lip? who already refl relieve and is better 'Who was relieved of pain and feels better?' This concludes arguments for analysing clitic omission as ATB extraction. From now on, I am going to assume that this analysis is correct. In the next section I am going to concentrate on a surprising case of clitic omission: one in which clitics arguably do not move out of the coordination but stay inside the first conjunct. 3 ATB extraction with seemingly no extraction 3.1 Problematic examples Take a look at the following example: (21) Zavolal jsem ho a představil známým, called past-aux^Sg hiniacc and introduced friends 'I called him and introduced him to friends.' 71 Let us go through this example more carefully. First, notice that the clitics jsem and ho are omitted in the second conjunct. Clearly we deal with clitic omission, which, as I argued, should be analyzed as an ATB extraction of clitics. Thus, both clitics jsem and ho originate as an auxiliary and an argument in both conjuncts. Later on, they both move out of the coordination. Notice that both clitics are linearly ordered at the end of the first conjunct. Since clitics undergo ATB extraction out of the coordination the verb preceding them must undergo movement out of the coordination as well. Thus, we end up with the structure (22) for example (21) . (22) [ calledj [ past-auxj him^ [conjp [yp *i *j *k 1 ["VP introduced tj t^ to friends] ] ] ] The problem is that in this structure the verb underwent movement from the first conjunct. But this movement violates Coordinate Structure Constraint and thus should render (21) un-grammatical! Furthermore, notice that after ATB extraction of clitics the first conjunct consists only of 'called'. Its movement out of the coordination violates another well-known constraint, the Conjunct Constraint (Grosu, 1981), which states that no conjunct may be moved. We have already seen (in Section 2.2) that the Coordinate Structure Constraint applies in Czech. (23) shows that the Conjunct Constraint applies in Czech, as well: (23) *Chlapec kterého jsem potkal a Petra. boy which past-aux met and Petr '*The boy which I met and Petr.' In short, we have ended up in a paradox. If we assumed that clitic omission is a case of ellipsis which targets clitic cluster we would have no explanation for the data discussed in the previous section (Section 2). But if we assume that clitic omission is a case of ATB extraction we expect the movement of the verb 'called' in (21) to violate the Coordinate Structure Constraint and the Conjunct Constraint and, therefore, we expect (21) to be ungrammatical, contrary to the facts. There are, as far as I can see, two possible ways out of this paradox. The first one is to assume that the Coordinate Structure Constraint and the Conjunct Constraint are structure-specific. They do apply in cases like (23) but are not applicable to (21). The second way out of the paradox is to assume that there is something special about ATB extraction of clitics. Something that enables them not to be pronounced outside of the coordination in cases like (21). In that case, the verb 'called' do not need to move out of the coordination either, and, therefore, no violation of the Coordinate Structure Constraint and the Conjunct Constraint takes place. I am going to follow the second route. 3.2 Clitic placement at the syntax-phonology interface One of the many issues connected to the study of clitics is a question of how to account for their placement. Is their second position a result of a phonological requirement, syntactic requirement, or both? In his recent work, Željko Boskovic (Boskovic, 2001) follows Franks (2000) and suggests that we should consider clitic placement an interface phenomenon. It is a result of an interplay between syntax and phonology. Following the Minimalist Program (see, especially, Chomsky 1995), Boskovic assumes that when a constituent moves it leaves a copy of itself in the original position. The two copies (one in the base position and the other in the target position of the movement) are indistinguishable 72 from each other. However, they do behave differently with respect to pronunciation. As has been noted many times in literature, it is usually only the highest copy that is pronounced. All the lower copies are deleted (see Nunes 2001 and Nunes 2005 for an extensive discussion and interesting explanation of why this should be so). It is this last point that, according to Boskovic, differentiates clitics from other moved constituents. Clitics enter computation with syntactic requirements. For the sake of argument let us assume that all clitics in Czech need to move via head movement to C.6 However, unlike most other words, clitics also come with phonological requirements that must be satisfied. Following Boskovic I assume that clitics in Czech need to lean on a host to their left.7 Recall that it is normally the highest copy of a moved constituent that is pronounced. Thus, we would expect clitics to be pronounced in their highest position, the head of C. However, this copy might violate clitics' phonological requirements: if there is no material higher in the sentence, clitics cannot lean on any host to their left and the sentence is ungrammatical. Boskovic (2001) suggests that in this scenario, a lower copy of clitics is pronounced: the one that satisfies phonological requirements. Let us go through one example: (24) Zavolal jsem ho. called past-aux^Sg himacc 'I called him.' Let us assume that the clitic 'him' is merged as a direct object of the verb and later on moves as a head to the projection which hosts the auxiliary. Furthermore, let us assume (following Veselovska 2004) that past auxiliaries are located in the head T. Thus, at the level of T, the structure looks followingly: (25) [rp [t past-aux hinij ] [yp [v ] [yp [v called ] tj ] ] ] The T projects further up. Following Migdalski (2006), let us assume that to satisfy EPP requirements of the T, the vP moves into its specifier. After that, C selects for the TP and clitics head move via the T into the head C. This is the final structure: 6 This goes contra arguments that Boskovic explicitly makes for Serbian, namely that clitics do not all move into the same position. However, the arguments on which this conclusion is based do not work in Czech. Thus, I assume that unlike in Serbian, all clitics in Czech do target the same position. If this turns out to be incorrect, the analysis that I am going to propose can still be maintained, with appropriate modifications. 7There are problems with this assumption for Czech because clitics in Czech can be either enclitics or proclitics. I will turn to this issue in Section 4. 73 (26) CP past-aux him However, in this case the highest copy of the clitics cannot be pronounced because it would violate phonological requirements. Therefore, a lower copy is selected: the one in which clitics sit in the T. The pronunciation of the lower copy gives us the correct word order for (24). 3.3 ATB extraction with seemingly no extraction explained Armed with an analysis of how clitics' syntactic and phonological requirements together derive their position let us move to the example discussed in Section 3.1. (27) Zavolal jsem ho a představil známým, called past-aux^Sg hiniacc and introduced friends 'I called him and introduced him to friends.' Recall what the problem was. Example (27) is a case of ATB extraction of clitics. Therefore, clitics must have moved out of the coordination. But if they did (so the reasoning went) then the verb 'called' must have moved out of the coordination, too, in violation of the Coordinate Structure Constraint and the Conjunct Constraint. There is a flaw in this reasoning, and, I believe, it should be obvious by now what the flaw is. In Section 3.1, I have assumed without discussion that what is pronounced is the highest copy of a moved constituent. In fact, this assumption is natural if one follows syntactic accounts to the second position of clitics and assumes that the second position is derived solely in syntax. However, we have seen in the previous section that this does not need to be so. More concretely, I presented an approach to clitic placement which argues that the second position is a result of a phonological requirement which leads to the pronunciation of a lower copy. This is one possible analysis of (27) which avoids violation of the Coordinate Structure Constraint and the Conjunct Constraint: first, the object is merged as the object of the verbs in both conjuncts and it adjoins to the auxiliary in the T, still separately in the two conjuncts. The T in both conjuncts project. The T in the first conjunct attracts the verb phrase consisting of 'called' into its specifier. The T in the second conjunct attracts the verb phrase consisting of 'introduced' into its specifier. Now, the two TPs are conjoined and C takes this coordination as its complement. The clitics jsem and ho undergo ATB extraction into the C. This is the final structure: 74 However, if the highest copy of the clitics was pronounced the clitics' phonological requirement would not be satisfied. Therefore, a lower copy must be pronounced. Counting the number of nodes, we arrive at the second highest copy: the one in which the clitics sit in the T in the first conjunct. Thus, this copy of clitics is pronounced and we end up with the correct word order without violating the Coordinate Structure Constraint or the Conjunct Constraint. There are at least two predictions that this approach makes. First, notice that full DPs should behave differently than clitics do with respect to examples like (27). This is so because full DPs do not have a phonological requirement that would force pronunciation of a lower copy. Since it is always the case that the highest copy of these DPs is pronounced examples like (27) with DPs in the place of the clitics should be ungrammatical. This is correct. Notice first that full DPs can license what looks like an object drop: (29) Petra jsem zavolal a představil známým. Petracc past-aux called and introduced friends 'I called Petr and introduced him to friends.' Since there is no object drop in Czech, I conclude that example (29) is a case of ATB extraction: 'Petracc' starts as an argument of the verbs in both conjuncts and undergoes ATB extraction out of the coordination. Thus, example (29) shows us that ATB extraction of full DPs is possible. Interestingly, an example parallel to (27), in which the clitic ho 'him' is substituted by the full DP 'Petracc' is ungrammatical: (30) *Zavolal jsem Petra a představil známým called past-aux^Sg Petracc and introduced friends 'I called Honza and introduced him to friends.' 75 As discussed above, this follows from the presented analysis. 'Petracc' comes with no phonological requirements that would force pronunciation of a lower copy in this example. Therefore, the highest copy of 'Petracc' has been pronounced in (30). Since the highest copy must be outside of the coordination (otherwise, there is no way to license object drop in this coordination), the verb 'called' must have moved out of the coordination, as well. But this movement of the verb violates the Coordinate Structure Constraint and the Conjunct Constraint! The second prediction that the presented analysis makes concerns interaction of clitic omission with the size of conjuncts. To license clitic omission in (27), repeated here as (31), clitics have to move out of coordination even though this movement is masked by the fact that the copy outside of the coordination is not pronounced. (31) Zavolal jsem ho a představil známým, called past-aux^Sg hiniacc and introduced friends 'I called him and introduced him to friends.' Crucially, clitics must be able to move out of the coordination, otherwise the structure is illicit. If we dealt with a coordination of conjuncts which are islands for clitic movement (i.e., if we dealt with a structure bigger than the TP) clitics should not be able to undergo ATB extraction out of the coordination and clitic omission should consequently be ungrammatical. This should be so even though clitics are not pronounced outside of the coordination. This prediction is also borne out. Notice first that clitic omission is possible in conjoined TPs in embedded contexts: (32) Petr řekl že jsem ho zavolal a představil známým. Petr said that past-aux^gg himacc called and introduced friends 'Petr said that I had called him and had introduced him to friends.' However, once we conjoin CPs clitic omission is impossible: (33) *Honza řekl že jsem ho zavolal a že představil známým. Honza said that past-aux^gg himacc called and that introduced friends 'Honza said that I had called him and that I had introduced him to friends.' As is well-known, clitics cannot move out of CP (see Progovac 1993, Veselovská 1995, Rezac 2005, or Dotlačil 2007 for different explanations of this fact). In (33), clitics have to undergo ATB extraction out of the coordination. However, since clitics cannot move out of CP this movement is illicit, and since ATB extraction is impossible, clitic omission cannot take place in (33). In conclusion, the grammaticality of clitic omission in (31) supports the analysis of clitic placement in Czech along Franks (2000) / Boskovic (2001) line. We have seen an argument for the language architecture in which phonology overrides syntax in deciding which copy should be pronounced (see Nunes 2005 for an approach to the pronunciation of copies which is compatible with this view). The next section is going to discuss one problem ignored so far: do we really have any reasons to assume that clitics come with phonological requirements? 4 Czech clitics and their phonological requirements As discussed by Fried (1994), Toman (1996), and many others, clitics in Czech do not need to lean on a host to their left. Clitics in Czech can be at the beginning of an intonational phrase 76 and lean on a host to the right (i.e., behave as proclitics) (example (34-a)). They can, in fact, even be at the beginning of a clause in colloquial Czech (example (34-b)) (see Lenertová 2004 for more examples and discussion). (34) a. Já # tvoje máma # jsem ti slíbila hračku. I # your mother # aux you promised toy T, your mother, promised you a toy.' b. Se mi včera narodil syn. (ok only in colloquial Czech) refl me yesterday born son 'My son was born yesterday.' But is the fact that Czech clitics can be either proclitics or enclitics important for my analysis of examples like (35)? (35) Zavolal jsem ho. called past-aux^Sg himacc 'I called him.' In fact, it is. Following Boskovic (2001), I analyzed (35) as a case in which phonology forces pronunciation of a lower copy of clitics. But if clitics can also be proclitics there is no reason to expect that phonology should force the pronunciation of a lower copy in this case. The highest copy would violate no phonological requirements after all. Since phonologically, clitics can be both proclitics and enclitics, shouldn't it be only syntax that derives clitics' second position (in standard Czech), with no help from phonology? One way out from this problem that I would like to suggest is that clitics are not underspec-ified for proclitization and encliticization. Instead, they are ambiguous. In (35) enclitics were chosen in the lexicon: since these need to lean on their hosts to the left, the pronunciation of a lower copy is forced. If proclitics were used instead the highest copy could be pronounced (for reasons unclear to me, this latter option is possible only in colloquial Czech). Surprisingly, (34-a) does not allow for the possibility in which clitics become enclitics (and end up not in the second, but third position in the clause): (36) ??Já # tvoje máma # slíbila jsem ti hračku. I # your mother # promised aux you toy 'I, your mother, promised you a toy.' This is surprising if clitics were ambiguous: why should we not be able to retrieve to enclitics in (36) which would force the pronunciation of a copy below the predicate? I tentatively assume that there is nothing wrong with choosing enclitics in (36). What goes wrong with this example is the predicate movement. The predicate 'promised' cannot move above clitics to support them. Notice that if the predicate cannot move from its base position both clitics jsem and ho must end up at the left edge of the intonational phrase.8 To be sure there are much more cases that one should analyze before jumping at the conclusion that the hypothesis of clitics being ambiguous between enclitics and proclitics really is viable. Unfortunately, since this would lead me too far afield from the topic of this paper, I have to leave them aside. I assume that the base position of ho cannot be pronounced for independent reasons - see Moro (2000) 77 5 Conclusion This paper has discussed cases of clitic omission as in (37): (37) Já jsem ho zavolal a představil známým. I past-aux^Sg hiniacc called and introduced friends 'I called him and introduced him to friends.' I argued that these cases should be analyzed as ATB extraction of clitics. However, this conclusion turned out to be problematic in cases of examples in which clitics seemed to stay inside the first conjunct: (38) Zavolal jsem ho a představil známým, called past-aux^Sg himacc and introduced friends 'I called him and introduced him to friends.' This, I argued, should be seen as evidence that clitics might be pronounced in their non-highest position, which is in line with Boskovic's analysis of clitic placement in Serbian (Boskovic, 2001). Based on the examples like (38), I argued that clitics' position in Czech is not solely derived in the syntactic part of the computation but is a result of an interplay between syntax and phonology. If this analysis is on the right track cases like (38) supports a particular view on the syntax-phonology interface. Under this view, phonology can alter word order that is derived in syntax by deciding which copy of movement done in syntax should be chosen for pronunciation. References Boskovic, Zeljko. 2001. On the nature of the syntax-phonology interface: cliticization and related phenomena. Oxford: Elsevier Science. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Massachusetts: MIT Press. Dotlačil, Jakub. 2007. Why clitics cannot climb out of CP: a discourse approach. Compton, R. et al. (eds.). Formal Approaches to Slavic Linguistics: The Toronto Meeting 2006. Ann Arbor: Michigan Slavic Publications 15:76-93. Dyla, Stefan. 1984. Across-the-board dependencies and Case in Polish. Linguistic Inquiry 15:701-705. Franks, Steven. 2000. Clitics at the interface. Beukema, F. & M. den Dikken (eds.). Clitic phenomena in European languages. 1-46. Amsterdam, Philadelphia: John Benjamins. —. 1993. On parallellism in across-the-board dependencies. Linguistics Inquiry 24:509-529. Fried, Mirjam. 1994. Seccond-position clitics in Czech: Syntactic or phonological? Lingua 94:155-175. Grosu, Alexander. 1981. Approaches to island phenomena. Amsterdam, New York: North-Holland. Johnson, K. 2006. Gapping is not (VP) ellipsis. Unpublished, ms. 78 Lakoff, George. 1986. Frame semantic control of the Coordinate Structure Constraint. Farley, A. M. (ed.). Papers from the Parasession on Pragmatics and Grammatical Theory. Chicago: University of Chicago. Lenertová, Denisa. 2004. Czech pronominal clitics. Journal of Slavic linguistics 12:135-171. Migdalski, Krzysztof. 2006. The Syntax of Compound Tenses in Slavic. PhD dissertation, Tilburg University, Tilburg. Moro, Andrea. 2000. Dynamic antisymmetry. Cambridge, Massachusetts: MIT Press. Nunes, Jairo. 2005. Linearization of Chains and Sideward Movement. Cambridge, Massachusetts, London, England: MIT Press. —. 2001. Sideward movement. Linguistic Inquiry 31:303-344. Postal, Paul M. 1998. Three investigations on extraction. Cambridge, Massachusetts: MIT Press. Progovac, Ljiljana. 1996. 1993. Locality and Subjunctive-like Complements in Serbo-Croatian. Journal of Slavic Linguistics 1:116-144. —. Clitics in Serbian/Croatian: Comp as the second position. Halpren, A. &; A. M. Zwicky (eds.). Approaching second: Second position clitics and related phenomena: 411-428. Calif: CSLI. Rezac, Milan. 2005. The syntax of clitic climbing in Czech. Heggie, L. &; F. Ord0fiez (eds.). Clitic and Affix Combinations. Amsterdam: Philadelpia: John Benjamins. Rivero, Mária Luisa. 1991. Long head movement and negation: Serbo-Croat vs. Slovak and Czech. The Linguistics Review 8:319-351. Ross, John Robert. 1967. Constraints on Variables in Syntax. PhD dissertation, MIT, Cambridge, Massachusetts. Schutze, Carson. 1994. Serbo-Croatian second position clitic placement and the phonology-syntax interface. MIT Working Papers in Linguistics 21:373-473. Toman, Jindřich. 1996. A note on clitics and prosody. Halpren, A. &; A. M. Zwicky (eds.). Ap-proaching second: Second position clitics and related phenomena: 505-510. Stanford: CSLI. Veselovská, Ludmila. 1995. Phrasal Movement and X-Morphology: Word Order Parallels in Czech and English Nominal and Verbal Projections. PhD dissertation, Palacký University, Olomouc. —. 2004. The Extended verbal projection in Czech: Three Variants of the verb be. FDSL 5. Leipzig. 79 Czech Cases and the Syntacticon: Poznámky k, o, okolo, nad něčím a pro někoho Joseph Emonds jeemonds@hotmail.com Czechs and non-Czechs alike generally feel that the Czech morphological case system is quite complex, especially as it manifests itself with prepositions or without them, but in phrases which correspond to PPs in say English.1 It is the purpose of this study to show that Czech cases in PP structures are more than "orderly"; they are in fact, with literally one clear exception, entirely regular and predictable.2 Of course, in order to do this, one cannot just accept every truism that has been handed down through traditional grammar. Moreover, it is necessary to crucially use a few highly motivated but innovative constructs formulated within generative grammar. If such moves were not necessary, the regularity of the Czech Case system would have become apparent before now. Generally, structuralist and generativist authors alike are overly reliant on traditional grammar's truisms. Even generativists seem to feel that these truisms are somehow "inherent in the data" and can be used at no cost. So in practice they prefer to perpetuate them rather than utilize theoretical constructs not yet approved in their perception by "current theory." Non-Slavic speaking linguists acquainted with Indo-European (I-E) morphological case soon sense that many uses of Czech's six (non-vocative) cases are far from unfamiliar.3 Nominative and accusative have the same form for inanimates, those termed both "neuter" and "masculine." Most predicate nominals and adjectives agree in case with the noun phrase they modify, whether in primary or secondary predication. Partitives are expressed in the genitive. Noun phrase objects of prepositions expressing motion toward a Goal are in the accusative. What seems bewildering though are the Czech cases used with Ps (or to express some English PPs without a P): accusatives, genitives, datives, locatives and instrumentals all find their places in its system. Compared with Classical Greek, Standard German, Icelandic, Latin and 1 I have had fruitful and enlightening discussions on this topic with both L. Veselovská and M. Martinková. I thank them both, and am very grateful to M. Martinková for carefully reading and commenting on various drafts of this paper and for supplying numerous useful references. Of course, neither is responsible for my errors of fact and interpretation. 2 At least among prepositions that are not highly lexicalized and idiomatic. An entertaining exercise for Czech readers may be to guess which grammatical P is irrevocably exceptional in my system. 3 Czech vocatives also have a familiar Indo-European feel, i.e., akin to those in Latin. In almost all paradigms, a vowel is added to nominatives of proper nouns: a high vowel with masculine stems and a mid-vowel with feminine stems. As vocatives are not related to sentence structures involving other categories, they don't properly speaking constitute a separate case. Compare them to the case structures discussed below in section 1. 81 Old English, in which 3 cases can appear after Ps, Czech cases in PP structures seem more arbitrary and less predictable. 1 Against the traditional definition of "dative" for the third case To my mind, the earliest generative attempts around 1980 to make sense of morphological case incorporated two insightful idealizations, but at the same time formulated them in terms of an unperceptive and unreflective lapse. The good news is (i) that cross-linguistically each basic case results from a characteristic case-assigning category, and (ii) these basic case-assigning categories are few in number, basically four: (1) a. "Nominative" is assigned by I to the DP closest to it. b. "Accusative" is assigned by V to the DP closest to it. c. "Dative" or "oblique" is assigned by P to the DP closest to it. d. "Genitive" is assigned by some category in N-projections to the closest DP. The bad news is that the field retains to this day these "names" of these cases as if they are somehow categories in their own right. Perhaps this gives the impression that generative grammar explains these traditional concepts, but in most cases it simply uncritically accepts them, e.g. even elaborating on the timeworn distinction between "structural" and "semantic" case.5 Emonds (1985: Ch. 1 and 5), argues in detail that no special case categories should appear in grammatical representations; "nominative" is nothing other than the appearance of I as a feature or index on nominal projections and likewise for the other cases. Consequently, the case features should be represented essentially as follows, where the traditional names for them have absolutely no theoretical status. (2) a. Nominal projections Dk in the nominative are notated Di. b. Nominal projections Dk in the accusative are notated Dv. c. Nominal projections Dk in the oblique case are notated DP. d. Nominal projections Dk in the genitive are notated Dq. 6 These notations involve more than terminological parsimony, even though that alone would justify them. They also remove prejudices that impede good analyses, e.g. the instinctive feel of many linguists (including generativists) that traditional terms like e.g. "datives" or "genitives" capture similarities across languages independent of actual empirical cross-linguistic 4 Latin grammars always remark that no P can ever be followed by its dative case, leaving only genitives, accusatives, and ablatives that Ps can introduce. 5 "Semantic" cases are those assigned before movement of object nominals in passives, while "structural" cases are those assigned after it. This leads to a problem with genitives, which government and binding treatments turned a blind eye to. As observed in Veselovska (1997), genitive phrases act "structurally" inside noun phrases, but "semantically" when they are complements to verbs. The minimalist recognition (Chomsky, 1995) that trees are constructed from the bottom up removes any problem, as well as the need for the semantic/structural distinction. Once assigned, case cannot change. DPs inside PP domains receive case before the PPs merge into larger structures ("semantic case"), while DPs not inside PPs merge in IP or larger DP domains without case ("structural case"). If such IPs or DPs then merge further, the cases on the DPs they properly contain cannot change, exactly as with "semantic case." 6 Veselovska (2001) argues, to my mind convincingly, that the case assigning category inside nominal projections is Q, for (existential) "quantifiers" including, in Czech, high numerals. I adopt this here throughout. 82 justifications. This harmful hidden assumption puts meaningless burdens of proof on any analyses in which Case X in is the basic equivalent of some otherwise named Case Y in L2, rather than what tradition calls Case X in L2. In addition, reducing "cases" to features or indices as in (2) (i.e. an accusative is a D or N projection with a V feature or index) makes possible some revealing structural analyses of how predicate nominals and adjectives receive morphological cases, including those in Czech (Emonds 2000: Ch. 8). For example, the Case feature Dv on a direct object assigns case to a following secondary predicate nominal. Such analyses cannot even be envisioned if one insists on the traditional case names formally unrelated to the case assigners. Although most work on case is unaware of such accounts, they nonetheless constitute unanswered and empirically based arguments in favour of eliminating, as in (2), traditional or any other ad hoc names for cases. Perhaps the most striking innovation in the generative approach to case is embodied in (lc) and (2c). Traditional grammar always uses the terms nominative, accusative and dative for the morphological cases of respectively the subject, direct object and indirect object of typical simple three place predicates. It furthermore treats the three cases as entirely unrelated. However, massive cross-linguistic syncretism between nominatives and accusatives (throughout I-E and not least in Czech) shows that the first two cases are indeed subcases of some more general verb-related "archi-case."7 But related to each other or not, both "nominative" and "accusative" noun phrases are associated in (la-b) with structural configurations closely connected with I and with V. Hence the generative treatment of these cases is not so far from their traditional associations with subjects and direct objects. For "datives," the situation is different. And indeed, the early generative proposal for case in e.g., Chomsky (1981) does not define datives as the "case of the indirect object." Rather, the generative "dative" was (and is) the case structurally associated with the structure of PP. Indeed, the surface form of indirect objects in many systems is not associated with PP at all. For example, in several languages lacking morphological case, benefactive applicative phrases, as studied in e.g. Baker (1988), typically surface in direct object position and are arguably accusative, sometimes by morphological as well as syntactic criteria. In Modern Standard Arabic, indirect objects also appear as accusatives, i.e. with the same case as direct objects.8 Pursuing this line, it seems like the "third case" or the "P-assigned case" in (lc) should be identified with the case used for expressing locations of a verb's action or state. In Classical Greek, Standard German, Icelandic and Old English, this case happens to also be that of the indirect object (the so-called "dative"). However, Latin's locative case is its ablative, not its dative.9 The third Czech case, judging from how many locative Ps govern it, is neither what is called the dative or the locative, but is rather the "instrumental." Such are the cases that express, in each of these languages, the Ps or PP structures translating under, over, between, in front of, behind, through and with, when these are conceived of as pure locations unassociated 7 Generative treatments have not recognized this. I pursue this topic elsewhere; it does not interact with this essay's subject matter except in one particular. The fact that nominatives and accusatives are partly "the same" can explain why Ps in I-E systems never "take a nominative." Traditional grammars, if they even note this fact, provide no explanation for it. 8 There is a underlying universal relation between indirect objects and PP structure, but it is not a surface relation. This position is cross-linguistically supported in Emonds (1993). 9 The Latin "dative" is a variant on the ablative and has different forms only in (most but not all) singulars; Latin plural datives and ablatives always have the same form. Moreover, in all paradigms where it differs from the ablative, the dative singular has a single form -i. Emonds (1985: Ch. 5) argues that this special allomorph appears in Latin PPs if and only if an introductory [P, GOAL] is 0. As discussed in section 7, the status of the Czech dative resembles to some extent that of the Latin dative. 83 with Goals or Sources of movements or transitions. Reinforcing this first conclusion for Czech, we find this language has a sometimes specially remarked use of the instrumental, the bare (preposition-less) "instrumental expressing place," exemplified in bold here with some examples from Holá (2000: 235): (3) Musíte jít podchodem a pak doleva. 'You must go through the underpass and then left.' Šli jsme jinou ulicí. 'We went by the other street.' Vlak projel malým nádražím. 'The train went past a small station.' Reka Vltava teče Prahou. 'The Vltava River flows through Prague.' I will further be developing the idea that Czech system is most closely related, in terms of other I-E systems, to that of Classical Greek (the latter's nominative, genitive, dative and accusative). In Czech, the closest structural relative to the Classical Greek dative is the instrumental. Of course, the Czech dative and locative cases will also be analyzed; in fact, the patterns surrounding these "additional" cases will be a crucial testing ground for refining and developing a theory of the lexicon that I have proposed in earlier work, so these recalcitrant cases are in fact the main motivation for pursuing this topic.10 This study will incorporate the "good news" on morphological case from Government and Binding (four basic cases each associated with a universal structure) and discard the "bad news." The names of the cases are just mnemonics for DPs structurally marked as occurring in certain constructions. I use the neutral descriptive term "oblique" for the case generally associated with a P of location, i.e. the Classical Greek dative, the Czech instrumental and the Latin ablative. 2 Probable universal syntactic features of Czech Ps Several works by Jackendoff (1977, 1983, 1990) and others have established the need to clearly distinguish noun phrases denoting a LOCATION of a verb's action or state from those denoting the GOAL of a verb's action. Only the latter are typically introduced with a counterpart to English to. This work has also shown that noun phrase SOURCES of a verb's action, indicated in English with from, share some similarities with GOALs, which he proposes to capture by treating Ps that introduce both types as "Prepositions of PATH". The Path Ps that introduce GOALs and SOURCE DPs differ in their grammar from the Place Ps that introduce LOCATIONS. Many relevant syntactic analyses by van Riemsdijk (1978, 1988, 2002) have brought out further differences between these two classes of Ps, which he differentiates by means of a feature that we can identify with ±PATH.n Van Riemsdijk's work on complex case systems of e.g, some Caucasian languages, as well as 10 Case systems outside Indo-European exhibit little if any relation between expressing simple locations with Ps and the case of indirect objects. Locations with P use the nominative in Turkish and the genitive in Modern Standard Arabic. Descriptions of Turkish follow common practice and call the case of its indirect objects "dative". 11 Van Riemsdijk and Huijbregts (1998) show that Ps of PATH often introduce PPs of PLACE as their complements, i.e., two Ps can succeed each other: from behind the barn, onto (<= to + on) the table, French [path en ] [place dessous ] de la table 'to underneath (of) the table'. We are not concerned here with sequence of Ps, but only with those Ps closest to DPs that assign them case. 84 traditional analyses of somewhat more familiar grammars of e.g. Sanskrit and Finnish, suggest that the basic features on P further divide [P, PATH] into those expressing "motion toward", "motion up to", and "motion into", in both directions (to and from). It appears that Czech cases do not grammatically distinguish these three subcategories of path, but only two in each direction, as in (4). We can specify those of the left column as +CONTACT and those of the right column as -CONTACT. (4) Czech grammatical P of +CONTACT | Czech grammatical P of -CONTACT motion up to or into: P = do+ genitive motion toward: P = k(e)+ dative motion off or out of: P = z(e) + genitive motion away from: P = od + genitive The grammatical features that seem to cross-classify Czech prepositions thus include at least LOCATION, GOAL, SOURCE and CONTACT. The feature -PATH stands for the conjunction [-GOAL, -SOURCE], but I am not sure if as a feature on D it has a theoretical status or is just shorthand. In terms of these features, we can cross-classify Czech Ps in a table, indicating at the same time the cases typically exhibited by their object nominals. 85 (5) Table of Czech cases on DPs objects of Ps Czech features on Ps that have DP objects Grammatical entries for P = 0 or P * 0 P, ±F, -f, where F: =±GOAL, iLOCATION, iCONTACT Open Class Dictionary (P t 0): P, +LOCATION, ±GOAL, +fj Positions of: x = modified XP, s = speaker, o = object of P F; = -LOCATION -GOAL, -SOURCE P,-F= 0 'with' (instrument) = s(e) 'with' Not defined, since Dictionary Ps must be +LOCATION. F; = +LOCATION -GOAL, -SOURCE In G-B terms, this case is "oblique." The Czech term is instrumental, and is in red font here. In Latin, this is "ablative." P,-F = 0 'through, by' (bare locational DP) s x o před 'in front of s o x za 'beyond' x o x mezi 'between' X o nad 'above' _s_(or other reference point) o x pod *below' s (or other reference point) P,+F; = o 'about' = v(e) 'at, in' = pfi 'near, during' = po 'over, along' = na 'on' This square to be analysed in section 5. (This will be the Czech "locative square.") F; = +LOCATION +GOAL -SOURCE The default is the super-case assigned by V, namely the Czech accusative, in blue font here. P = 0 (indirect objects) P = (az) k(e) 'toward' This square to be analysed in section 6. (This will be the Czech "dative square.") Exception: do + genitive: '(up) to' The same case as assigned by V, namely accusative. o '(to) against, than' přes '(to) across' před '(to) in front of, mezi '(to) between', nad '(to) above', etc. na 'onto', purpose, duration mimo 'outside, besides' pro 'for' po 'up to, during' v(e) 'on' (time), some idioms za 'for' (in exchanges), 'as' F; = +LOCATION +SOURCE -GOAL The Czech genitive case is in green font here. The genitive here is similar to that in Classical Greek. It is unclear whether +SOURCE should be +GOAL or -GOAL. P = 0 (partitives) P = od(e) '(away) from' z(e) 'out of, off bez 'without' blízko 'near from' nedaleko 'not far from' okolo 'around' vedle 'next from' kromě 'except' vyjma 'except' (?)Lexical items with [p,source0]: během 'during' koncem 'at the end of podél 'along' u 'at the place of, near' včetně 'inclusive of Some one way semantic implications related to case can be appreciated immediately. For example, the genitive expresses motion away from while the accusative expresses motion into. Another tendency evident in Table (5) is that the majority of Ps take objects in the accusative, genitive or instrumental cases, while not many Ps take datives or locatives. That is, in the presence of P, not all cases are equal, an asymmetry which requires an account. 11 Emonds (2000, Ch. 2) argues that no actual categorial difference separates P and the feature LOCATION. The interpretation "spatial or temporal location" is nothing more than the presence of P at Logical Form ("LF"). For a mechanism by which P can be marked as not interpreted as LOCATION in LF, see section 8. 86 3 Correspondences between Czech and Classical Greek Prepositional Cases This section will not yet approach the difference between what in Table (5) are termed "grammatical" and "open class" prepositions, in part because neither traditional grammar, nor structuralism, nor generative grammar provides any clear criteria to work with. This section focuses only on the rows of Table (5), making no distinction between its columns. This step leads us to investigate three basic case types among Czech objects of P: those traditionally called the instrumental/ locative (rows 2 and 3), the dative/ accusative (row 4), and the genitive (row 5). Prior to specifying a case-assignment theory, these labels have no content other than descriptive mnemonics for phonological spellings. Table (6) expresses two pivotal correspondences between Czech and its ancient Indo-European cousin Classical Greek. Rows 1-3 in Table (6) correspond to rows 3-5 in Table (5). We for the moment ignore case usages lacking overt Ps, such as indirect objects. (6) Cases on objects of overt Ps expressing location (not extended to temporal P): +LOCATION Czech Classical Greek -GOAL (pure location) instrumental/ locatives datives +GOAL, -SOURCE datives/ accusatives accusatives +SOURCE genitives genitives Two traditionally named cases in Table (6) have corresponding uses: accusatives and genitives. I turn first to an important characteristic common to the two systems (in row 3), the fact that genitive case is used for "motion away from" and also for "privative" prepositions expressing concepts such as instead of, except, without and not far from. 3.1 An extended l-E use of genitives common to Classical Greek and Czech The use of the genitive to express separation from a SOURCE is not "natural", that is, a widespread pattern imposed by Universal Grammar. A number of non-I-E languages with a clear genitive case mark, perhaps the best studied being Japanese, entirely reserve the genitive for use within noun phrase projections. That is, whatever the complexities of Japanese structures, one thing is for sure, if one sees the genitive no 'of, then one is inside a nominal structure. A Japanese genitive phrase with or without an immediately dominating PP can never be a constituent of a verbal projection. In contrast, both the Romance and Germanic I-E families have an impressive range of genitive constituents of verb phrases, such as introduced by e.g. English of: His talk was of no interest; we think of her as intelligent; they heard/ spoke/ knew of that fact; John is of a different opinion. The French genitive constructions with de 'of in verbal projections differ from the English ones, but are equally if not more varied. Moreover, I take it the English genitive constructions are typical within Germanic and those of French within Romance. Languages in these same families with morphological genitives (Latin, German, Icelandic) also exhibit such genitives in various verbal constructions (unlike Japanese). However, one pattern conspicuously absent in both Romance and Germanic is any systematic reflex of genitive case for objects of Ps expressing motion away from. This particular property of early I-E (characteristic 87 of Sanskrit) is preserved in only some descendent families, but nonetheless is clear in both Czech and Classical Greek. Thus, the treatment of Greek Ps of locations and the required cases of their objects in Niederle, Niederle and Varel (1991: 214-223) is organized around a Table on their page 213 with the following columns:12 Col. 1 Ps that take accusatives, express goals, and answer "where to?" Col. 2 Ps that take genitives, express "point of departure," and answer "where from?" Col. 3 Ps that take datives, express "location" and "(at rest) contact", and answer "where?" They also refer to such genitives as "genitives of separation." Clearly, their Table's columns correspond respectively to rows 4, 5 and 3 of the Table (5) above of Czech Ps. Thus, Greek translations of Czech Ps of motion away from and some other Greek Ps indicating motion away from take object nominals marked with genitive case. The Roman spellings here don't indicate the accents on the final vowels of Greek Ps. apo '(away) from', Czech od ek/ ex 'out of, Czech z(e) (7) hyper 'from above downwards' hypo 'from below upwards'13 kata 'from above' para 'from close to'14 These are only some of the most studied Greek Ps with genitives, but they suffice to establish that using this case to express "motion away from" is systematic in Greek, parallel to the Czech Ps in the last row of Table (5). And also as in Czech, several Greek Ps take genitives simply by lexical stipulation, independent of their meanings: anti 'against', dia 'through' and pro 'in front of'.15 We can thus propose that Czech and Classical Greek share the essential aspect of how genitive case is spelled out in PPs. The symbol 7rgenitive represents the phonological forms of the genitive inflection, which are too varied to concern us here. (8) Lexical entry for Genitive Case Marking (Czech and Classical Greek): <[ +N ]___>, { DQ / [ Dp, SOURCE ] }, genitive I here explain the form of lexical entries that appear in this study, based on the lexical theory developed in Emonds (2000, 2003, 2005a). The material inside <... > is a word-internal sub-categorization frame, i.e. the D; are suffixes on a stem of category +N, where +N encompasses the nominal categories N, A and D, precisely those that exhibit morphological case. The idea that case suffixes are of category D is justified in Veselovska (2001); cf. note 5. Since a noun-internal location is not an interpretable position in universal grammar for any of D;, PLURAL, or SOURCE, these latter categories can be spelled out as suffixes only by virtue of some LF-independent licensing device that constrains the mappings from syntax to phonological form (PF). In this framework, the general device employed for this is "Alternative Realization" (18), formulated and exemplified in section 5.3 below. 12 German Ps that take genitives are a group of lexical items without this semantic basis. 13These two Ps correspond to Czech pod and nad. Their objects receive different cases according to the scheme in Table (6). The glosses in (7) are relevant for genitive objects. 14Greek para is like Czech pfi. The gloss in the text of 'from' is again only relevant for genitive objects. The grammatical item is -CONTACT; the possibility of three different cases after it suggests that this feature is grammatical in Greek. 15Niederle, Niederle and Varel (1991) discuss some further Greek Ps whose uses with genitives remain unclear to me: amphi, epi, meta, peri and pros. 88 Consequently, a "mini-tree" associated e.g. with a genitive expressing motion away from is (9), where the higher P, SOURCE, DP, and PLURAL are all in their canonical (= LF-interpretable) positions. (9) PP [P, SOURCE] Dp, aPLURAL [np N] N [Dp, SOURCE, aPLURAL] lexical noun stem 7t genitive In lexical entries such as (8), the syntactic specifications appear to the left and the phonological ones to the right. This reflects a central claim that I argue for in Emonds (2003), to the effect that grammatical items are lexically identified (i.e. accessible in processing) by their unique grammatical context and category feature combinations.16 Finally, we can now also state how to specify Ps that take genitives lexically. For example, the fact that Czech do 'to' is [+GOAL, -SOURCE] and yet takes a genitive object is clearly exceptional in Table (5). Consequently do is treated as a lexical exception throughout this study. Using the widespread generative term for item-specific lexical case marking, the object of do is thus a "quirky genitive."17 (10) Quirky Case. P, +GOAL, +CONTACT, , do When no blank occurs inside the context specification , an item is a free morpheme and occurs with a full phrase complement, whose lexical head is required to have the feature F;. In contrast, I claim that open class items are uniquely identified by their phonology, and moreover subgrouped into searchable "cohorts" by their initial consonant clusters (Marslen-Wilson 1987, 1990). Thus, a single open class Dictionary item such as open can be an A, V, or even N (a golf tournament), but a suffix such as the English causative -en is unique by virtue of its syntax; -en can be grouped as a single lexical item with phonologically distinct causative suffixes -ify, ize, and 0, but not with its homophone, the passive suffix -en. It is then correctly predicted that free morphemes with different initial consonants can constitute a single lexical entry only if they are grammatical {bad/worse; go/went; she/her). M. Martinková (pers. comm.) informs me that the Russian translation v(o) of Czech do is not exceptional or "quirky" with respect to Table (6); its object DP is accusative. Not unexpectedly, some Czech verbs also take "quirky case" complements, e.g. ptát se 'ask (of someone)' and bát se 'fear' take quirky genitives; věřit 'believe' and rozumět 'understand' take quirky datives; plýtvat 'waste' and opovrhovat 'despise' take quirky instrumentals. 89 3.2 Further properties of the system of overt Ps that express LOCATION Generally, one may ask whether and how the rows of Table (6) should be grouped. There are three plausible possibilities: one is that GOAL and SOURCE Ps have a common PATH feature (+SOURCE would also be +GOAL); a second is that Universal Grammar does not group them (+SOURCE would then be -GOAL); and a third is that pure LOCATION should be grouped with SOURCE but not GOAL. The least option might seem unmotivated semantically, but in fact the Latin system, which conflates both in a single "ablative case," seems to suggest this. For exposition, I treat Ps of GOAL and SOURCE as not sharing a feature, but do not investigate this question further here. The previous subsection has tried to capture the common expression of SOURCE Ps in Czech and Classical Greek, schematized in Table (6), row 3. The next section will turn to row 2, which proposes a general equivalence of Czech instrumentals and locatives with Classical Greek datives for expressing LOCATIONS. Here the traditional case names obscure the parallelisms, which is why I employ a neutral yet traditional term "oblique case." Finally, let us turn to the use of the accusative for "prepositions of motion toward," as in row 1 of Table (6). Both Czech and Classical Greek robustly exhibit a widespread property of Indo-European morphological case systems, namely the use of the accusative for their objects. (The Czech variant called the dative is treated in section 7.) Focusing here on the accusative, the issue arises, is this use of it stipulative? Does it arise from lexical marking on various and sundry Ps, or is there something more general behind this uniform use of accusatives for Goal DPs? 4 The l-E accusative as a default morphological realization of Case The fact is, the I-E morphological accusative has an essentially default function in several case-marking contexts (i.e. embedding in syntactic structures, as opposed to dislocated topics, vocatives, citation forms, etc.). The same I-E languages in which (i) "accusatives are used for motion to" also use them in nominative positions for all nouns of "neuter gender". For example, the nominative of Latin neuters like helium 'war' has a superficial accusative suffix -to. The following traditional truism expresses this, but provides no explanation: (ii) "For neuter nouns, nominative and accusative are always the same". In fact, it seems that the Indo-European accusative is simply the general mark of constructional case, i.e. the case assigned when case must be assigned, but when no other case is available. As a third example, when the finiteness constituent I is unspecified for features and hence cannot assign nominative to its subject, what emerges in Latin and other languages is a proposition whose form is, using another unexplained shibboleth of traditional grammar, (iii) "accusative with infinitive." In light of the accusative usages (i)-(iii), we need not say that the GOAL prepositions of Czech, Classical Greek or any other I-E language are lexically stipulated to take accusatives. Rather they are unspecified for requiring any case. This perspective has been suggested in Veselovska (2001), who proposes that accusative is an unmarked structural case on DPs. Then, because (some version of) the Classical Case Filter of Rouveret and Vergnaud (1980) requires DPs to have case generally, if no other Case assignment applies, Case Marking is supplied by 90 the following highly general default lexical entry.18 (11) Lexical entry for Accusative Case Marking (Indo-European): [ ~t"N ]___, Dx, 7t accusative This perspective accounts for why accusative case appears after Ps of such varied meanings in (5), not just those of spatial GOALS. Consider the range of functions of the bold Ps with italicized accusative objects in (12), from Naughton (1999: 301-304). To je pro mě. Jdu pro mléko. Cekám na tramvaj. Jsem tu jen na týden. , , Jsi o rok starší než iá. (12) Pršelo (po) celý den. Ve středu Nevěříš v Boha? Koupil to za pět korun Udělám to za hodinu. Certainly these usages cannot result from some extension or contortion of a special rule involving motion toward. Alternatively, to claim that so many Ps take the accusative via lexical specifications would amount to a loving embrace of irregularity. In contrast, the simple claim here involves no lexical or grammatical stipulations; the objects of these Ps are accusatives merely by default. These D projections must have case, yet no rule spells out case on them other than (11). When (11) applies e.g. to an oblique DP in a Goal phrase or to the nominative Di of a neuter gender lexical subject, their syntactic case features in LF nonetheless remain respectively, DP and Di, even though the morphological realizations are phonologically accusative. As a default, rule (11) follows the other more specific Case Marking rules, i.e. those in Czech that spell out Ds as Instrumental (i.e. the Oblique Case DP), Genitives, and (Animate Gender) Nominatives. However, (11) never "overwrites" such spell outs, since it applies to only those Dx that have not been assigned phonological content.19 It moreover does not conflict with a requirement that distinct complement XPs of a single predicate exhibit different syntactic case features in Logical Form.20 'This is for me.' 'I go for milk.' 'I wait for a tram.' 'I'm here only for a week.' 'You're (by) a year older than me.' 'It rained (for) all day.' 'On Wednesday' 'You don't believe in God?' 'He bought it for five crowns.' 'I accomplish it (with)in an hour.' Thus, there is no special Accusative Marking statement that specifies the category V or Dv. The complex phonological alterations among accusative allomorphs, represented in (11) by 7raccusatiVe, are not treated here. One type of phonological content that can be specified in a lexical entry is a null allomorph, such as the Czech feminine and neuter genitive plurals. Such null morphemes must be distinguished from categories that are null for theoretical reasons. See Emonds (2005). This very general and intriguing pattern is called the Logical Form Case Filter in Emonds (2000: Ch. 8). It explains several previously mysterious or unremarked restrictions on combinations of XP complement types. One of the possibilities is having "no case" in Logical Form, a situation that arises with clausal complements and predicate attributes. 91 5 Locative vs. Instrumental: boundary and properties of Syntacticon items 5.1 A lexical entry for Czech instrumental endings For discussion, I repeat in (13) the two upper rows of Table (5): (13) Czech Prepositions not related to GOALs or SOURCES F; = -LOCATION -GOAL, -SOURCE P,-F = 0 'with' (instrument) = s(e) 'with' Not defined, since Dictionary Ps must be +LOCATION. F; = +LOCATION -GOAL, -SOURCE In G-B terms, this case is "oblique." The Czech term is instrumental, and is in red font here. In Latin, this is "ablative." P,-F = 0 'through, by' (bare locational DP) s x o pied sox za x o x mezi X o nad _s_(or other reference point) o x pod s (or other reference point) P,+F; = o 'about' = v(e) 'at, in' = pfi 'near, during' = po 'over, along' = na 'on' (This is the Czech "locative square.") As observed in textbooks of Czech, the instrumental and locative cases have in common that they "can denote a fixed location". In this way they contrast with genitives and datives, which "can denote a change of place" (Naughton 1999: 66). Interestingly, both the least semantically specific, s(e) and 0 'with', and the most semanti-cally specific of the P in (13) select object DPs in the instrumental. This is a typical "elsewhere" distribution. That is, a special statement is needed for the Ps in violet font (section 5.3 below), and then a simple default statement assigns instrumental case to the rest of the Czech oblique objects which are neither GOALS nor SOURCES, indicated here shorthand with -PATH. (14) Lexical entry for Instrumental Case Marking on Czech nouns, singular and plural: <[ +N ]____(PLUR)>, [ Dp, -PATH ] { m, i, 0 }21 Any phonological specifications in the entries of this study are meant to suggest only the essential aspects of the various suffixes, ignoring the phonetic detail. Thus, (14) doesn't attempt to specify the phonological realizations of the three instrumental allomorphs or the conditions on choosing among them.22 In proposing his Mirror Principle relating morphological and syntactic positions, Baker (1985) observes that case affixes generally follow plural affixes if the two are distinguishable. However, as (14) stipulates, the Czech instrumental suffix, whose most common realization is -m-, seemingly precedes its plural -y/i; adjectival inflection directly reflects this. That is, Standard Czech instrumental plurals always end in -y/i (a rare exception being dvema 'two'). Nonetheless, certain aspects of the morphophonology of Czech instrumental suffixes stand out: (i) The feminine singular allomorph seems to be a high vowel whose value for ±FRONT is determined by the stem-final segment, and then ... a«=>... ou, and (ii) the allomorph is otherwise-m or 0. The zero allomorph blocks palatalization of final non-palatalized consonants in front of the plural -y/i (nom. plur. pdni 'sirs' vs. instr. plur. pdny). 92 5.2 The nature of the instrumental/ locative distinction We next examine a difference in (13) between what I claim are open class Dictionary items (in column 3) and "grammatical" items (in column 2). In several studies, I have claimed that the defining difference between two entirely different types of lexical items resides in that fact that open class items are always associated with some purely semantic features f not used in syntax proper. Before going into this difference more, let us ask: Are there actually purely semantic features needed for the Czech Ps in the third column in (13)? I believe there are. The meanings of these latter P are indeed more complex than those of the P in column 2. This is obvious for mezi 'between'. The other Ps in column 3 require some kind of frame of reference, such as an observer or gravity (to determine up vs. down). Consider: if two rocky masses are floating close to each other in deep space with neither observer, star nor the earth as a known reference point, one could not be said to be "beyond/ in front of/ above/ below" the other (column 3). But they could be said to be "on/ in/ near/ with" each other (column 2). These reference points and an orientation toward them are thus part of the lexical meaning of only the Ps in column 3 and are indicated by purely semantic features fj.23 This conclusion is not far from that reached in an extensive discussion of the semantics of Place in Leech (1969: esp. sections 8.1-8.5). The English Ps translating Czech Ps that take locatives are said to involve only "dimensionality", while those Ps whose Czech counterparts take instrumentals express "relative position" and "orientation". For these latter notions, he introduces a number of features in addition to those needed for simple locations, which I take to be examples of semantic fj.24 Now exactly which features F are syntactic and which f are semantic is not obvious, prior to syntactic investigation or to constructing a lexical theory. But given the well justified open vs. closed class distinction in grammatical phenomena far from prepositional meanings, and given a methodological commitment to the idea that syntax is not arbitrary, we look for and actually find a needed distinction. In particular, the need for "orientation features" fj for the P in column 3 in Table (13) sheds light on what has previously looked like a semantic morass. That is, the different cases assigned by Czech Ps reveal the boundary between the syntactic feature complex [P, -PATH] and some further Ps with purely semantic orientation features fj. In several works, most collected in Emonds (2000; 2007), I have developed the idea that the timeworn distinction between open and closed classes is not some kind of continuum or some vague property of lexicons without central implications for syntax. In fact, I find that all of syntactic theory turns crucially on the different behaviors of these two kinds of lexical classes. The open classes are strictly limited to four lexical categories, N, V, A, and P, which I term the Dictionary. The closed classes include all other categories, and crucially closed subsets of the lexical categories, and this component I call the Syntacticon. In terms of this section, both s(e) and the locative-assigning Ps of Czech are in its Syntacticon, while its other Instrumental-assigning Ps are in its Dictionary. The limited scope of this study does not allow for any more than a summary of the fundamen- 23 Since the locative P u 'at the house of seems to have a semantic feature f indicating something like "the area/ house of, we might expect it to take instrumental case. But it takes a genitive, which I assume is "quirky", like the genitive with do in (10). Compare the French P chez, which has only the "house of meaning. 24 Leech's intuitive theory nonetheless fails to predict certain aspects of Czech case distribution. First, his semantic dichotomy would lead to expecting locative case for the bare "instrumentals of location" in (3), since these do not express orientation or relative position. Second, whenever a P of relative position implies further some kind of separation of two objects (bez 'without', krome 'except', koncem 'at the end of, nedaleko 'not far from'), Czech uses a genitive rather than an instrumental, suggesting a role for the feature SOURCE, which in Leech's system has no special status. I conclude that syntactic features and the structure of the lexicon make better case predictions than his purely semantic investigation. 93 tal differences between Dictionaries and Syntacticons, which I reproduce with minor changes from Emonds (2005a). Not all the terms in (15) are discussed in this study. (15) Dictionary Syntacticon Defining property: Items with purely semantic features / YES NO a. Syntactic properties: i. Grammatical categories in the inventory N,V,A,P ALL ii. "Late insertion" possible during syntax and at PF NO possible hi. Items with "alternatively realized" features; cf. (18) below NO possible iv. Full suppletion inside paradigms (go/ went;); cf. note 16 NO possible b. Phonological properties: i. Items conform phonologically to "primary vocabulary" possible YES ii. Bound items heading compounds have relatively less stress NO YES hi. Phonetically zero morphemes (see section 8) NO possible c. Intermodal and processing properties: i. Open classes; adults can coin neologisms YES NO ii. Interface with non-linguistic memory and culture YES NO hi. Processing look-up in terms of initial consonant cluster YES NO iv. Processing look-up in terms of syntactic addressing NO YES v. (?) Limited to Broca's area of the brain NO YES 5.3 The Czech "locative square": a lexical entry for the locative suffixes As seen in the previous section, the relevant P (o, v(e), pfi, po, na) requiring locative endings that "replace" the instrumental ones should be specified as [P, F;, -fj], i.e. they lack any semantically specific features f characteristic of open class Dictionary items. Yet in order to get locative case, P must have some contentful syntactic feature(s) F; other than P itself. For when P has no such additional feature, it is realized as s(e) or 0, giving rise to Instrumental case spell outs in (14). So we need to now determine what this F for locatives is. The key to the nature of F is a curious gap in the gamut of Czech prepositions. L. Veselovska (pers. comm.) has pointed out that at presents a difficulty for Czech-speakers learning English. This item is taught to Czech learners as conveying the sense that its object is "dimensionless" or a "point" or an "abstraction" lacking internal structure. (16) We should stop { at/ in } Barcelona for petrol. Cf. We should spend our vacation {in/ ??at } Barcelona. He won't stay at the University for long. The people at the pub were noisy. I saw a police car at the last stop light. 94 Translations of these PPs into Czech must use prepositions that imply some kind of dimension or relative position; see for example Peprnik (1995) who typically uses Czech u. On the possible "dimensionality" and quirky case of u, see note 23; English at has no such implications. That is, both at and with seem to lack a feature that Leech associates with (positive) "dimensionality". For this concept of conveying that an object DP has spatial dimensionality, I propose a syntactic feature of P labeled EXTENSION. Except for the completely unmarked s(e), all closed class Czech Ps that are -PATH are then +EXTENSION: o 'about', v(e) 'in', pfi 'near', po 'over, along', na 'on'. From this perspective, with is a pure exemplum of [P, -PATH] with no further specification. 25 These considerations allow us to write a lexical entry for the Czech locative affixes. The following entry does not reflect the similar morphological shapes of locative and dative singulars in Czech; section 7.2 returns to the possible significance of this. (17) Lexical entry for Czech Locative Marking: <[+N]___>, EXTENSION, nlocative We now come to the crux of the puzzling "locative square" in (5) and (13). Why does (17) never spell out locative case on objects of Czech open class Ps of EXTENSION, for example those meaning 'on top of or 'in front of or 'alongside'? A revealing answer apparently requires further restricting a fundamental notion that distinguishes Syntacticon from Dictionary items. The relevant principle, which I and others have justified in too many works to list, is given in Emonds (2000; 125): 26 (18) Alternative Realization ("AR"). A syntactic feature F associated in UG with category B can be alternatively realized in a closed class grammatical morpheme under X°, provided that X° is the lexical head of a sister of Bj. Now Locative Marking (17) is clearly a case of AR, where F is EXTENSION, B (and BJ) is P°, and X° is [+N]°. The closed class grammatical morphemes are the DP spelled as it locative. Note that , as indicated in (15a-iii), open class morphemes never exemplify AR. However, the earlier formulated Czech case assignments for genitives (8), accusatives (11) and instrumentals (14) are also legitimate and in fact typical instances of AR under heads [+N]°. In all three entries, P is again at least one of the values of B, while the copied features F of P here include P itself (that is, in all three rules P can be a value of X in the case feature Dx). Therefore, up to this point, the AR of Locative Marking presents no formal difference from the other case assignments. And yet it is empirically more restricted, since no open class Czech P can be spelled out as (i.e. assign) a locative case suffix. To express this difference, it appears that we must supplement AR with (19). (19) Restriction on open class AR. Lexical entries can alternatively realize features of open class (Dictionary) items of category B only if they spell out the category B itself. Though appearing at first glance ad hoc, this modification is an important and justified restriction. For example, it (apparently correctly) forbids say, different agreements on finite verbs that reflect only whether an open class subject noun is count or mass. And in the situation under discussion, since P is not mentioned in Locative Marking (17) and its appearance would 25 In the terms developed here, we can probably maintain that Ps of PATH never have an EXTENSION feature. That is, PATH and EXTENSION are mutually exclusive. 26 A summary of the many constructions from many different languages that have been or can be handled in terms of AR appears in Emonds (2000, section 4.4 and 4.5). 95 be superfluous, the restriction (19) derives the fact that (17) cannot apply to object nominals of open class Czech P. 6 The crucial characteristic of true case-marking It should now be noted that Locative Marking (17) is formally not actually a "case-marking rule", because it specifies no category with a case feature. In this study Case Assignments, as seen in (8), (11) and (14) always have entries containing Dx where X is a case-assigning category. Leaving aside for the moment a possible relation to the dative, this lexical entry (17) for locative endings simply spells out special endings on nominal categories +N in the presence of certain prepositions. These spell outs amount to a more salient way to signal the presence of the special feature complex [P, EXTENSION, -fj.]. For convenience, though the term has no real theoretical status, we can call such a pattern, in which AR spells out secondary features of a case-assigner on the assigner's object, a "quasi-case". Thus, the Czech locative is a quasi-case, as is the Latin dative, compared to its true oblique case, the ablative (see note 9). This asymmetric relation between Instrumental Case Marking (14) and Locative Marking (17) can explain why the latter takes preference over the former. A first way to explain this might invoke Panini's and Kiparsky's "Elsewhere Principle", by which rules that apply in a more limited context, here (17), have precedence over those such as (14) that are more general. Another line of thought may be more revealing. In a wide variety of syntactic patterns in terms of AR, the grammatical morphemes that alternatively realize canonical positions and features are not themselves part of LF. That is, such AR morphemes, for example those of quasi-cases, are not interpreted in their surface positions. They appear in these positions only subsequent to "Spell Out" and so are inserted in PF. However, Case Assignment rules such as (14) plausibly apply during a derivation prior to Spell Out (and are used in LF). If so, when Locative Case (17) applies in PF, it must "overwrite" the instrumental case spellings, thus rendering ungrammatical any instrumental case on objects of Czech grammatical Ps of EXTENSION. Such overwriting plausibly distinguishes true case lexical insertions used in LF (they never overwrite each other), from quasi-cases used only in PF (they overwrite previous Spell Outs). In any case, whether due to the Elsewhere Condition or to a difference in component (narrow syntax vs. PF), Czech Locative Marking, linked to the specific feature EXTENSION on [P, -GOAL], pre-empts, overwrites, or otherwise precludes its Instrumental Case Marking. Just as the Latin dative is not an autonomous case (Emonds 1985, Ch. 5), I conclude that there is no autonomous locative case in Czech. Its locative DPs are indeed oblique case complements of [P, -GOAL], but additionally alternatively realize the syntactic feature EXTENSION of this introductory P. The new Restriction on AR (19) predicts why only a small subset of grammatical [P, EXTENSION] can take such locative objects. This account thus reduces Czech non-vocative cases to five (nominative, genitive, oblique, accusative and dative), which is closer to but still one more than the four expected by the theoretical approach to case summarized in (1). 96 7 The Czech "dative square": a second quasi-case 7.1 Alternative Realization of the feature GOAL Let us now focus on the remaining Czech case that falls outside of the restricted lists of (1) and (2), namely the dative. Datives with no overt P are used for indirect objects and also for "benefactive" nominals as in uvařil kamarádovi oběd 'made lunch for a friend'. Dative case is also always found with the grammatical P k(e) 'toward'. But overall, it is striking in how few structural contexts this case appears. Dative seems to simply spell out (alternatively realize) the feature GOAL of a P which has no purely semantic features f.27 The preposition ke has the syntactic feature -CONTACT in many if not all of its uses. Although do 'to' expresses [GOAL, +CONTACT], its object is a lexically stipulated quirky genitive (10). 97 (20) Lexical entry for Dative Marking on Czech Nouns:28 r , f PLUR, (vowel)m ) <[+N]___>, GOALJ ,K }~ } I high vowel \ Besides the Syntacticon item k(e), a very few Dictionary items such as díky 'thanks to', vůči 'compared to', kvůli 'because of, (na) proti 'against, opposite to' are generally listed as prepositions that take the dative. Given their lexical specificity, we can plausibly maintain that these few open class items take "quirky datives," consisting of lexical specifications for DP complements with a GOAL feature. Since the only way a DP (rather than P) can have such a non-canonical feature is via AR (18), these quirky datives are necessarily of the form [PP [p, goal 0 ] - DP ], which structure is identical to that of indirect objects and benefactives. Their dative inflection is then spelled out by (20). Since the feature GOAL is a canonical feature of P, instances of GOAL being spelled out in dative inflections are further instances of AR. Again, it seems that the complex of P features copied in these inflections cannot have any purely semantic features f, i.e. Dative Marking (20) like Locative Marking (17) is limited to, and is therefore a test for determining, which Ps are in the Syntacticon. Since do 'to' takes quirky genitives, the remaining GOAL Ps are k(e) 'to/ toward' and 0. The Czech dative is thus not so much a case used in several structures as a means of morphologically realizing on DP the presence of a minimally specified P expressing GOAL. In this it completely resembles the Latin dative. The dative is thus also a quasi-case; whenever such Ps of GOAL have any further specification in terms of semantic features f, then in fact their objects appear with the accusative case, as seen clearly in the right column of Table (5). Therefore, there is no special oblique case for these GOALs, only the default "super case" of accusative, which is assigned by the general default entry (11). 7.2 Speculations on the "morphological (in)dependence" of quasi-cases There is nonetheless an apparent difference between Latin and Czech datives with respect to their morphophonological realizations. The Latin dative is strikingly parasitic on the true oblique case, differing from the ablative only in the singular, and then only on some noun classes and some feminine agreeing adjectives. And in the patterns where a Latin dative singular differs from the ablative, it always spells out the same way, -i. That is, a single morphophonological statement specifies completely the phonological differences between its datives and its ablatives (Emonds 1985: Ch. 5). We can call this property the "morphological dependence" of the Latin dative quasi-case. Several textbooks on Czech suggest that its locative is dependent (in the same sense as the Latin dative), though at first paradoxically, this dependence seems to partly be on the dative, itself a quasi-case. Specifically, the locative plural forms depend on genitive morphology and the singular forms on dative morphology. Hence, rather than being phonologically independent as presupposed in (17), perhaps Czech locative morphology somehow "piggybacks" on that of 28Much of the phonological detail, e.g. exactly which vowels appear in dative suffixes, is due to Czech (morpho-)phonology. Of interest is that the dative "high vowel" corresponding to -a in singulars is a palatalizing -e, which perhaps should be represented as-ye with a high on-glide. In certain paradigms, stem-final unpalatal-ized dentals are separated from the dative suffix -i by means of epenthetic -ov-, possibly a "conspiracy" to remain unpalatalized (Czech datives don't seem to induce any palatalization that is not arguably part of the stem). 98 true cases. Here is a possible scenario. Let us first look at the relation in plurals. Czech genitives independently require two allo-morphic statements whose detailed phonology, as throughout, we must leave aside. Also as throughout, Dq represents genitive case.29 (21) a. Genitive Plural on Nouns. , [ Dq, PLUR ], { u, i, 0 } E.g. muž-ů 'of men', radost-i 'of joys', měst-fy 'of towns' b. Elsewhere Genitive Plural.<[+N]___>, [ Dq, PLUR ], (front vowel)ch Nouns of adjectival form such as hajných 'of gamekeeper's wives' and vstupných 'of entry fees' follow this pattern. Now it turns out that (21b) is used only for genitive and locative, possessive pronouns, demonstratives, etc., but also for locative plural nouns N. So we need a statement (22) stating that in PF (not in LF) these nouns acquire a feature of an adjective [+N, +V] that contrasts with some feature of a noun [+N, -V]. (22) Locative . DP [ DQ , +V ] / [ +PLUR, +EXTENSION,___] Singular locatives are more closely related to datives, but nonetheless involve some rather minimal phonological differences: (23) Locative. DP GOAL / [ -PLUR, +EXTENSION,___]; further, (i) delete -u following to (in modifiers), and (ii) -u =>- -ě in -ANIMATE nouns. Keep in mind that if the "syntactically formulated" (22)-(23) replace the independent morphology statement for locatives in (17), they are nonetheless both instances of AR (18) and so have effect only in PF. That is, the syntactic features they introduce, namely Dq, +V and GOAL respectively, have no relation to LF interpretations. (22)-(23) express patterns that are remarked, even emphasized, in Czech textbooks. They give the impression that locative morphology pervasively depends on dative morphology. And in fact, since (22) contains only syntactic features and no phonology, (23) does constitute a single morphophonological statement that specifies completely the phonological differences between Czech locatives and its genitives and datives. Hence, exactly as the Latin dative is related to its ablative (oblique) case, the Czech locative quasi-case is morphologically dependent on its genitive and datives. In view of this conclusion, it is tempting to ask whether even Czech datives might be dependent on its true oblique instrumental case. Probably only a "believer" in this study's approach, for example the author, might try. But it is curious how often the Czech datives can be derived from instrumentals by applying the following rough "switching" algorithm: (24) Recipe for Czech dative forms based on instrumentals. (For this recipe, a final surface -ou must be treated as a form with an underlying final -m.) a. If an instrumental feminine singular noun ends in a vowel, make it short. b. In other instrumentals ending in a vowel, do the minimum to make them end in -to. c. If an instrumental ends in -to, do the minimum to make it end in a vowel. In this paper, I don't really stand by this recipe - or even by (22)-(23) - because they all depend 29 My contention is that true cases are morphologically complex and thus independent. The fact that the morphology of a true case is irreducibly complex, as in (21), supports this view. 99 on unformulated "universal optimality phonology" that determines what constitutes "doing the minimum". But it seems natural that vocalizing a final -to consists in retaining its roundedness so that the corresponding vowel is u. In some perhaps exceptional situations, this recipe may falter or need modification. More tinkering is not the purpose of this study, but the very fact that a switching algorithm seems psychologically plausible supports (25). (25) We cannot be sure whether Czech dative suffixes are morphologically independent, like those of a true case, or morphologically parasitic on the instrumentals. In particular, since it is not implausible that Czech speakers might internalize a kind of "vowel vs. -to switching" to make datives, a rather strong hypothesis can at least be entertained: (26) Quasi-case affixes. These differ from true cases (i) in spelling out only Syntacticon items (19) and additionally (ii) in being morphologically parasitic on true case affixes. One can of course question whether Czech children can internalize three non-trivial algorithms such as some perfected versions of (22)-(24). The alternative is that they internalize the Declension Tables (spread out before me) that contain more than 100 allomorphs for datives and locatives alone, which are presented as somehow independent and moreover don't include clitics and all the possessive pronoun forms. Does the Czech brain learn 3 or so clever tricks at an early age, or is it doomed to plod through memorizing 100 plus allomorphs (for only 2 cases)? That is the question posed in this subsection. 8 Null vs. Phonological Spell Outs of minimally specified P Table (5) of Czech prepositions and associated cases embodies the claim (15b-iii) that only Syntacticon items, and not open class items, can have null allomorphs. Some detailed implications of this claim are discussed in Emonds (2005b). Actually, most syntactic frameworks take for granted the possibility of phonologically null grammatical items, and do not propose null open class items. In view of this, a central question is, under exactly what conditions can a grammatical P be null? Let me state here a principle I have defended in many studies which answers this question. (Emonds 1987; 2000: 135) (27) Invisible Category Principle (ICP). If all marked canonical features F on B are alternatively realized, except perhaps for B itself, then B may be empty.30 Now Case Marking lexical entries that spell out the various values of Dx as in (2), are a sort of formal prototype of Alternative Realization (see again section 5.3). Consequently, if all the marked features of such a P are spelled out in its object, P can be empty.31 We can see one effect of (27) with the use of dative case for expressing indirect objects. Cross-linguistically, in languages without morphological dative case, indirect objects can almost invariably and usually must be introduced with a minimal P expressing GOAL (English to, French a, Japanese ni, etc). Presumably, such P spell out the feature complex [P, GOAL, CONTACT]. Moreover, since to (French d) seems less marked than toward (French vers), 30 As discussed in section 4.4.4 of the work cited, the fact that a category may be empty almost always leads to the situation where Economy requires that it must be empty. 31 The Accusative Case Marking entry (11) does not alternatively realize the case category P nor the feature GOAL. Therefore, it cannot license an empty unmarked P of "motion toward". 100 +CONTACT is plausibly the unmarked value of this feature. Hence in order to satisfy the ICP (27) only the marked canonical feature GOAL of an indirect object P need be alternatively realized. Since the Czech entry (20) indeed alternatively realizes GOAL as dative inflections, this P introducing its indirect objects is 0. Minimally contrasting with this null P is the preposition k(e), which spells out [P, GOAL, -CONTACT]. Here -CONTACT, a marked canonical feature of P, is not alternatively realized. Hence this P must be phonologically overt. A similar dichotomy distinguishes the "bare instrumentals" of location in (3), from PPs introduced by the overt P s(e) 'with'. Keeping in mind that the LF interpretation of P is simply LOCATION (note 11), bare instrumentals express semantically a vague sense of the place of the action. Moreover, even the use of this case for which it is named, for expressing instruments, apparently has an unmarked P similar to that of indirect objects. That is, an instrument as in jet autem 'go by car' structurally indicates nothing more than a general location of the action; implying that "instrument" is not actually an explicit LF concept. The notion of "means" or "instrument" must be no more than pragmatic implication, due to cars being generally items designed for purposes of transport.32 Now the situation is different with accompaniment, jet autem se studentem 'go by car with a student'. The student is not in any sense the locus of the going, but in this phrase rather more like someone who is also going, a sort of secondary agent. That is, in "accompaniments" the sense of unmarked location is absent in LF. To express this, I use the following lexical formalism, developed in more detail elsewhere. As in note 11, a P present in LF signifies LOCATION in the most general sense. Similarly, all other syntactic categories have such general senses: D has reference, A indicates properties, V indicates actions, etc. I claim that a marked syntactic "cancellation" feature +0 allows the basic categories to lose their general interpretive sense: [D, +0] loses reference (in expletives); [A, +0] loses its property interpretation (e.g. in verbal passive participle affixes); [V, +0] loses its activity interpretation (in stative verbs), etc. Similarly, those Ps which do not denote any type of location, such as English of, agentive by, the with of accompaniment, despite, etc. are all similarly lexically specified as [P, +0]. Such is also the situation for Czech [P s(e)\ 'with' when its object represents (a) accompaniment, (b) emotion or (c) mental perception or activity. The P carries no sense, even a vague one, of the action's physical location. Hence, the locational nature of P must be cancelled in LF by the specification [P, +0]. This marked feature +0 precludes a null P, so no s(e) in (28) can be omitted. (Examples of M. Martinková gratefully acknowledged.) (28) a. Tom vešel s kamarádem dovnitř. 'Tom went with a friend inside.' Objednali si margaritu se soli. 'They ordered a Margarita with salt.' b. Sel do toho s velkým odhodláním. 'He went into it with great determination.' Mluvil o ní s láskou. 'He spoke about her with love.' c. Chci pokoj s výhledem na moře. T want a room with a view on the sea.' S chutí začali pít, ale pili s mírou. 32 Of course, when something is used as an instrument, there is no interest in more detail about its location with respect to the action, so no specific preposition is used. 101 'With gusto they started to drink, but they drank with restraint.' There are no different "senses" of with here, only a general means of syntactic juxtaposition lacking any implication of location, in both Czech and English. Now, the cancellation feature +0 of s(e) and with is both marked and unrelated to any Alternative Realization. So if s(e) with the feature +0 were not spelled out, the ICP (27) would be violated. The cancellation feature +0 thus explains the alternation pattern of s(e) with a null morpheme in the top row of Table (5). The ICP is an integral part of this explanation, as it is also in explaining the difference between Czech datives with and without k(e). 9 Conclusion: Czech has the same four structural cases as other l-E languages. We started by observing that Czech Ps seem to "take" five different morphological cases on their objects, counter to some more studied I-E systems which apparently allow only three such cases. But we have now seen that the Czech locatives and datives are not actually full-fledged cases. Like the latter, they alternatively realize the categories and features of introductory Ps. But unlike them, lexical entries for datives and locatives don't spell out case features of the form Dx as in (2), but rather only syntactic sub-categories of P: GOAL and EXTENSION. Consequently, the "fine-tuning" of Alternative Realization (18), by means of the new Restriction on AR (19), predicts that dative and locative assignment is permitted only in the presence of closed class P, that is, Ps that are in the Czech Syntacticon. Turning around the perspective, studies of the Dictionary/Syntacticon dichotomy have not previously had clear criteria for assigning individual Ps to one or the other lexical component. The syntax of Czech cases, however, provides sharper ideas and criteria. For example, there are very few GOAL Ps in the Czech Syntacticon, only k(e) and 0, while there are somewhat more LOCATION items, namely the Locative-assigning Ps that are core items expressing EXTENSION or what Leech (1969) calls "dimensionality." The Czech special "quasi-cases," the Locative and the Dative, interestingly share many phonological spell outs (section 7), which further reinforces their similar theoretical status. Beyond this, however, the Czech syntactic system of morphological cases and of Ps does not differ in any essentials from those of its non-Slavic I-E relatives, especially Classical Greek, with which it shares the properties of (i) using Genitives for "motion away from" and (ii) expressing objects of Dictionary Ps, many expressing GOALs, with what is, following the suggestion of L. Veselovska, the unmarked structural case Accusative. To me, the most intriguing remaining question is how to formally express in the lexicon or PF the similarities of Czech datives and locatives regularly remarked in books that teach Czech. These similarities, once they are captured, should shed some light on a notion of "morphological (in)dependence." That is, the allomorphs of the true case suffixes naccusative, ^genitive and it oblique, seem to be significantly more diverse (and different from each other) than the quasi-case suffixes ^dative and ^locative- But for the moment, I cannot fully explain the phenomena, but only note that Czech textbooks suggest that it is a very real one. References Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16:373-416. 102 —. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Brame, Michael. 1984. The Head-Selector Theory of Lexical Specifications and the Nonexistence of Coarse Categories. Linguistic Analysis 10:321-325. Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foris. —. 1995. The Minimalist Program. Cambridge: MIT Press. Emonds, Joseph. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris. —. 1987. The Invisible Category Principle. Linguistic Inquiry 18:613-632. —. 1993. Projecting Indirect Objects. The Linguistic Review 10:211-263. —. 2000. Lexicon and Grammar: the English Syntacticon. Berlin: Mouton de Gruyter. —. 2003. Two Types of Lexical Access: Why Syntacticon Items can be Null. Theoretical and Applied Linguistics at Kobe Shoin 6:1-24. —. 2005a. The Computational Lexicon. English Linguistics 22:232-266. —. 2005b. Syntactic Conditions on Phonetically Empty Morphemes. Broekhuis, H., N. Corver, R. Huybregts, U. Kleinhenz &; J. Koster (eds.). Organizing Grammar. Linguistic Studies in Honor of Henk van Riemsdijk: 111-121. Berlin: Mouton de Gruyter. —. 2007. Discovering Syntax: Clause Structures of English, German and Romance. Berlin: Mouton de Gruyter. Holá, Lída. 2000. Czech Step by Step. Prague: Fragment Publications. Jackendoff, Ray. 1977. X-bar Syntax. Cambridge: MIT Press. —. 1983. Semantics and Cognition. Cambridge: MIT Press. —. 1990. Semantic Structures. Cambridge: MIT Press. Leech, Geoffrey. 1969. Towards a Semantic Description of English. London: Longman Group. Marslen-Wilson, William D. 1987. Functional Parallelism in spoken word recognition. Cognition 25:71-102. —. 1990. Activation, competition and frequency in lexical access. Cognitive Models of Speech Processing. G.T.M. Altmann, ed. London: MIT Press. Naughton, James. 1999. Colloquial Czech, 2nd edition. Oxon: Routledge. Niederle, Jindřich, Václav Niederle, and Ladislav Varel. 1991. Mluvnice řeckého jazyka. Praha: Scriptum Publishers. Peprník, Jaroslav. 1995. Angličtina pro pokročilé. Olomouc: FIN publishers. Riemsdijk, Henk van. 1978. A Case Study in Syntactic Markedness. Dordrecht: Foris. —. 1998. Categorial feature magnetism: the endocentricity and distribution of projections. Journal of Comparative Germanic Linguistics 2:1-48. —. 2002. The unbearable lightness of GOing. Journal of Comparative Germanic Linguistics 5:143-196. — and Riny Huijbregts. 1998. Interface in Space-How Natural Language Expresses Spatial Relations. Paper presented at the Third International Symposium on Language, Logic and Computation. Batumi. 103 Rouveret, Alain and Jean-Roger Vergnaud. 1980. Specifying Reference to the Subject: French Causatives and Conditions on Representations. Linguistic Inquiry 11:97-201. Veselovska, Ludmila. 1998. Possessive Movement in the Czech Nominal Phrase. Journal of Slavic Linguistics 6(2):255-300. Veselovska, Ludmila. 2001. Agreement Patterns of Czech Group Nouns and Quantifiers. In: Corver, N. & H. van Riemsdijk (eds.). Semi-Lexical Categories: 273-320. Berlin: Mouton de Gruyter. 104 Mixed Nominals in Czech Petr Karlík pkarlik@phil.muni.cz 1 Introduction In the generative tradition, the mixed categories pose a problem from the very beginning. The thing is that regarding the postulate of endocentricity of syntactic constructions, in the lexical-based models the structures showing morphological and syntactic properties of two different "categories" X and Y (even though their lexical heads are single words, i. e. X or Y) present a potential problem. Standard generative insight into this type of mixed categories is that they have the external syntax of a X projection and the internal syntax of both a Y projection and of X or Z projections. For mixed categories, it is typical that a phrase, exclusively a constituent of XP, co-occurs with a constituent exclusively found in YP. Another specific property of mixed categories is their internal word structure the root of which is a lexical head of a mixed phrase. If, according to the lexicalist hypothesis, syntactic categories as V, N, A are related to morphological forms, mixed categories prototypically show anomalous morphological forms of words as noun, verb, adjective: it often happens that mixed category phrases are headed by words which appear to be morphophonologically ambiguous or neutral between the two categories of the mixed phrase. A stereotyped example which complies with the above mentioned description of a mixed category are Italian phrases headed by the infinite/ sostantivato (1). The distribution of these phrases is identical with the distribution of DPs, see (la) and (la1), while their internal syntax is hybrid: In the internal structure, determiners, possessive and qualifying adjectives, i. e. expressions being typical constituents of a phrase with a lexical head N appear before eseguire. As for the postpositions, eseguire can be followed by a direct object and by adverbs, i. e. expressions being typical constituents of a phrase with a lexical head V: (1) il suo continuo eseguire la canzone impeccabilimente the his/her continual perform.INF the song impeccably 'his continually performing the song impeccably' (Zucchi 1993:55) a [il suo continuo eseguire la canzone impeccabilimente] mi piaceva a' [la cancone] mi piaceva Another instance of a mixed category are English phrases headed by the verbal gerund (2). Identical distribution as with DP shows their external syntax of nominal projection, see (2) - *The present study is a part of the project MSM 0021622435, the goal of which is an analysis of the Czech mixed categories from the diachronic point of view. The present-day Czech data and their analysis should be one of the conceptual and empirical diagnostics for this research. 105 (2a), their accusative case assignment and adverbial modification show their internal syntax of verbal projection, see (2) - (2b). At the same time, in the internal syntax of those phrases, expressions appear which are known as typical constituents of NPs, namely possessives (2'), but not determiners (2"). Further, there can be occurrences of expressions which are not constituents of noun phrases or unembedded vPs or higher projections containing the vP; e.g., in (2'") the subject receives accusative case. Morphonologically, verbal gerunds are ambiguous between the two categories of the mixed phrase as well (vdestroying the book x Ndestroying of the book): (2) [John deliberately destroying the book] annoyed everybody (Alexiadou (2005), slightly modified by the author) a. [my friend] annoyed everybody b. John deliberately destroyed the book (2') John's / his destroying the book (2") *that / *the destroying the book (2"') John / him destroying the book In Czech, syntactic mixed categories carrying nominal and verbal properties similar to Italian phrases headed by the infinito sostantivato or English phrases headed by the verbal gerund apparently do not exist. On the other hand, phrases headed by words like (a) stavění, (b) stavba seem to be potential candidates for mixed categories, even though their properties are different from those defined as mixed category typical. The external distribution of these phrases is identical with DP (3) - (3a) and their internal structure shows prototypical morphosyntactic features of nominal phrases, i. e. no accusative case assignment but genitive case assignment (3b), and no adverbial modification but adjectival modification (3c). In common with DPs, possessives can occur within them (3d). Moreover, nominal properties can be seen in the inflection of stavění and stavba realizing the features Gender, Number and Case: (3) [stavění / stavba domu] ho unavil-o/a a. [Marie] ho unavila b. staví dům x stavění / stavba *dům stavění / stavba domu c. rychle staví x *rychle stavění / stavba rychlé stavění / stavba d. Petrovo kolo Petrovo/a stavění / stavba x *Petrovo čte The aforementioned properties of Czech phrases headed by stavění and stavba make them comparable to English phrases headed by the nominal gerunds (mixed nominalizations), as (4), and to phrases headed by the derived nominals, as (5), with both external and internal syntax of DP.1 Those English phrases - unlike the phrases with verbal gerunds2 - do not have accusative case assignment, but an of-PP one (see (4), (5) x (2)), they can have a determiner (while this is impossible for the verbal gerunds (see (5a) x (2"))), they have adjectival modification while the verbal gerunds have adverbial modification (see (5b) x (2)), etc.: 1In the generative tradition, the differences among verbal gerunds, nominal gerunds and derived nominals and phrases headed by them have been described innumerable times, from Lees (1960), Chomsky (1970), Abney (1987), Grimshaw (1990), Marantz (1997) to Harley & Noyer (2000), Borer (2003), Harley (in press) and others. The mentioned studies have been used by current authors as argumentation in favour of influential generative theories. It is not my intention to comment on all the mentioned analyses, nevertheless, I am going to use some of the data presented in them. 21 take into account only the differences relevant for the analysis of Czech data. Therefore, I omit the differences concerning particle shift, e. g., H. Harley & R. Noyer (1998). 106 (4) [John's destroying of the book] annoyed everybody (5) [John's destruction of the book] annoyed everybody a. the destroying / the destruction of the manuscript b. the deliberate destroying / destruction of the manuscript Verbal features are essential in order to consider the phrases headed by words of the type stavění / stavba with distribution identical with NP/DP instances of a mixed category. In both the external and in the internal syntax of the phrases sub (3) no verbal features can be seen so far; nevertheless, they can be expected at least in the internal structure of the words whose root is their lexical head: there is an apparent relation between stavění / stavba and stavět and, if there is (in the lexicalist framework) [ystav-ě-t], there must be a derivation towards []Ní[N[N[vstav-ě] Nn] ?i] -inflection], [Nevstáv] Nb] -inflection]. Verbal gerunds having syntax properties similar to related VPs, while nominal gerunds and derived nominals have the properties of NP/DP, predict that an auxiliary can be present within phrases headed by the former (6) and, that this is not possible within phrases headed by the later (6a), a well known fact: (6) John's having criticized the play annoyed us a. * John's having criticized of the play annoyed us For this paper, another well known fact is more important, namely that in auxiliary constructions aspect can be expressed on the verb, but not tense or mood: (7) Mary's eating the last piece of cake (Baker 2005:11) a. Mary's having eaten the last piece of cake b. *Mary's will eating / willing eat... c. *Mary's might eating / mighting eat... It is exactly the potential to express aspect (but not tense and mood) which lines up the Czech phrases headed by stavění with the English phrases headed by the verbal gerund: Part of the internal structure of the word stavění can be the same affixes which in the parallel Czech VPs realize either the aspect: prefixes (8) or suffixes (8'), or the so-called Aktionsarts: prefixes (9), or suffixes (10). On the other hand, in the Czech phrases headed by stavba the prefixes and suffixes are present in the internal structure of the phrase just exceptionally (which may be idiosyncratic), see (8) - (10) x (8a) - (10a). At the moment, we do not have any evidence to claim that in both types the affixes realize the aspect identically with VPs (to be mentioned furthermore): (8) stavění- -dostavění st avětip^dost avětpf a. stavba- -dostavba (8') dostavění- -dostavování dostavětp^dostavovatipf a. dostavba- -0 (9) střelení- -zastřelení Střelitp^zastřelitpf+feature x a. střelba- -*zástřelba (10) hraní- -hrávání hrátip^hrávatipf+feature y a. hra- -* 0 The present data provide some preliminary support for a hypothesis that phrases headed by the words of the type stavění are "more verbal" then the phrases headed by the words of the 107 type stavba. I will try to support this hypothesis by an analysis in the non-lexicalist theoretical framework because in the lexicalist analysis, similar conclusions were reached by L. Veselovská (2001). Since I am not a generativist, my argumentation will consist in gathering of the relevant data and in relating them to classical analyses of nominalizations. 2 Analysis Starting from the standard non-lexicalist hypothesis (A. Marantz, 1997, and many others), according to which the nominalizations are Spell-Out categorially neutral ^/P in a D-context: An acategorial root becomes a nominal by a head raising of the structure containing it (and any features associated with typically verbal features) to a nominal head realised by a nominal suffix. Tentatively, it can be presupposed that if the structures of the type stavění and stavba are nominalizations, then they are structures on Spell-Out which are a result of a derivation: this derivation is entered by a categorially neutral lexical head root ^/stav and the categorial properties of the projected phrase are determined by the functional heads received during the syntactic derivation. It is obvious that structural portions of such a projected phrase must be smaller than TP because (among others) there is no Nominative in the phrases with stavění a stavba: Petr čte x *Petr četba / čtení. 2.1 The first step of the analysis (with a direct empirical support) means to distinguish the nominalizations of the types stavění and stavba. I am using an independent observation of T. Scheer (2001): The quantity of a vowel in the prefix za- is determined by a [verbal] / [nominal] feature quality of the first suffix adjoined to a root. If the suffix bears a feature [nominal] the vowel is long, if the suffix bears a feature [verbal] the vowel is short. I have verified that this finding is generally valid for all the prefixes with a final vowel.3 (The addition is that the second step of the Scheer's algorithm is the following: A long prefix gets reduced if the root contains a long vowel (the law of the three mores): zá-stav 0-a x za-stáv-k-a.) The data in (11) show that the prefixes in both types of nominalizations behave differently: a. pří-SPĚv-(e)k vý-hr-0-a ná-stav-b-a zá-klop-k-a (11) b. při-spív-á-n-í vy-hr-á-n-í na-stav-e-n-í za-klop-e-n-í c. při-spív-á-m vy-hr-a-(j)i na-stav-í-m za-klop-í-m The structures of the type stavba show at their prefix (which can be a part of them) that the first suffix merging to the right with the root bears the feature [nominal]; these are the root-derivates; see (a). On the other hand, the structures of the type stavění show on the prefix that the first suffix on the right merging with the root has the feature [verbal], see (b). It is confirmed by the examples in (c); therefore these cannot be root-derivates, but necessarily stem-derivates. It means (if this analysis is correct) that a mixed category, under standard concepts a phrase whose categorial features appear to change at some point in its projection line (P. Ackema &; A. Neeleman (2004), among others), are only the nominalizations of the stem-derivate type because only those provide the change of the categorial features in the course of derivation: At first, in the syntax the acategorial root obtains verbal features through merging with a Theme-head and then, in the certain phase of derivation as a result of a merge 3A more complicated situation arises if together with the vowel quantity its quality changes: both průřez and prořez, but, expectedly, průvod, průchod... and, unexpectedly, prodej, prohoz. 108 of this "verbal" structure as a complement of a functional head little n, the categorial features of a phrase change. 2.2 I propose a common feature of all root-derivates is that they are licensed by a feature [event], which the root obtains from Encyclopedia: ^PÁB - pád-fy(-fy/...) (meteoritu), yOÁS - jas-ot(-fy/... ) (diváků), yVoL - vol-b(-a/...) prezidenta etc. This hypothesis is supported by the fact that the roots without an event structure, e. g., -\/bagr, -\/such, are from Encyclopedia furnished by the features making them ready to merge into a D-context: bagr-fy(-fy/...), or into an A-context: such-$(-ý/...). If they are to be interpreted as mixed nominals, they need a larger portion of a verbal-associated structure, minimally a ThP, because the Theme-head is the node that furnishes the embedded structure with the features contributing to its event-interpretation, i. e. with the features obtained by the roots of the type already in Encyclopedia, not in syntax. Therefore: lov-0(-0/. ..) x *bagr-0(-0/...), but : bagr-o-vá-n-í(-0/...); *such-0(-0/...), but: suš(-i)-e-n-í(-0/...), or sch(n)-u-W(-0/. ..); as for the analysis of root-derivates see P. Karlík (in press). 2.3 On the other hand, even though the stem-derivates are furnished with event features of the root from Encyclopedia, they are not licensed by it. See the contrast: -\/chod+event: chodSTOOt.nom - choz-e-n-ístem.nom x ^mlad.event: 0root-nom - mlád-nu-t-ístem.nom / (o)mlaz-e-n-istem-nom- Therefore, they can be derived from all the structures furnished (by the merge with the Theme-head in the syntax) by the feature [verbal]:4 [uč-e]-n-í, [[uč-i-tel]-ov-á]-n-í It is not quite sure whether the portion of a structure which is merged in the D-context is smaller than vP or whether the vP is contained in it. I will try to summarize the data which can be a basis of an adequate analysis. 2.4 Observing the morphophonological structure of the stem-derivates represented by, e.g., (12), (12) děl-á-n-í(-0/...) we can see that it contains minimally terminal nodes filled by Vocabulary Items (VI) -á-, -n-, -i and the inflexion material -0/... (The structure (12) does not provide any evidence for terminal nodes filled by a 0 element, nonetheless, they can be assumed, cf. do-děl-á-vá-n-í-). At this moment, I pass over the realization of features in the head Th (here, by VI -á-), supposing with a high level of probability that those features are of a verbal nature (supporting data see above). I consider important to find data which could be a base for reasoning what kind of features are realized by VI -n- (standing in complementary distribution with -t- (mlaz-e-n-i / mlád(-n)-u-t-i), uninteresting for the present paper) and, in what kind of a head? To compare (12) with (12a) and (12b) may help to find the solution: (12) a. děl-á-n(-0/...) 4In the first place, stem-derivates are not derived from non-lexical verbs which do not take aspectual prefixes but only suffixes: muset/musívat (číst) - 0, stát se/stávat se (učitelem) - 0, nechat/nechával (ho zavřít) -0. 109 b. děl-a-n(-ý/...) There is a reason to ask a question whether the attachment of the identical material (-n-/-t-) to the stem of the stem-derivates (12), passive participles (12a) and passive adjectives (12b) is random or motivated. One of the options is that this suffix is inserted into a head responsible for a passive morphology. Stem-derivates, passive participles and adjectives share a part of the morphophonological form [děl-á-n-]. At the same time the head into which (in passive participles) VI -n- (or -t-) is inserted is, according to classical analyses, responsible for the passive nature of the phrase. If we accept Harley's (in press) hypothesis, that the analysis of structures proposed for a form must also be contained within the analysis of any structure derived from that form,5 it necessarily follows that also the stem-derivates must be structures with a passive morphology. The idea that all the nominalizations (more exactly: all the process nominals), including structures without a visible passive morphology, are passive structures has been known since H. Borer (2001) and others. Borer's hypothesis has been used by M. Engelhardt &; H. Trugman (1998) to explain (13) in Russian: (13) konspekt lekcii brata x *konspektirovanie lekcii brata According to this analysis, the nominal konspekt does not contain any VP: the genitive case of the subject brata is licensed by D, whereas the complement lekcii is inherently case-marked by the lexical head konspekt. But on the other hand, konspektirovanie contains a projection of V incorporated into the head noun: genitive internal argument occupies the [Spec,VP] position of the VP contained within the DP (hence it is subject) as a result of passive formation prior to the incorporation of V-to-N. Within process nominals, the internal argument being the subject, not the complement, the only source for the licensing of the genitive case is D. The proposed analysis claims that the stem-derivates contain a large portion of a verbal structure including a vP projection. The putative evidence supporting the analysis of process nominals as passive forms concerns: (a) the availability of instrumental agents (see further), (b) failure of the incorporated verbal head to assign accusative case to its internal argument etc. This analysis is not entirely unproblematic for the approaches cooperating with distributed morphology, because its part is not a syntax-based approach to morphology (e.g., no acategorial roots occur). For the Czech language, it is not even descriptively adequate, since Czech process nominals can take just one genitive case6 (like in Russian), but, unlike in Russian, they allow just one genitive case at non-process nominals, too. (The other genitive surfaces as a Possessive, in both cases): (14) *projekt reformy premiéra *projektování reformy premiéra premiérův projekt reformy premiérovo projektování reformy The empirical support disputing the hypothesis of the passive status of nominalizations comes from the data which show that nominalizations and passive participles face the mentioned (Harley's) requirement of containedness in such manner that they cannot be considered related. Data in (15) show a discrepancy at unaccusatives, unergatives, argumentless verbs and at non- 5The present hypothesis is necessary in the syntax-based morphology and is confirmed by the following data: At stem-derivates, the genitive DP can be interpreted both as A-l and A-2, or, as Poss, respectively (učení Pavlaa-i/a-2/poss)- At action nominalization, the genitive DP is necessarily interpreted as A-2 or Poss (učitel Pavlaa-2/poss) since A-l is realized by the suffix -tel- (c.f. P. Caha - P. Karlík, 2005). Therefore, in the structure Inaučitelnost Pavla (matematice) the DP Pavla is necessarily interpreted as A-2 but in the structure with a stem-derivate containing the structure of action nominalization, there is another head licensing A-l present, and so the genitive DP is necessarily interpretable as A-l (učitelování Pavlaa-i)- 6Two of the Genitive cases can possibly occur only if one of them is a lexical case: zbavení ženy starostí / zbavit ženu starostí (J. Panevová, 2000). 110 actional transitives, data in (16) at reflexives: (15) blednutí, běžení, sněžení, dostání x *blednut, *běžen, *sněžen, *dostán (16) umývání se x *umýván se Next: If a stem-derivate really is a passive structure it should have in a DP (17) a genitive DP with the interpretation A-2 (an internal argument) in a subject position, i. e. in [spec, nP], analogically as is a DP with the A-2 interpretation in the subject position with a nominative morphology , i. e. in [spec, TP] in CP with passive participles (17). The contrast in (17a) exploiting the binding theory shows that a nominative has the properties of a subject while a genitive does not. It means that not even these data do not support the passive nature of stem-nominalizations: (17) kritizování učitele žákem x učitel je kritizován žákem a. *kritizování učitelei svými žákem x učiteli je kritizován svými žákem Also, most of the native Czech speakers do not consider grammatical the structure with the Possessive a-2 (17'), either: (17') *učitelovOi kritizování svými žákem I take this evaluation for relevant because for many Czechs the nominal Possessive with the interpretation A-l binds an anaphora (see (18) x (18a). It can mean that the pronominal Possessivea-i has the properties of a subject while a genitive does not (19). (This is not true for pronominal Possessives; see the contrast (19)): (18) a-iPetrovOi kritizování svéhoi učitele a. *a-2PetrovOi kritizování svými učitelem (19) ?PetrovOi pobíhání ve sve; pracovně x *pobíhání Petrai ve sve; pracovně (19') ???moje pobíhání ve své pracovně Another pair of examples show that a possible evaluation of a Possessive as a subject in the phrases with nominalizations contrasts with an unambiguous evaluation of a Possessive as a non-subject in the phrases with genuine nouns (cf. important example (20)): (20) ?PetrovOi pobíhání ve sve; pracovně x *Petrovo křeslo ve své pracovně The presented data give a good reason to conclude that the analysis of the phrases headed by stem-derivates as passive structures is not compatible with the distributed morphology and it is not descriptively adequate, either. Therefore, I can announce another hypothesis: The feature contributed by a head into which VI -n-/-t- is inserted is [gender]. This hypothesis can be easily tested, using the already mentioned finding that a nominal suffix attached to the root as first prolongs the prefix with a final vowel. Data in (21), especially the contrast (21a) x (21b) plus comparison of (21a) and (22) show that the suffix -n- has not only a nominal feature but also a gender feature. Since the gender feature is not interpretable it must be eliminated during the derivation. It can be eliminated by checking the under Agree by a closest accessible node containing gender feature: the deriváte then agrees (as an adjective) with an NP: [gender: mase. / fem. / neutrum]. Alternatively, this feature must be eliminated by Merge with a head containing the feature [gender: value] and, in this case, the VI -%- [gender: neutrum] is inserted and a deriváte is a noun: 111 (21) (a) zá-chyt-n-ý x (b) za-chyt-i-t zá-chyt-k-a (22) za-chyc-e-n-ý dopis / za-chyc-e-n-á zpráva za-chyc-e-n-í(-0/...) It seems that a structure [Fi - Fn- Th - yj\7 with the event and aspect interpretation generated by a syntactic process can be considered a universal structure which can be merged further: If it is merged as a complement of little n, the result is a NP/DP, if it is merged as a complement of little v, the result is vP, in the style of the distributed morphology. What follows from this analysis is the finding that at the stem-derivates the chunks of the structure merged with a D-context contain a VP, but don't contain a vP. The acceptance of this analysis, first proposed by Marantz (1997) and accepted by others, can be supported by the following data: The absence of little v brings about the fact that the complement is not assigned an accusative case (it correctly predicts the contrast: čte knihu x čtení *knihu). Further, the absence of little v causes that a [Spec, vP] position is not available, therefore, no external argument is licensed. This analysis correctly predicts the following facts: In a transitive sentence with a potential object Petr napomíná, the DP Petr is interpretable necessarily only as a A-l („agent") since it has been inserted into the structure syntactically, through little v, and then moved from [Spec, vP] to [Spec, TP]. In the DP Petrovo napomínání is the Possessive Petrovo possessor (inserted either through the head little n or through the head Poss) and exactly only at the stem-derivates and picture nouns it can be reinterpreted - on the principle of coercion, e.g. - both as an „agent" and as a „patient" (Petrovoposs > a-i/a-2 napomínání). At the nouns not contaminated with verbal features it can be interpreted only structurally, as a possessor: Petrovoposs auto). The same is true about the interpretation of a Genitive DP: napomínání Petraposs > a-i/a-28 x letadlo prezidentaposs. If in the structure of a phrase another DP or Possessive is present, their thematic interpretation is computable from UTAH: Petrovoa-i napomínání Pavlaa-2, Petrovoa-2 napomínání Pavlem a-\- Also the analysis according to Marantz faces problems, e.g., with the already mentioned occurrence of an Instrumental adjunct. The instrumental presupposes a passive morphology (23) .9 Other problems pose the reflexives (see below), the occurrence of vP adverbs (in the post-subject position) (24) while the sentential adverbs in this position are unacceptable (24a).10 In (25) we can see that those adverbs do not occur at the root-derivates either: (23) *kritizuje Petrem x je kritizován Petrem *poslední kritika románu Petrem x poslední kritizování románu Petrem (24) vyčištění Vašeho oblečení rychle a spolehlivě (Vám nabízí čistírna ...) rychlé a spolehlivé vyčištění vašeho obleku (Vám nabízí čistírna) a. *vyčištění Vašeho oblečení pravděpodobně (Vám nabízí čistírna) pravděpodobné vyčištění Vašeho oblečení (Vám nabízí čistírna) (25) čtení té detektivky rychle (je skoro hřích) x *četba té detektivky rychle (je skoro 7F represents other heads of functional projections above Theme, e.g., two heads for a lexical aspect. The Theme can combine with a genuine y/ from the Encyclopedia (uč-e-n-í) or with a complex structure derived by a syntactic process from a -J (uČ-1-TEL-ov-á-n-í). 8 As for the enforcement of the interpretation of the genitive DP as A-2 shown by the contrast of napomínání Pavla a-i/a-2x napomenutí Pavla a-2 at the stem-derivates with the feature [aspect: perfective] see below. 9 A counterexample can be found in the Infinitival phrases in ECM-constructions: Petr nechal [Marii přečíst básničku] x Petr nechal [básničku přečíst Marií]. 10 A similar Czech example was brought by L. Veselovská (2001); as for the analysis of the phenomenon, see J. Fu - T. Roeper - H. Borer (2001). 112 hřích) At the moment, we give up the solution because it is necessary to deal also with the properties of the head into which the suffix -i- is inserted. The example (26) brings an assumption that the Czech stem-derivates are mass nouns: (26) Balení těch dárků ho unavilo x *Dvě balení těch dárků ho unavila Stem-derivates as mass nouns behave analogically with pluralia tantum, namely they co-occur with generic and not with cardinal numerals (26a); the predicate agreement confirms that they are in singular (in the same way as the pluralia tantum are in plural (27)): (26) a. *Dvě balení těch dárků ho unavila x Dvojí balení těch dárků Petra unavil-o/*-a (27) Jedny housle ležel-y/*-a na pohovce It looks that all the stem-derivates are compositionally mass nouns, and therefore, in accordance with standard opinions (e.g., H. Borer (1999-2001), H. Harley (in press)) process nominals. If in general it is true that mass nouns can be turned into count nouns (e.g., genuine nouns in syntax with the help of classificators: dobytekmass > *dva dobytky > dva kusy dobytkacomít x pescomít -dva psi > *dva kusy psa), it is necessary to expect that also the stem-derivates as the instances of mass nouns allow for this change. The shift of mass nouns into count nouns is here started by a context, as shown in (28). It manifests itself both in the fact that stem-derivatescount can occur in singular and in plural (29), and in their idiomatic interpretation as result nominals denoting especially a result (hlášení, pohoštění, čalounění ...), a place (propadání, stoupání ...), means of activity (krmení, osvětlení, (lyžařské) vázání, oblečení ...) etc.: (28) Balení mass ho unavuje x Balenícount je vodotěsné. (29) *Dvě balenímass těch léků ho unavila x Dvě balenícount těch léků byla v ledničce The presented data show that -í- could be VI manifesting the feature [mass]; therefore the feature [gender] has a value "neuter", cf. stáf-í, list-í cukr-ov-í, kř-ov-í pan-stv-í, sochař-stv-í ..., pod-hůř-í, ná-vrš-í, Po-lab-í.11 The distinction process nominals (PN) x result nominals (RN) is relevant for the analysis of stem-derivates because each of the types shows a different syntactic behaviour (since J. Grimshaw, 1990) an obligatory component of all nominalization theories): (a) Only subjects and complements of PN can be interpreted as arguments: (30) PetrovoA-i oblečeníPN MarieA-2 trvalo dlouho / PetrovoA-2 oblečeníPN trvalo dlouho PetrovoPoss oblečenÍRN se sušilo na dvoře (b) PNs can compositionally express the values of the feature [aspect] (psaní - napsání, opsání - opisování)12 and the values of the feature [quant] (psaní - psávání). - RNs have only one aspect form (idiosyncratically, either perfective, or imperfec-tive), but not an aspect meaning deductible from it: 11 It is necessary, though, to explain a single exception, namely paní [Fem, Count], and numerous neuter nouns with the feature [count] (pondělí, ústí ...). 121 suppose that a good diagnostics can be seen in the preposition během selecting a complement with the feature [aspect: imperfective]: během balení x *během zabalení / během zahalování. 113 (31) BalenípN / zabaleníPN toho léku trvalo chvilku a. BaleníRN / *zabaleníRN toho léku bylo poškozeno (32) Jeho propadánípN / propadnutípN už v základní škole nás mělo varovat a. Navštívili jsme Rudické propadání^ / *propadnutÍRN (c) PNs can be negated, RNs can not: (33) BalenípN / nebalení pn svačiny do sáčku se mu vymstí a. BalenÍRN / *nebalenÍRN léků bylo poškozeno vodou (d) PNs can be modified by vP adjectives (and adverbs) marking the presence of an external argument, RNs can not: (34) Rychlé / úmyslné baleníPN těch léků ho překvapilo a. *Rychlé / *úmyslné baleníRN těch léků je vzduchotěsné (e) PNs cannot be modified by the adjectives denoting the properties of actual objects, RNs can: (35) *Modré / *papírové baleníPN těch léků ho překvapilo a. Modré / papírové baleníPN těch léků je vzduchotěsné (f) PNs are not accessible for diminution, RNs are: (36) Při psanípN / *psaníčkupN toho dopisu zpíval a. PsanípN / psaníčkoRjsr od milenky ho potěšilo (g) PNs allow reflexiveness, RNs do not: (37) PetrovOA-i holeníPN se trvalo dlouho a. Petrovoposs *holenÍRN se zůstalo nepoužito It is well known that the syntactic behavior of the reflexive se allows the interpretation that is ceases to be a clitic and it agglutinates as a postfix into the internal structure of a stem-derivate: (38) shows the contrast with the predicted clitic-behavior, (38a) shows the contrast with the behavior of a reflexive phrase with an overt case morphology, (38b) shows that se blocks the interpretation of a Possessive as an A-2: (38) rychle se oholí / *rychle oholí se x *rychlé se oholení / rychlé oholení se a. *oholení sebe Petra x oholení se Petra b. PetrovoA-i seznámení se s Marií x Petrovoa-i/a-2 seznámení s Marií 114 2.5 If we observe at least some of the data pertaining to the root-derivate syntax, we can see that some of them can be read both as a PN and as a RN (39), others only as a RN (17) and, some only as a PN (18) which is their idiosyncratic property : (39) VýchodRN z kina je ucpaný Při východupN z domu zahladil po sobě všechny stopy (40) Sprcha^n je ucpaná Při *spršepN / sprchováníPN zad ucítil chlad (41) Ozývala se dlouhotrvající střelbapN *StřelbaRN / střela rn vylétla z hlavně velkou rychlostí Interestingly, root-derivatespN behave differently from stem-derivatespN: root-derivatespN cannot be reflexive (42), they cannot take negation (43), their aspect is not deductible from their internal structure (44), manifestation of their internal argument is not obligatory (45), they cannot co-occur with VP adverbs (46), they require a SC with jako support (471): (42) hraní se / si na prezidenta x *hra se / si na prezidenta (43) nečtení novin x *nečetba novin (44) několik let se táhnoucí stavění / *dostavění stadionu x několik let se táhnoucí stavba / dostavba stadionu (45) po *objevení došlo k poklesu úmrtnosti x po objevu došlo k poklesu úmrtnosti (46) ?čtení té detektivy rychle je skoro hřích x *četba té detektivky rychle je skoro hřích (47) rozhodl se pro odejití (jako) první x rozhodl se pro odchod jako první An important property shared by root-derivatespN and stem-derivates, but different from genuine nouns (see (48) x (48a)) is the ability of their subjects to bind anaphoras: (48) ?Petrovo procházení se ve své zahradě / ?Petrova procházka ve své zahradě a. *Petrova lavička ve své zahradě 3 Conclusions Neither the analysis of stem-derivates supposing a portion of a verbal structure containing little v, nor the analysis without little v can predict all the contrasts presented in this paper. One of the analytic options is the analysis according to H. Harley (2005) which cannot be properly tested yet. The presented data and their analyses can be relevant for a diachronic description already in this non-definite version. Hopefully, they can be a starting point for an analysis of the relation between neuter stem-derivates with a feature [mass] realized by inserting of the VI -%- , at which only the consonant endings are visible (M. Ziková, in press) (Sg: Nom, Gen, Dat, Acc, Loc: kopáníS, Instr: kopání-m; PI: Nom, Gen, Acc: kopání-^, Dat: kopání-m, Loc: kopání-ch, Instr: kopání-mi) and feminine stem-derivates with full visible adjective inflection (Nom:kopan-á, Gen, Dat: kopan-é, Acc: kopan-ou ...). 115 References Abney, Steven Paul. 1987. The English Noun Phrase in Its Sentential Aspect. Ph.D dissertation. MIT, Cambridge, MA. Ackema, Peter and Ad Neeleman. 2004. Beyond Morphology. Oxford: OUP. Alexiadou, Artemis. 2005. Gerund Types, the Present Participle and Patterns of Derivation. Maienborn, C. &; A. Wöllstein (eds.). Event Arguments: Foundations and Applications : 139-152. Tübingen: Niemeyer. Baker, Mark. 2005. On Gerunds and the Theory of Categories. Ms, Rutgers University. Borer, Hagit. 1999-2001. The Form, the Forming, and the Formation of Nominals. Handout. —. 2003. Exo-Skeletal vs. Endo-skeletal Explanations. Syntactic Projections and the Lexicon. Moore, J. &; M. Polinsky (eds.). The Nature of Explanation in Linguistic Theory: 31-55. CSLI and University of Chicago Press. Bresnan, Joan. 1997. Mixed Categories as Head Sharing Constructions. King, T. H. & M. Butt (eds.). Proceedings of the LFG97 Conference. Caha, Pavel and Petr Karlík. 2005. Where Does Modality Come from? Hansen, B. & P. Karlík (eds.). Modality in Slavonic Languages:61-72. München: Otto Sagner. Chomsky, Noam. 1970. Remarks on Nominalization. Jacobs, R. & P. Rosenbaum (eds.). Reed-ings in English Transformational Grammar: 184-221. Waltham, MA: Ginn &; Co. Engelhardt, Miriam and Helen Trugman. 1998. D as a Source of Adnominal Genitive in Russian. Boskovic, Z., S. Franks & W. Snyder (eds.). FASL 6. The Connecticut Meeting 1997:114-133. Ann Arbor: Michigan Slavic Publications. Fu, Jingqi, Thomas Roeper and Hagit Borer. 2001. The VP within Process Nominals: Evidence from Adverbs and the VP Anaphor do-so. Natural Language and Linguistic Theory 19:549-582. Grimshaw, Jane. 1990. Argument Structure. Cambridge, MA: MIT Press. Harley, Heidi. 2005. Syntactic Event Structure and Nominalizations. Handout. —. (in press) The Morphology of Nominalizations and the Syntax of vP. Giannakidou, A. &; M. Rathert (eds.). Quantification, Definiteness, and Nominalization. Oxford: Oxford University Press). Harley, Heidi and Rolf Noyer. 1998. Mixed Nominalizations, Short Verb Movement and Object Shift in English. Tamanji, P. N. & K. Kusumoto (eds.). Proceedings of NELS 28:143-157. Amherst: GLSA. —. 2000. Formal vs. Encyclopedic Properties of Vocabulary: Evidence from nominalisations. Peeters, B. (ed.). The Lexicon-Encyclopedia Interface: 349-374. Amsterdam: Elsevier. Jelínek, Milan. 1967: Jména dějová. Danes, F., M. Dokulil et al. (eds.). Tvoření slov v češtině 2: 562-653. Praha: Academia. Karlík, Petr. (in press) Několik drobných poznámek k dějovým jménům. Sborník prací Filozofické fakulty brněnské univerzity A 55. Lees, Robert B. 1960. The Grammar of English Nominalizations. The Hague: Mouton & Co. Marantz, Alec. 1997. No Escape from Syntax: Don't Try Morphological Analysis in the Privacy of Your Own Lexicon. U. Penn Working Papers in Linguistic 4.2: 201-225. 116 Panevová, Jarmila. 2000. Poznámky k valenci podstatných jmen. Hladká, Z. & P. Karlík (eds.). Čeština - univerzália a specifika 2:173-180. Brno: Masarykova univerzita. Scheer, Tobias. 2001. The Rhythmic Law in Czech: Vowel-final Prefixes. Zybatow, G., U. Junghans at al. (eds.): Current Issues in Formal Slavic Linguistics: 37-48. Frankfurt am Main: Lang. Veselovská, Ludmila. 2001. K analýze českých deverbálních substantiv. Hladká, Z. & P. Karlík (eds.). Čeština - univerzália a specifika 3:11-28. Brno: Masarykova univerzita. Ziková, Markéta. In press. Phonological cyclicity: case markers in the Czech nominal declension. Proceedings of FDSL 6.5 (Nova Gorica). Zucchi, Alessandro. 1993. The Language of Propositions and Events. Dordrecht: Kluwer. 117 1, 2, SE Lucie Medová and Tarald Taraldsen lmedova@Princeton.edu, knut.taraldsen@hum.uit.no 1 Introduction The goal of this article is to contribute to the understanding of the morphosyntactic properties of the reflexive clitic (henceforth: se) in Czech (and Slavic and Romance languages in general). In particular, we want to suggest a way of unifying its various uses, although we will focus mainly on the purely reflexive (1) and the impersonal se (2), and we want to see how se relates to the lsi and 2nd person clitics, with which morphology suggests that it forms a natural class1. (1) reflexive se Karel se učesal. Karel7voM.5G.Ai SEacc combedM.5G 'Karel combed [his hair].' (2) impersonal se Hrály se karty a zpívalo se. playedF.PL SEACC cardsAr0M.PL.F and sung^c SEACC 'There was some cards playing and singing.' / 'People were playing cards and sang.' We will begin by outlining a unified analysis of reflexive and impersonal se. Then, we will discuss certain problems which this analysis seems to run up against. We entertain two possible ways to analyze se, first as an analogue to antipassive morpheme and second as a spell out of an abstract self, a part of a special possessive-like construction. As the second type of analysis handles better the empirical evidence, we elaborate on its consequences. In the last section, we return to the connection we started with at the beginning, trying to link the morphology of the reflexive clitics with person and case. 2 The raising analysis of reflexives We first present a somewhat simplified version of an analysis partially assimilating the derivation of sentences with reflexive se to the one standardly assumed for (one type of) impersonal 1 Abbreviations are the usual ones, case: NOM(inative), (GEN)itive, (DAT)ive, (ACC)usative, (LOC)ative, (INS)trumental, (OBL)ique, (ERG)ative, (ABS)olutive, gender M(asculine), MA=masculine animate, MI=masculine inanimate, F(eminine), N(euter), number: SG=singular, PL=plural; DO=direct object, IO=indirect object, IA=internal argument (subsuming both DO and 10 objects), EA=external argument. The Czech examples (if not cited) are from Correct Czech (spisovná čeština), as required by the second author. 119 se. As we imply that our discussion should extent both to Slavic and Romance reflexive clitics, we show examples both from Czech and Italian, the representatives of each group. 2.1 First step: Reviving Kayne (1986) Routinely, a sentence with impersonal se like (3)2 is given a parse like (4)3, where se somehow licenses the occurrence of PROar6 in Spec-vP. The arbitrary PRO gives rise to the (obligatory) human agent reading. (3) impersonal se a. Noviny se čtou ráno (komiksy večer). newspaper^voM.PL se read3.p£ morning comic strips evening 'People read newspaper in the morning, comic strips in the evening.' b. I giornali si leggono la mattina. the j0urnals7v0M.PL se read3.PL the morning 'People read newspaper in the morning.' (4) IP A reflexive se, on the other hand, is usually treated as an anaphoric object clitic (Burzio (1986), Dobrovie-Sorin (2005); for Czech Panevová (1999)). So, (5) on the reflexive reading would be analyzed as in (6), begging the question how the impersonal and the reflexive se are related to each other4: (5) reflexive se a. Chlapci se myjí každé ráno. boysnom.pl se wash3.p£ every morning b. I ragazzi si lavano ogni mattina. the b0ys7v0M.PL se wash3 PL every morning 'The boys bath themselves every morning.' 2The Czech version of (3), with the present tense and imperfective verb, implies somewhat deontic modality, cf. One should read newspaper in the morning, comic strips in the evening or habituality; in the text, we ignore this implication. 3To reduce clutter, we do not show V-raising to I (used agnostically to denote whatever functional head hosts a preverbal subject in its Spec). The (final) location of SE is given as I on the basis of Kayne (1986), Cinque (1988) and much other work. The choice of PROarb as the external argument (EA) accords with Kayne (1986). (A different choice will be made below.) 4However, there is a family of analyses represented by Grimshaw (1982) and Reinhart and Reuland (1993), for instance, which would see SE as the morphological reflex of a lexical operation converting a transitive argument structure to reflexive one. 120 (6) IP myjí t j každé ráno Kayne (1986), however, points out the reflexive reading of (5) might actually come from a structure exactly like (4) except that PRO, the EA, doesn't have arbitrary reference, but is interpreted as controlled by the DP raised to Spec-IP: (7) IP On this account, impersonal and reflexive SE both license a PRO in the EA-position. This analysis has the virtue of unifying the two SEs. But encounters a number of problems, two of which we will concentrate on below. 1. it seems to make it harder to understand why reflexive SE seems to group with the 1st and 2nd person clitics, if the latter are still treated as ordinary object clitics 2. by analyzing the subject of a reflexive sentence like (5) on a par with the subjects of unaccusatives, it predict - contrary to the facts - that reflexive verbs behave like unac-cusatives with respect to auxiliary selection across Italian dialects5. Reflexives, however, select auxiliaries by and large like unergatives. Moreover, unlike unaccusatives, reflexives don't allow ne-extraction from a postverbal subject Alboiu, Barrie, and Frigeni (2004) propose a way of updating Kayne's (1986) analysis which would eliminate the problems in 2. Adapting Hornstein's (1999) proposal that control is raising to a ^-position, they claim that the reflexive reading of (5) is the product of a derivation that raises the IA to Spec-vP before it is moved to Spec-IP, yielding (8) instead of (7): 5For Slavic, it is much harder to establish the unaccusative - unergative distinction, as the tests routinely used are unreliable. In our approach, true reflexives just must be unergative, as they 'pick up' the external ö-role. 121 (8) IP On this view, the subject of (5) on its reflexive reading is an EA (the bearer of the agentive 6-role) as well as an IA, and reflexives are thus expected to group with transitives and unergatives rather than with unaccusatives. On the face of it, however, such an approach runs straight against the ^-criterion: the single element chlapci in (8) accumulates two distinct ^-roles, one in the IA position and another one in the EA position. In the next subsection, we present an elaboration that cancels this violation. 2.2 Second step: #-roles and Case Medová (to appear) develops an analysis which shares the virtues of Alboiu, Barrie, and Frigeni (2004) account, and in addition accounts for the fact that a sentence like (9), with DAT SE, has a reflexive reading, but not an impersonal one; unlike (5) (repeated here as (10)), which is generally ambiguous between a reflexive reading and an impersonal one: (9) reflexive DAT Si a. Chlapci si myjí ruce každé ráno. boys^voM.pl S^dat wash3.PL handsacc.pl every morning b. I ragazzi si lavano le mani ogni mattina. the boysnom.pl ^dat wash3.pl the hands every morning 1. REFL: 'The boys wash their hands every morning.' 2. *LMP: *'One washes the boys' hands every morning.' (10) reflexive ACC SE a. Chlapci se myjí každé ráno. boysnom.pl SE wash3.pl every morning b. I ragazzi si lavano ogni mattina. the b0ys7v0M.pl SE wash3pL every morning 1. REFL: 'The boys bath themselves every morning.' 2. IMP: 'One bathes the boys every morning.' Her analysis incorporates Starke's proposal (Starke (2005); applied to Czech data by Caha (2006)) that a DP is in general embedded inside one or more Case-layers ordered in accordance with a universal hierarchy, for us, DAT > GEN > NOM. For example, the NOM subject of a run-of-the mill sentence like (11-a) might actually appear in Spec-vP as (11-b), with the NOM, the NomP-layer, embedded inside some oblique Case K: (11) a. John ate an apple. 122 b- KP K NomP Nom John The NOM surfaces because the NomP is allowed to raise to Spec-IP from inside the larger KP, stranding K in Spec-vP6: It is further assumed that the stranded K will be part of the substructure lexicalized by the verb root, under the conception of lexical insertion developed by Starke. That is, lexical insertion can target non-trivial subtrees. Like Starke, Medová (to appear) also assumes that only the sister of the highest head can subextract. This assumption is a crucial element in the explanation for (9). Another crucial ingredient is the assumption that an EA, the element in Spec-vP, must be embedded inside a layer of oblique Case. For concreteness, we take the relevant K head to be GEN(itive). Similarly, Kayne (1993) and Mahajan (1994) have it that the EA argument is born as an OBL case, pointing toward the evidence from ergative languages. We return to the ergative connection in the section 3.1. With this set of assumptions, each of which is independently motivated, Medová (to appear) predicts the lack of an impersonal reading with DAT se. To reach the NOM position in Spec-IP, the DAT IA corresponding to the subject of (9) must be able to shed two layers of oblique Case, as depicted in (13). 6Whether or not there should be the trace (copy) of the raised NomP, depends on theoretical commitments irrelevant to our present concerns. 123 NomP IP chlapci That is, the DP must raise in two steps: first to a GEN position, stranding the DatP-layer, and then, to the NOM-position, stranding the GenP-layer. But the only GEN-position between the DAT IA's initial position and the NOM-position is the EA-position, Spec-vP, and by moving to this position, the raised IA will pick up the agentive 8-vo\e in addition to the one it received qua DAT IA, and a reflexive reading is produced7. Thus, to produce a reflexive reading, the DP has to have exactly two layers to shed: one in the place it has been born (DatP-layer in (13)) and another one in the EA-position, which, as we argued below, we take to be GenP8. Finally, concerning the o'-cirterion, notice that - strictly speaking - it is not exactly the same element that picks up the different 6>-roles, as shown in (13). So, the 8-vo\e in the IA position is associated with the DatP-layer while the EA 8-vo\e with the GenP-layer. (We briefly discuss 7Another fact predicted by this is that a sentence like (i) has an impersonal reading, but not a reflexive one (for some DAT, see Medová (to appear) for discussion): (i) Děti se vracej rodičům. kidsjvom.pl SEacc return3.PL parents/?at.pl 1. *REFL: 'The kids return (themselves) to their parents.' 2. IMP: 'The kids are being returned to their parents.' The DAT in (i) c-commands the initial position of the IA raised to Spec-IP, so that Relativized Minimality would prevent a GenP to raise to Spec-vP from inside the IA, given that the GenP inside the non-reflexive DAT argument is closer to Spec-vP. 8If we take it seriously, we wind up with the prediction that only DAT (and animate, as discussed further) DPs should be able to reflexivize. It is a welcome result for a variety of Romance languages that have the animated DO introduced by a preposition a which is taken to be a DAT case marker, as in (i); inanimate nouns are not introduced by a. In Slavic, the ACC-to-GEN shift for MA nouns could be seen as an instantiation of an OBL case. See Medová (to appear) and section 4.2 for discussion. (i) Vi a Antonio / (*a) un libro. seei.sc.past a Antonio / a book 'I saw Antonio / a book.' 124 an alternative to this proposal in section 3.3.) While perhaps unusual, the assumptions (each of them motivated, however) we made allow us to discuss the role of se in the derivations of reflexives and impersonals, as, so far, the se itself seemed rather superfluous. Also, we motivate our assumption that the Case layer associated with the EA is, indeed, GEN. 3 Internal Arguments have GEN inside Under the assumptions so far, it is essential that the IA has a GEN inside: to derive a reflexive means to move the GenP (a GEN-Case layer) to the EA position as the first step of the reflexive derivation. Further, the NomP (a NOM-Case layer) moves from under the GenP to the NOM position, as depicted in (13). But why should the layer associated with the IA position be GEN? And what exactly is se needed for in the structure (13)? There are two potential ways to substantiate the claim that the IA that derive reflexives contains in a way GenP (an OBL-Case layer, in any event) in it and that this GenP is in a particular way connected to an EA. The first option draws a parallel between the ergative languages and NOM/ACC languages via assimilating the antipassive morpheme of ergative languages to se-morpheme of Slavic or Romance. The second option builds on the observation that GEN appears in possessive constructions and on a particular view of reflexivization going back to Helke (1971). We will discuss both options in turn, starting with the antipassive connection first. 3.1 Antipassive Oblique case can be seen on a DO, if - for a reason - the DP does not appear in the DO position (which, essentially, translates as 'not being marked by a structural (ACC) case'). One particular instantiation of the OBL-marked DO is seen in antipassive constructions, characteristic for ergative languages9. So, in Chukchi (Chukotko-Kamchatkan), a transitive sentence has the EA in ERG case and the DO appears in ABS, as shown in (14-a). In antipassive, indicated by a special morpheme -tko- in Chukchi, the EA argument is in ABS case and the DO bears now DAT case, shown in (14-b). Notice, furthermore, that the ERG case marking of the EA is by an OBL case: a connection to our claim that EA are GEN. (14) a. 9tlg-e keyrj-gn penrg-nen fatherEpG beavABS attack-3SG:3SG.AOR 'The father attacked the bear.' b. 9tlg-9n penr9-tko-y?e kayrj-etg fatherj4B5 attack-AP-3SG.AOR beavDAT 'The father rushed at the bear.' (Kozinsky, Nedjalkov, and Polinskaja (1988)(2c,lc)) The DO thus starts out as an OBL case (DAT in (14-b)) and 'peels' to the structural DO-position - in parallel to the (oblique) DP in (13) 'peels' its way to the NOM(=structural case) position. Arguably, then, the DO has to remain OBL because the (structural) DO position is filled up with a different morpheme: -tko- in (14-b). This opens for an attractive analysis of 9But notice recent literature on antipassive constructions in NOM/ACC languages as well, for instance Postal (1977) on French, Ndayiragije (2006) on Kirundi (Bantu) and Say (2005) on Russian sja. 125 SE in (13): suppose, that in parallel to Chukchi's -tko-, the Czech SE 'blocks' the (structural) DO-position. The DP Karel in (13) starts out as DAT (an OBL case) and should move now to the DO-position, but it cannot as the position is filled by SE. Instead, it has to move to the EA position and further up to the NOM. The idea of SE being a DO-position 'filler' is relatively well-supported. First, the SE itself is ACC-marked, if compared with the DAT-marked Si10. Second, SE indeed looks like an object clitic, as shown in (15) for Czech and Italian. Kayne (2000) argues that to complete the line m- (pronoun with the root to) for 1st person, t- for 2nd person, the s- type pronoun, not the /-type (as the definite articles) should be taken for 3rd person. Plausibly, the same could be argued for Czech. Czech ACC DAT Italian lsi person mě mi 1st person mi 2nd person tě ti 2nd person ti reflexive se si reflexive si 3rd person M&N ho mu 3rd person M&N lo F ji jí F la Third, when SE appears in the sentence, no other ACC-marked DP is possible; the other argument appears in (PP-introduced) OBL case (16). (16) a. Karel se směje Ivoně. K překročení Alp cross/jvj? AlpsAcc.pl crossing Alpsgen.pl 'to cross the Alps' 'crossing of the Alps' 127 More precisely, we assume that the possessor DPs are in general introduced to the derivation as DatP. The motivation for this claim comes from a variety of languages. For instance, in French (22), the possessor is indeed introduced by the preposition d, a DAT case marker. (22) Notre ami a tous. our friend to all 'Our common friend.' On the other hand, some possessors appear as GEN (notre in (22)). In our approach, as before, GenP emerges from under the DatP by a movement to a higher position in the complex DP, as shown in (23). (23) DAT possessor DP "~^ GEN possessor DP Possessor This higher position, we take it, is absent from the complex DPs headed by SELF, in other words, the PossessorSELF remains a DatP until the derivation reaches an EA position to which the GenP can raise from inside the possessive DatP, as shown schematically in (24)12. This way, the derivations we have, are consistent with the assumption that the highest head is always left behind under movement. (24) DP - IP Possessor We will take it that self denotes something, which, by convention, can only be a part of an animate / human being. So, only animate DPs will combine with self and therefore only 12In the section 4.2 we suggest an alternative. 128 animated DPs can reflexivize, as discussed in section 4.1. Finally, the full derivation of reflexives (5), repeated here as (25), under these assumption would run as shown in (26). (25) Chlapci se myjí každé ráno. boysnom.pl se wash3 PL every morning REFL: 'The boys bath themselves every morning.' (26) IP spells out IP chlapci We leave it an open issue whether animate nouns are always introduced in the derivations as possessors of self, or, in fact, occur as possessors of self only when needed for the derivation of reflexives to converge. In the former case, we would have to say that the self part spells out independently (as se) just in case the derivation splits the possessor from self (an option taken in section 3.3). Potentially, the facts that we are going to discuss in section 4.2 argue for this view. At this point, however, we will compare the antipassive se approach with the Possessor of self analysis. We want to see which analysis can explain more. 3.3 Open competition: Antipassive against Possessor of self analysis 3.3.1 DAT reflexive The antipassive analysis of se links the morpheme se to the DO position to 'block' it, so that no other DP can move into it. But as far as we see, there is no possibility to extend this view to DAT reflexive clitics Si. Moreover, from a comparative perspective, to the best of our knowledge, antipassives on indirect objects don't exist. On the other hand, the Possessor of self analysis does provide a neat extension to DAT reflexives. The derivation of a DAT reflexive sentence (9), repeated here as (27), would proceed as (28) shows. (27) Chlapci si myjí ruce každé ráno. boystvom.pl SVdat wash3.PL hands^ccPL every morning refl: 'The boys wash their hands every morning.' 129 (28) IP spells out IP chlapci Still, one might ask why exactly self is in the derivation. We have, as suggested above in section 3.2, two potential answers: first, self is in the derivation of reflexives simply because it is an inherent part of the structure of animate nouns. We take up this line in the section 4.2. On the other hand, self can be required for the ^-criterion, adapting essentially Kayne's (2005) approach, as an alternative to the approach we took in section 2.213. So, if the complex DP splits up in the course of the derivation (as, indeed, happens with reflexives), then both parts must be spelled out. 3.3.2 Impersonals Under the Possessor of self story, impersonals are rather neatly derived. The example of impersonal se is repeated below in (29). (29) Noviny se čtou ráno (komiksy večer). newspapernom.pl se read3.PL morning comic strips evening 'People read newspaper in the morning, comic strips in the evening.' On our approach we just cannot say that the obligatory human agent reading of impersonal se comes from a PROar6, as is the standard assumption. Instead, we suggest, in line with the Possessor of self analysis, that the EA of an impersonal se construction is, indeed, a complex DP, a Possessor of self of a particular kind: the possessor, in our view, is a silent human agent, labeled as man in (30). The complex DP in the EA position needs to be pronounced and the only way to do so is se. 13We assumed that the ^-criterion is satisfied by the 0-roles being picked up by different Case-layers. 130 (3°) IP spells out IP MAN If we assume that the complex animate DP is born in the EA position, we need to make sure that the NomP layer of the man does not get to the NOM position: it would ban the DP from the DO noviny from getting there. But notice that the NomP layer of the EA argument man is buried down under the GenP layer, so, under Relativized Minimality, the NomP layer of the DP from the DO position is closer to the NOM position in IP. Additionally, the movement of the whole complex DP from the EA position to the NOM position in IP should be disallowed as well. We would like to link the impossibility of such a movement to the fact that there are no NOM reflexives. The most striking example (or, rather, lack of such) come from Icelandic. A class of Icelandic verbs has oblique subjects (that bind anaphors and show other NOM-subject properties) and NOM objects: the expectation, then, is to see the oblique subject binding a NOM reflexive anaphor, the NOM counterpart to ACC sig. But such a form just does not exist, see Taraldsen (1994) Taraldsen (1995) for discussion. Finally, we predict that the SE seen in impersonals must be GEN: we assume that the EA position is associated with GEN. However, there is a piece of evidence that SE, indeed, can be GEN: given that every instance of ACC case has to shift to GEN under nominalizations and given the fact that SE appears in nominalizations as it is (as discussed in Oliva (2001) and Hron (2005)), we might take it that SE is indeed both GEN and ACC form. (Morphologically, GEN and ACC pronouns have the same form in Czech.) Notice that if SE in impersonals is the GEN form, as predicted by our analysis, we might be a step closer to understanding the impersonals with ACC marked original DO, as shown in the Slovene example (18). We leave this issue for further research. The antipassive analysis of SE does not extend to derive impersonals. As a matter of fact, it is not uncommon for a language to use the same morpheme to create antipassives and reflexives (Chukchi (Chukotko-Kamchatkan), Yidiji, Lardil (Australian), Greenlandic, Halkomelem (Salish), Medova (in progress) for discussion). However, non of the languages uses the reflex-ive/antipassive morpheme for an (equivalent to) impersonal. So, the fact that Slavic (and Romance) does might be because of the homonymy between the EA form (GEN) and the DO form (ACC) pointed to above. To summarize, it seems that on the face of it the Possessor of SELF analysis can both derive the DAT reflexive Si constructions and the impersonals; the antipassive analysis cannot do neither.14 In the next section, we thus discuss a potential extension for the winning analysis. 14 Yet, the outcome is different for the anticausative use of se exemplified in (i), that we did not discuss at all. In particular, our Possessor of self analysis does not lend itself in ant obvious way to derive such a 131 4 Extension of Possessor of SELF analysis In the following section we will briefly discuss a set of other facts that show that animate nouns are different from inanimates in ways that might suggest that animate nouns always come with a SELF build in. 4.1 How much are animates different? First, observe that quite generally, only animate nouns derive true reflexives. The examples in (31) are absolutely parallel with only one contrast: while (31-a) has an (masculine) animate noun, example (31-b) has a (masculine) inanimate; yet, the readings are diametrically opposite: true reflexive in (31-a) and impersonal in (31-b).15 (31) a. Karel se umyl. Kare\nom.ma.sg SE washM.sG REFL: 'Karel washed (himself).' b. Talíř se umyl. plate/voM.M/.sG SE washM.5G IMP: 'The plate was washed.' Another piece of evidence that animates are different from inanimates comes from morphological case marking of masculine nouns in Czech. (This evidence is, however, only partial as it concerns only masculine nouns.) Still, MA nouns show an extra piece of morphology in DAT/LOC.SG and NOM.PL nouns, as summarized in the table (32). (32) case MI plate MA sir DAT/LOC.SG talíř-i pán-OV-i NOM.PL talíř-e pán-OV-é Third piece of evidence is based on Czech possessive adjectives. Generally, possessors appear in GEN case (33). For animate nouns, there is another option: the possessor can be expressed as a possessive adjective (34). Crucially, however, the -ov (for MA) or -in (for F) suffixation structure: if we assume that only animate nouns are Possessor of self (as discussed further in 4.1), there is no reason for anticausatives (with an inanimate noun in NOM) to be derived with se. On the other hand, the antipassive analysis of se could be extended to the impersonals along the following lines. se in (ACC) reflexives 'blocks' the structural DO position. Now, if the structural position for EA is, as we argue, associated with an oblique case (GEN) (that is, it crucially is not the NOM position), then se can 'block' this position. As a consequence, the DP from the DO position (or, more precisely, from the OBL case-layer) can move to the NOM position. That is, we predict that se can 'block' both the (structural) EA position (associated with GEN case) and the (structural) DO position because se is both ACC and GEN marked: the same outcome as above. (i) (Karlovi) se rozbily brýle. Karelyat.sg se brokepL glassesjvoM.pl 'The glasses broke (on Karel).' 15We acknowledge that there is a (marginal) impersonal reading of (31-a), but, importantly, there is no reflexive reading of (31-b). If we say that the distinction is purely pragmatic, we lose a possible explanation of the observed syntactic restrictions. However, even animate nouns can participate (more or less marginally) in impersonal constructions (as in (31-a) above) suggesting that animates can be coerced to inanimates. Conversely, in relevant (fairy-tale) contexts, inanimates can be conceived as animates. 132 that creates possessive adjectives is unavailable for inanimate nouns, as shown in (34-a) and (34-b)16. (33) a. tajemník fakulty secretaryMA5G facultyGE7V.sG 'the secretary of the faculty' b. noha stolu legF.sG tableGE/v.5G 'the leg of the table' c. prsten Jarmily Noskové ringM/.5G Jarmila NoskováGE7V.sG 'Jarmila Nosková's ring' d. motorka Karla Noska motorcycleF.5G Karel NosekGE7V.5G 'Karel Nosek's motorcycle' (34) a. *fakult-in-0 tajemník faculty-lN-MA5G secretaryMASG b. *stol-ov-a noha table-OV-F.5G legF.5G (Veselovská (1998)) c. Jarmil-in-0 prsten Jarmila-IN-M/.5G ringM/.sG 'Jarmila's ring' d. Karl-ov-a motorka Karel-OV-/voM.F.5G motorcycleF.5G 'Karel's motorcycle' There are at least two questions: first, why only animates have possessive adjectives and second, why only MA nouns have the OV in DAT/LOC.SG case endings. We will not make any specific proposals as to why it is that only animate nouns form possessive adjectives, but we will make a suggestion about the specific complex DAT/LOC forms of MA.SG. nouns, those shown in (32). In fact, one might analyze a form like Michal-ov-i as the DAT form of a complex DP, where Michal-ov is in fact a possessive adjective possessing SELF and the DAT/LOC -i marks the case of the complex DP headed by SELF. If this is correct, then, at least, animate SG. nouns must always be introduced as the possessors of SELF, since Michal-ov-i is the only 16We abstract away from the restrictions on possessive adjectives formation: only MA or F singular nouns can form possessive adjectives and only one item can appear in the possessive adjective, ruling out (i). Apparently, even inanimate could (marginally) have ov in the history of Czech, as shown in (ii). (i) Jarmil-in-0 (*Nosek-in-0) prsten Jarmila-IN-m/.sg Nosek-iN-m/.sg ringm/.sg Intended:'Jarmila Noskova's ring' (ii) k opraveni ohne oltaf-ov-a to repairzMT flamec^jv alta,rPoss-gen '(about) the repair of the altar's flame' (Gebauer (1929):159) 133 possible DAT/LOC form of Michal17. 4.2 The animacy requirement So far, we work with the assumption that animate nouns can combine with self in a kind of possessor structure. The analysis developed up to this point commits us to the view that the DatP Possessor of self does not shrink to a GenP internally to the DP in the derivation of a sentence with reflexive se, since otherwise raising the GenP containing the non-pronominal part of the possessor to the EA-position would violate the requirement that the topmost Case-layer must be left behind under movement. But nothing precludes the possibility that a DatP Possessor of self may have its GenP raised to a DP-internal Gen-position in derivations in which the possessor does not raise to the EA-position, as discussed in (23). This in turn opens up for a solution to a problem that has nothing to do with reflexives. In Czech and other Slavic languages, singular masculine animate nouns (MA) have GEN case as direct objects, as opposed to the masculine inanimate nouns (MI), in the literature referred to as the ACC-to-GEN shift, as shown in (35). (35) Vidím Michal-a / hrad. Seei.5G.PRE5 Michal-GEAr.5G.MA / CastUGC.5G.MJ T see Michal / a castl.' Usually, this is regarded as a morphological quirk, an instance of syncretism between the GEN and the ACC. But significantly, GEN MA.SG direct objects group with oblique DPs (DPs bearing INS, LOC, DAT or 'real' GEN case) with respect to the obligatoriness of resumptive pronouns in what-relatives, as observed in Toman (1998). Whereas relativized NOM or ACC DPs do not require a resumptive pronoun, oblique DPs do, as shown in (36), and the 'Genitivus pro Accusativo' DPs do as well: compare (37-a) to non-obligatory resumptive pronoun ho with 'real' ACC of the MI nouns in (37-b). (36) a. To je ten muž / ten nápoj, co jsem *(mu) this is the majinom.ma / the drinkAroM.M/ what be^scPRES himDAT,SG.m včera propadla, yesterday fall.forf.sg 'This is the man / the drink that I felt for yesterday.' b. To je ten muž / ten nůž, co sis *(n0) tak this is the m