Clitics, Verb (Non-)Movement, and Optimality in Bulgarian

This paper addresses the issue of whether functional categories head separate projections. In particular, Pollock (1989) claims that auxiliaries head functional projections. I argue here that Bulgarian clitic auxiliaries do not head separate syntactic projections. I further argue that the respective order of Bulgarian clitics and verbs is not the result of syntactic movement (e.g. Long Head Movement; Rivero, 1994). Nor does it result from a postsyntactic re-ordering at PF (e.g. Prosodic Inversion; Halpern, 1995). I develop an alternative analysis that makes syntactic movement or PF re-ordering completely unnecessary. The new analysis is couched in Optimality Theoretic terms and builds on the non-syntactic view of clitics advocated by Klavans (1985) and Anderson (1992). The analysis incorporates linear order constraints proposed in Prince and Smolensky (1993) and Anderson (1995); it demonstrates how the simple OT mechanism of ranking violable constraints can yield the complex distribution of the interrogative particle li with a minimum number of (independently needed) assumptions. The Optimality Theoretic account is parsimonious in that it posits minimal trees, minimal movement, and global evaluations of syntactic/PF structures. The constraints that do most of the work are interface constraints (many of which are independently needed in more traditional syntactic accounts). To the extent that this parsimonious analysis is successful at handling the facts previously accounted for in terms of extended trees, it provides one substantial argument against the Pollockian view that functional categories always head syntactic projections. CLITICS, VERB (NON)-MOVEMENT, AND OPTIMALITY IN BULGARIAN* Géraldine Legendre Johns Hopkins University November 1996 I. ISSUES AND CLAIMS This paper addresses the issue of whether functional categories head separate syntactic projections or not. Much work in generative syntax assumes that they do, following Pollock (1989). Among other things, this assumption has led to a non-minimal view of tree structure, e.g. gigantic trees, proliferation of projection labels, and abundant syntactic movement. I argue here that in at least some important cases, functional categories do not head syntactic projections, based on a case study of Bulgarian clitics. According to Rivero (1994a), Bulgarian is one of the many Slavic languages which shows Long Head Movement (henceforth LHM): movement in the syntax of a verbal head across another head to a clause-initial position. This kind of movement deserves particular attention because, among other things, it violates a basic principle of Government-Binding Theory, the Head Movement Constraint (Travis, 1984). Moreover, under current views of economy shared by both the Minimalist Program (Chomsky, 1991, 1992, 1995) and Optimality Theory (Prince and Smolensky, 1993; Grimshaw, 1995; Legendre et al 1993, 1995, in press), movement is a costly operation which has to be motivated. In Rivero’s account, LHM is motivated by a ban against clitics in initial position of a clause, a constraint I will refer to as the Tobler-Mussafia (TM) Law, as it is known in Medieval Romance studies. The pattern is exemplified in (1). Throughout the paper, clitics are in italics. (1) a. Az sßm mu go dal. I have him-dat it-acc given ‘I have given it to him’. b. Dal sßm mu go. ‘(I) have given it to him’. Bulgarian is a pro-drop language. In the presence of an overt subject, auxiliary and pronominal clitics follow the subject and precede the non-finite verb (1a). In the absence of an overt subject the non-finite verb appears in initial position and clitics follow (1b). Rivero (1994a) proposes that the non-finite verb dal in (1b) moves across the auxiliary head sßm to a C position. This is an instance of last resort movement which saves an otherwise o ungrammatical structure. In a series of papers, Rivero extends this analysis to a wide variety of languages, including other South Slavic languages (Serbo-Croatian), West Slavic (Slovak, Czech), Romance (Old Spanish), and Celtic (Breton). In Bulgarian, Rivero’s analysis is empirically undermined in several ways. A first complication is that some clitics escape the TM Law. This is for example the case of the future particle šte: (2) Šte mu go dadete. fut him-dat it-acc give ‘(You) will give it to him’. The behavior of šte forces Rivero to analyze šte as a non-clitic modal and to stipulate that the modal particle constitutes a barrier for LHM. A second complication is the distribution of the interrogative particle li which marks yes-no questions. Li observes the TM Law but is otherwise completely movable. (3) a. Vizdal li go e? seen Q him-acc has ‘Has (he) seen him?’ b. Šte go viñdaš li? fut him-acc see Q ‘Will (you) see him?’ 2 Legendre Clitics, Verb (Non-)Movement, and Optimality in Bulgarian c. Ne šte li go viñdaš? neg fut Q him-acc see ‘Will (you) not see him?’ In (3a-b), li is postverbal; in (3c), it is pre-verbal. In (3a), li precedes all other clitics; in (3b), li is the last clitic in the cluster which is interrupted by the verb; in (3c), li is the penultimate clitic. Rivero (1993) claims that li is generated in C . Its status as a bound morpheme forces incorporation of the verb to C, with other clitics remaining o in IP in (3a). In (3b,c) however, the modal particle and the negation each constitute a barrier to movement of V to C. Hence, li must be lowered, right-adjoining to V in (3b) or left-adjoining in (3c). In Rivero’s 1993 comparative analysis of Bulgarian and Serbo-Croatian, li lowering is claimed to be idiosyncratic to Bulgarian. A third complication which Rivero’s analysis does not deal with is the following. In the presence of the negative particle ne, the placement of li is rigid: it occurs strictly pre-verbally in third position following the element which follows ne: (4) a. Ne mu li izpratix knigata? ‘Didn’t (I) send him the book?’ b. Ne si li mu ja dal knigata? ‘Haven’t (you) given him the book?’ c. Ne šte li go viñdaš? ‘Will (you) not see him?’ d. Ne znaeš li? ‘Don’t (you) know?’ One obvious generalization emerges from the data introduced so far. They all involve clitics, a category of elements whose analysis remains controversial. In the tradition of Pollock (1989) and others, Rivero does not distinguish clitic auxiliaries from non-clitic ones in the syntax: the present perfect auxiliary sßm heads an AuxP just like the past perfect auxiliary bjax despite the fact that the former is a clitic while the latter is not (Hauge, 1976). Rivero does not discuss any evidence for clitichood. Rather, she distinguishes functional auxiliaries like sßm which license LHM from lexical auxiliaries like šte which do not license LHM. This distinction is problematic because some lexical auxiliaries, including šte, optionally allow LHM (Embick and Izvorski, 1994). This is shown in (5). (5) a. Šte e izpil konjaka. will is drunk the cognac ‘(He) will have drunk the cognac.’ b. Izpil šte e konjaka. Furthermore, LHM in (5b) is completely unmotivated and this remains an unresolved problem in Rivero’s analysis. As Embick and Izvorski (1994) note, LHM is predicted to be obligatory because it is construed as a last resort movement. In (5), it is only optional and constitutes a marked option according to my informants. In this paper, I propose to reconsider the Bulgarian facts discussed by Rivero by focussing on clitics rather than movement. My analysis differs from Rivero’s in two fundamental ways: one, I follow Anderson (1992, 1995) in claiming that clitics are introduced at PF rather than in the syntax. Two, constraints are ranked and can be violated, as proposed in Optimality Theory (Prince and Smolensky, 1993). These two assumptions lead to a comparatively simple account of Bulgarian which can be summarized as follows: roughly speaking, the distribution of clitics follows from the competition between constraints which favor placing clitics close to the left edge of the clause and constraints which favor placing clitics away from the left edge of the clause. The outcome of every competition depends on which constraints are at work in any given instance (i.e. which functional features are being expressed in the clause). Thus, the outcome is computed for each possible set of features. Crucially, a single ranking of constraints will be shown to be responsible for the variety of empirical facts introduced above. The paper is structured as follows. Section 2 presents a critique of two other analyses: Halpern (1995) and Embick and Izvorski (1994). It leads to the conclusion that the common treatment of auxiliary clitics as heads 3 Legendre Clitics, Verb (Non-)Movement, and Optimality in Bulgarian of syntactic phrases is at the root of the problems that are exposed. Section 3 focusses on an alternative view of clitics which is developed in Klavans (1985) and Anderson (1992). Theoretical and empirical considerations are laid out in support of the conclusion that clitics are not present in the syntax. The case of the future particle šte and the negative particle ne is examined. It is argued that both are clitics; their distinctive property arises from the fact that they violate a constraint satisfied by other Bulgarian clitics, the TM Law. An OT analysis of Bulgarian which incorporates this distinction is developed in Section 3. Section 4 reconsiders the issue of optional LHM in Bulgarian. It is argued that participle fronting is not optional. An OT analysis is offered which exploits the OT notion of input by proposing that the two alternative structures in (5) correspond to two different inputs. Section 6 summarizes the results. 2. OTHER ACCOUNTS 2.1 PROSODIC INVERSION An alternative proposal to Rivero’s LHM is made in Halpern (1995) which relies on a PF operation called Prosodic Inversion (henceforth PI). Enclitics have phonological requirements that can be met by moving the clitic (generated in the leftmost position in their syntactic domain) to the r