On the Semantics of Pronominal Clitics and some of its Consequences

Abstract Recent work on the acquisition of the binding conditions suggests that pronominal clitics (PCs) encode the presence of an unsaturated argument position. In other words, PC-constructions encode functional abstraction: the argument position related to the PC is re-opened. This interpretation represents a radical departure from traditional analyses (in virtually every syntactic framework, including HPSG and Principles&Parameters), which take PCs to reduce the valence of the predicate to which they are linked, either in the lexicon (HPSG) or in syntax (P&P). In this contribution, I will provide conceptual and empirical motivation for this radical reinterpretation of PC-constructions, by claiming that it considerably enhances the prospects of explanatory adequacy in (at least) the following domains: (a) the acquisition data relative to Principle B Effects in Romance languages; (b) the familiar vs. bound-variable interpretation of PCs; (c) the diachronic relationship between clitic left-dislocation constructions (CLLD) and PC-constructions; (d) the properties of Romance CLLD which are still in need of a deep conceptual account, like the (optional) presence of a resumptive clitic and the recursive nature of the topic projections in the left-periphery. Key words: pronominal clitics, interface conditions, interpretable features, topicalization, left periphery, locality effects, resumptive clitics. *Different versions of this paper have been presented at the University of Utrecht (Utrecht Anaphora Workshop, October 9-10, 1999) as a joint work with Sergio Baauw, and at the University of Trieste (International Meeting of Generative Grammar, February 28 – March 3, 2001). I am grateful to the audiences there for useful comments and discussion. For their stimulating comments on earlier drafts of this article, I am particularly indebted to A. Moro, G. Rigau, A. Tomaselli and two anonymous reviewers. This work has been largely inspired by the analysis of clitic acquisition phenomena developed in Baauw 2000. I am deeply grateful to Sergio for the long discussions we had on many of the topics involved. Needless to say, I am responsible for all errors and shortcomings. Catalan Journal of Linguistics 1, 2002 41-69 On the Semantics of Pronominal Clitics and some of its Consequences* Denis Delfitto Università degli Studi di Verona Dipartimento di Linguistica Letteratura e Science della Comunicazione Via S. Francesco, 22. 37129 Verona [email protected] Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 41 1. The semantics of pronominal clitics The main controversy in the syntactic literature on pronominal clitics (special clitics in Zwicky’s 1977 terminology) revolves around the choice between theories that have clitics as a kind of affixes with special selection properties (see especially Klavans 1985) and theories that take them to be a kind of pronominal arguments with special syntactic distribution (see especially Kayne 1975). This contrast is often referred to as the opposition between the so-called base-generation approach (clitics are generated in their surface position on some syntactic host as pieces of inflectional morphology) and the so-called movement approach (clitics are generated in canonical argument position and undergo movement in syntax). More recently, there have been analyses that try to concile the affixal status of clitics with the advantages of a movement analysis, under the hypothesis that movement involves a clitic-related silent category (see especially Sportiche 1996). Within lexicalist approaches to the theory of grammar, the cliticization process has been regarded as a rule which applies to verbs subcategorizing for complements in order to reduce the number of these complements by transferring the relevant information in the lexical structure of the complex verbal head containing the clitic. For instance, the lexical entry for Italian lo legge (‘reads it’) can be represented as follows in theoretical frameworks such as HPSG (Monachesi 1995: 89): (1) [COMPS <>, CLTS{NP[acc][3sgm]}] What has largely gone unnoticed is that all these theories (despite the different labor division which they propose between lexicon and syntax or, within syntax, between base-generation and movement) essentially share the same basic assumption concerning the semantics of pronominal clitics: clitic-constructions are propositional, since the clitic discharges one of the argument positions of the (verbal) predicate. When we add lo (‘it’) to legge (‘read’), we basically map a transitive verb (logical type <e, <e,t>>) into an intransitive verb (logical type <e,t>): lo legge gets the semantic representation λx (x lo-legge), to the effect that, by adding a further argument to lo legge, we actually get a proposition. Of course, theories differ as to where this semantic operation (consisting in saturating one of the argument position of a verbal predicate) actually takes place: either in the lexicon (HPSG) or in syntax (with further choices to be made, for syntax-oriented theories, as to what exactly fills the clitic-related argument position in clitic-constructions). In this contribution, I would like to challenge this uniform semantic analysis of clitics and clitic-constructions. More precisely, I would like to show that an alternative semantic analysis of clitic-constructions is not only possible, but in effect also desirable on conceptual and empirical grounds. This alternative analysis is based on the hypothesis that clitic-constructions are unsaturated expressions: the argument position related to an (object) clitic is re-opened, in the sense that it is interpreted as a bound variable (a variable bound by a λ-operator). As a consequence, sentences involving one or more pronominal clitics are predicates, for42 CJL 1, 2002 Denis Delfitto Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 42 mally represented as λ-abstracts. This entails that when we add lo to legge, we do not saturate the object argument position: lo legge, as well as legge, is a one-place predicate, which is represented as in (2) as a matter of bare output conditions on syntactic representations at the interpretive interface: (2) lo legge (read-CL3sgm) →λx λy (x legge y) However, there is an important difference. In the case of the verbal head legge, (2) is simply a convenient logical form used to represent the fact that legge is, semantically, a two-place predicate. In the case of the complex head lo-legge, the representation in (2) is partially encoded in syntax: what is encoded is the presence of a λ-operator binding the clitic-related variable y. In other words, I propose that clitic-constructions are grammatical tools designed to encode λ-abstraction over the argument position(s) of (verbal) predicates. The relevant bare output conditions can be made precise along the following lines. Suppose that pronouns are translated into free variables at the interpretive interface, due to the interface condition formulated in (3) (cf. Heim and Kratzer 1998): (3) If αis a pronoun or a trace, i is an index and g is a variable assignment whose domain includes i, then ||αi||g= g(i). Consider further the two following refinements of the condition (3). First, let us assume that the role played by indexes in (3) is taken over by ϕ-features, in accordance with minimalist guidelines (cf. Chomsky 1995): a pronoun is in the domain of a variable assignment function g not because of the fact that it is endowed with the index i but because of the fact that it is endowed with ϕ-features. Second, let us assume that g has a certain bunch of ϕ-features in its domain only if these ϕ-features are the proper subset of a larger feature-cluster which crucially includes some substantive feature (typically, the [+animate] feature proper to full pronouns, as discussed in Corver and Delfitto 1999): (4) g includes AGR only if AGR ⊂Σ, where Σis a set of features among which there is at least a substantive feature (where AGR = a bunch of ϕ-features). Given these assumptions, clitic-constructions may be assumed to encode λabstracts in the following way. Clitics have been argued to undergo movement in two distinct steps: first, the clitic moves as a whole (that is, as a DP), creating a spec-head configuration with the lexical head to which it is related; second, the clitic moves as a head (that is, as a D) (see Corver and Delfitto 1999 and the references cited there). After the second step, clitic-movement gives rise to abstract syntactic configurations such as (5) ([NP e] is supposed to express the fact that pronominal clitics are transitive structures involving an empty NP-complement of D; for a critical discussion of this hypothesis, see Corver and Delfitto 1999): On the Semantics of Pronominal Clitics CJL 1, 2002 43 Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 43 (5) Di…[DP ti[NP e]] The fact that the empty DP in (5) is in a checking configuration with the lexical head to which it is thematically related ensures that it is enriched with the substantive feature required for it to be interpreted as a free variable (at the interface, the DP will be translated into a variable x which is visible to the variable assignment g). At the same time, the head of the X-chain (that is, D) is translated into a λ-operator, as an instance of a more general output condition at the interpretive interface. Namely, notice that D is nothing else, by hypothesis, that a bunch of ϕ-features (i.e. AGR). When it moves, it creats an AGR-chain, whose head is arguably in an A’-position. A’-chains involving AGR generally encode functional abstraction, as is shown by the interpretation of structures involving Quantifier Raising (cf. Heim and Kratzer 1998): (6) a. No linguist is a genius. b. (No linguist) [λx (x is a genius)] Given the interface condition (3) (and the proposed replacement of indexes with AGR), the λ-operator in (6b) is encoded by the index of the QP no linguist (in our terms, by the AGR-element realized on the QP in A’-position). The QP itself provides the argument of the λ-abstract created by the abstract AGR-chain. The difference with respect to clitic-movement is that the D-head has no semantic content (by hypothesis, it is nothing else than AGR): it simply contributes to encoding formal objects of the sort λx (…x…), without providing an argument for the λ-abstract, that is, without saturating the predicate.1 44 CJL 1, 2002 Denis Delfitto 1. An anonymous reviewer points out that reference to the substantive feature [+animate] seems misleading in the case of partitive clitics such as Romance en/ne. This criticism would be justified if I were willing to maintain that clitic-movement is intended to provide the clitic-chain with the substantive feature [+animate].This is not the case. In Corver and Delfitto 1999, it is proposed that the real issue is underspecification: clitics are underspecified for one of the substantive feature [+animate] and [-animate] and clitic-movement is intended to provide the clitic-chain with one of these substantive specifications. The asymmetry between ‘animate’ and ‘inanimate’ actually holds only for strong pronouns, which are typically specified as [+animate] and only marginally as [-animate] (cf. the discussion in Corver and Delfitto 1999). The same reviewer also complains about the lack of complete symmetry between the logical forms created by clitic-movement and the logical forms created by Quantifier Raising, in that only the latter gives rise to the canonical tripartite structures that constitute the syntax of generalized quantifiers. This criticism is based on a misunderstanding. With clitics, there is no tripartite structure because of the fact that there is no generalized quantifier as the head of the A’-chain (hence no tripartite structure based on the relational interpretation of determiners). As explicitly proposed in the main text, I interpret A’-chains involving an abstract AGR element as encoding functional abstraction. The λ-abstract may be saturated by a generalized quantifier (even in the case of clitic-constructions, as in Qualcuno, lo ho visto ‘someone, I’ve seen him’) or by a referential (empty) topic, whose logical type is <e>. Contrary to what is suggested by the anonymous reviewer, the empty topic cannot provide the restriction clause within a tripartite structure: empty topics simply instantiate a distinct logical combination with λ-abstracts with respect to generalized quantifiers. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 44 The conclusion is that the interpretation of clitic-constructions as unsaturated expressions follows from the syntactic properties of clitic movement and some plausible bare output conditions concerning pronominals and abstract AGR-chains. In this paper, I want to provide empirical evidence in favor of the functional interpretation of pronominal clitics. In section 2, I will contend that the functional interpretation favors a straightforward analysis of acquisition facts and language change facts, while allowing a unitary semantics for pronominal clitics. In section 3, I will show that the functional interpretation is an important heuristic tool in order to establish important comparative differences among different types of topicalization constructions. In particular, I will develop the basic insight that clitic-constructions are usually analyzed as (possibly hidden) clitic left-dislocation structures: the (possibly null) dislocated topic is the argument of the λ-abstract created by clitic-displacement. In the last section, I will extend this analysis by showing that it leads to an elegant solution of some long-standing problems concerning the role and the distribution of clitic-resumption in Romance clitic left-dislocation structures. 2. Empirical evidence for the functional interpretation of pronominal clitics 2.1. Acquisition facts2 Recent acquisition research has revealed that the so-called Delay of Principle B Effect (DPBE) which has been found in English and other Germanic languages such as Dutch does not show up in Romance languages (Italian: McKee 1992, French: Hamann and Philip 1997, Spanish: Baauw, Escobar and Philip 1997), at least when the pronoun involved is a clitic. In a nutshell, the interpretation assigned to (7a) by English children is quite different from the interpretation assigned by Spanish children to the Spanish equivalent of (7a), given in (7b): (7) a. The girl is pointing to her. b. La niña la señala. More precisely, English children exhibit a clear non-adult-like behavior in admitting coreference between the subject and the pronoun in approximately 50% of the cases (for adults, the coreference construal is excluded by the application of Principle B of Binding Theory). On the contrary, Spanish children behave (almost) adult-like, in excluding coreference in approximately 90% of the cases. For English and Dutch, it is widely accepted that the DPBE is a consequence of the processing limitations of the children, under the hypothesis that the application of Principle B in (7a) involves a transderivational computation whereby children have to compare a binding construal (the pronoun is a variable bound by a λ-operator) and an accidental coreference construal (the pronoun is a free variable). Since this proOn the Semantics of Pronominal Clitics CJL 1, 2002 45 2. This section is largely based on the research results achieved in Baauw 2000. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 45 cessing burden arguably exceeds the short-memory resources of the kids, the linguistic computation breaks down before the children are able to decide whether accidental coreference is admitted. We thus expect children to adopt a guessing behavior, and this seems indeed confirmed by the experimental results (coreference is admitted in around 50% of the cases; cf. especially Grodzinsky and Reinhart 1993). This line of analysis is strongly corroborated by the finding that English children virtually behave adult-like when the subject is not an object-referring linguistic expression (in this case, the prediction is that the accidental coreference construal is simply irrelevant, to the effect that Principle B directly applies to exclude coreference between the subject and the pronoun). In effect, coreference is correctly excluded in around 85% of the cases in sentences like (8) below: (8) Every girl is pointing to her. How is the functional interpretation of pronominal clitics relevant for this whole issue? Notice that there is no straightforward explanation for the fact that Spanish (and, more generally, Romance-speaking) children do not exhibit the DPBE with pronominal clitics. The main reason is that the property involved in the derivation of the DPBE (processing troubles brought about by a transderivational computation) are not easily parametrizable: we do not expect Spanish children to follow a different strategy in assessing whether coreference is admitted, nor are we ready to concede that their performance skills are higher than those of their English peers (for an in-depth discussion of the shortcomings of other analyses that have been proposed, see Baauw 2000). However, the solution to this acquisition puzzle is straightforward if we adopt the functional analysis of clitic-constructions: since Romance pronominal clitics are interpreted as bound variables as a matter of bare output conditions, the accidental coreference construal (based on the interpretation of the pronoun as a free variable) is simply excluded in cases like (7b), to the effect that binding (hence Principle B) will apply to rule out coreference between the subject and the pronominal clitic. In a sense, adopting the functional analysis of clitics allows us to analyze (7b) in the same way as (8): in both cases binding is the only legitimate construal, because either the antecedent or the pronoun is not objectreferring (the antecedent in (8) and the pronoun in (7b)). I conclude that the proposed semantics elegantly fits these acquisition data.3 2.2. Language change data If clitic-constructions are unsaturated expressions, they have to be combined with one or more arguments (depending on how many pronominal clitics are present) in order to get a proposition. One of the combinatorial possibilities that arise is that the argument that combines with the λ-abstract be a syntactic topic. This entails 46 CJL 1, 2002 Denis Delfitto 3. For a critical discussion of the issues raised by the proposed analysis and further arguments in favour of it, see Baauw 2000. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 46 that the reason why (9a) is interpreted as a sentence (a saturated expression receiving a truth-value) is that (9a) has the structure shown in (9b), with an empty topic combining with the λ-abstract: (9) a. Marcello lo legge. ‘Marcello reads it.’ b. [Top e] [λx (Marcello legge x)] In other words, the reason why clitic-constructions are interpreted as sentences is that they involve an additional structural layer, where a (possibly null) topic is realized (semantically, the topic saturates the clitic-related argument-position). The conclusion is that clitic-constructions have the syntax of Romance clitic left-dislocation4or, to put it in more salient terms, that Romance clitic left-dislocation is one of the most prominent syntactic formats to which pronominal clitics are expected to give rise. This important consequence of the present proposal is apparently confirmed by the observation that in the period of transition from spoken Latin to Romance pronominal clitics were arguably first used in combination with a phonologically realized dislocated argument. So, in one of the oldest documents of the Italian language, the so-called Placiti Cassinesi (a tenth-century AD legal document), pronominal clitics appear in a construction which responds positively to all diagnostics for clitic left-dislocation (cf. Cinque 1990, Vincent 1997): (10) Sao ko kelle terre, per kelle fini ke ki contene, I know that these lands within those bords that here are described trenta anni le possette parte Sancti Benedicti for thirty years them owned the monastery of Saint Benedict ‘I know that these lands, within the bords that here are described, for thirty years were owned by the monastery of Saint Banedict’ As is well-known, the source of pronominal clitics in Romance is constituted by the Latin pronoun ille (originally a demonstrative), whose use started as a means of marking objects or, more generally, internal arguments as referring to items which are taken as given, that is, familiar in the discourse context. In this perspective, it is quite telling that the Latin pronoun ipse, which is the source of 3rd pers. pronouns and of definite articles in some Romance varieties, never represents the etymon of object clitics in Romance. The point is that ipse was used not simply to mark a topic, but also crucially to signal topic-change, that is, to resume non-familiar topics (a use that dates back to the classical usage; cf. Vincent 1997). Under the present analysis, the natural hypothesis to be put forward is that when the heirs of ille came to be used to encode functional abstraction over the clitic-relatOn the Semantics of Pronominal Clitics CJL 1, 2002 47 4. For a detailed presentation of the syntactic properties of Romance clitic left-dislocation, see especially Cinque 1990. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 47 ed argument positions, the topic that ille used to mark as familiar at some preceding stage came to be realized in the dislocated topic position typical of clitic leftdislocation structures. This was not a minor change, of course: it involves what has often been characterized as the passage from dependent-marking grammars, where the status of arguments is encoded through morphological marking on the arguments themselves, to head-marking grammars, where the status of arguments is encoded through marking on the verbal head (in this case, by means of clitic morphology). Since clitic-structures involve the presence of dislocated topics, this shift was functional to the preservation of flexible word-order after the disappearance of case-morphology on the arguments of the verb. The continuity between these two stages is constituted by the fact that clitics encode λ-abstraction over argument-positions that were originally saturated by familiar topics, that is, by expressions referring to entities that were salient in discourse. On these grounds, the grammar of cliticization clearly envisages empty topics (as in 9a) besides phonetically realized topics, as in (10). Since the dislocated topic refers to a salient discourse entity, it need not be phonetically realized: it can be syntactically represented as an empty topic (cf. (9b)) which is easily identified in discourse. As a consequence, there is no need to hypothesize a passage from grammatical systems where clitics resumed overt dislocated arguments to grammatical systems where clitics resumed extra-sentential discourse entities, contrary to what is proposed in Vincent 1997: «…the origin of clitic pronouns is to be found in contexts involving pronominal resumption of a focal and dislocated argument. From these contexts, it generalizes to others in which the co-referential or antecedent item is elsewhere in the discourse or extra-linguistic context» (Vincent 1997: 161). In effect, I would like to propose that the dislocated argument never was a focal element, it was a familiar topic since the very beginning; and resumption of extra-sentential antecedents simply does not exist, since it can be reduced to resumption of phonologically unrealized dislocated topics. The grammar of clitic left-dislocation is virtually the same as the grammar of simple clitic-constructions (even though the element saturating clitic-constructions need not be a left-dislocated topic).5So, if we are not surprised to find clitics com48 CJL 1, 2002 Denis Delfitto 5. A reviewer points out that this claim appears to be too strong, in view of certain prima facie asymmetries between clitic-constructions and clitic left-dislocation. A first case concerns anaphorically dependent DPs that can be resumed by a clitic but cannot be left-dislocated. This is shown by the contrast between (i) and (ii) in Catalan: (i) Tots dos estan molt contents perquè s’han comprat la mateixa camisa, them both are very happy because they bought the same shirt però jo la trobo més aviat lletja but I find it’ rather ugly (ii) *La mateixa camisa jo la trobo més aviat lletja the same shirt, I find rather ugly’ Although I fully agree on the judgements (that can be easily extended to other Romance languages, among which Italian), there is a straightforward line of analysis for this sort of contrast Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 48 bining with overt dislocated topics in the oldest stages of Romance, we have serious reasons to believe that the speakers using these kind of constructions were also able, since the very beginning, to understand and produce simple clitic-constructions. On the Semantics of Pronominal Clitics CJL 1, 2002 49 that is fully compatible with the claim that simple clitic constructions are hidden left-dislocation constructions. Namely notice that the empty topic in (i) need not be identified with the dislocated nonreferential DP in (ii). In the discourse situation involving (i), it is quite possible that the empty topic is identified with something like the definite description «the shirt bought by them both», on partial analogy with the anaphoric mechanisms proper to relatively complex cases of pronominal anaphora, like ‘donkey-anaphora’. In other words, the ungrammaticality of (ii) need not have a direct import on the interpretation of (i), since there are rather obvious alternative ways to identify the content of the empty topic. Another potentially problematic case concerns clitic-doubling, as it surfaces in Catalan and many Spanish varieties. If clitic-constructions involve an empty topic, clitic-doubling should be impossible. In fact, this criticism would be sound if one could show that there is no way of interpreting the doubling lexical DP as the referential expression saturating the λ-abstract encoded by the clitic-construction. However, the interpretation of the doubling DP as a subject of predication that saturates a λ-abstract is in fact the interpretation of clitic-doubling which has been proposed in a recent paper by J. Aoun for Lebanese Arabic (cf. Aoun 1999). His analysis entails that the doubling DP gets generated in some subject position and that the rest of the sentence undergoes massive displacement, much in the spirit of Kayne’s Antisymmetry hypothesis. This proposal is supported by substantial evidence, among which one should mention matters of scope interpretation. Of course, it is not clear whether this analysis can be generalized to all instances of cliticdoubling. But it should be clear that my hypothesis, far from being incompatible with the existence of clitic-doubling, paves up the way for a new analysis of clitic-doubling. In this respect, I should remind the reader that the essential feature of my proposal is that clitic-constructions give rise to unsaturated λ-abstracts: what I predict is the existence of a semantic subject of predication (saturating the λ-abstract in order to give rise to a proposition), independently of the specific syntactic execution of this subject of predication (which may be realized as a left-dislocated constituent, as a doubling constituent or even, as in the case of bound clitics to be discussed below, as a grammatical subject). These remarks lead me to the discussion of a third potentially problematic contrast: the Principle C violation effects found in (iii) are completely absent in (iv): (iii) Al Joaniprokva dir que no li’ajudaries [i≠k) ‘Joan, (s)he said that you won’t help’ (iv) Prokva dir que no li’ajudaries [i≠k] or [i=k] ‘Joan, (s)he said that you won’t help’ As anticipated above, the solution is pretty straightforward: the coreferential construal corresponds to a logical form in which the saturating expression is the grammatical subject pro. In other words, bound clitics correlate with logical expressions where two instances of the same variable are bound by a λ-operator,along the lines of (v) (for the relevance of this analysis in an acquisition perspective, see Baauw 2000): (v) [λx (x va dir que no ajudaries x)] (pro) Finally, I would like to mention the fact that overt dislocated topics tend to get a contrastive flavor that is not detected with empty topics. Here, I will simply propose that constituents interpreted contrastively need phonological realization: as a consequence, empty topics will be identified with discourse referents that cannot be contrasted with other (perhaps more salient) discourse referents, as is the case with overt dislocated phrases. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 49 More generally, I will try to show that the key for the understanding of the syntactic properties of topics lies in the different ways in which semantic binding is encoded in morphosyntax (either by means of clitic-constructions or by means of (feature-)movement to a non-L-related checking position in the left periphery). 3.2. Blocking effects, adverb effect adverbials and the status of CLLD-topics English topicalization gives rise to a large variety of intervention effects that do not play any role in CLLD. It is fair to say that a part of these effects are still poorly understood and represent a long-standing issue in locality theory (for a survey and updated discussion, see especially Culicover 1996, from which many of the examples below are drawn). A case in point is constituted by the status of multiple topicalization and by the blocking effects triggered by intervening topics on subsequent topicalization and wh-movement. Starting with multiple topicalization, it is not impossible to find cases where it seems to give rise to perfectly legitimate sentences, as in (18) below (notice that embedding does not affect grammaticality): (18) a. (They told me that) to that man, liberty we would never grant. b. (They told me that) liberty, to that man we would never grant. However, there are cases where multiple topicalization is seriously degraded, as in (19): (19) *I said that that book to me Maxim gave. As for the blocking effects of embedded topics on subsequent topicalization, Culicover emphasizes that certain constructions can be rescued by adopting special prosodic patterns, as is the case in (20a) below: (20) a. To Terry, I think that the MONEY, Lee gave. Unfortunately, cases as in (20b) cannot be rescued by any particular prosodic pattern11: (20) b. *The money, I think that Terry, Lee gave. An empirical paradigm of comparable complexity is also detected when we consider the blocking effects of embedded wh-phrases on subsequent topicalization. The following examples (due to D. Pesetsky) are sufficient to show that it is not difficult to find cases where the dislocation of a topic over a wh-phrase leads to grammatically degraded results: 56 CJL 1, 2002 Denis Delfitto 11. See Culicover 1996 for an interesting line of analysis of the kind of contrast exemplified in (20). Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 56 (21) a. On a nice day like this, what are you worrying about? b. ?A book like this, why should I buy? c. ?*This book, to whom should I give? There is, however, a subset of the data with respect to which the results are virtually uncontroversial. Intervening topics uniformly block subsequent wh-movement. This fact is exemplified by the sentences in (22), where the wh-phrase is moved across the topic to the left-periphery of the same clause, and by the sentences in (23), drawn from Browning 1996, where the wh-phrase moves higher up to the left-periphery of the main clause: (22) a. ??The man to whom that book I gave… (Rizzi) b. *You know what in Scotland they eat. (Hudson) c. ??I won’t be able to remember what on a nice day like this I was worrying about. (Pesetsky) (23) a. *Who did Leslie say that, this present, Kim gave to? b. *To whom did Lee think that, this present, Robert gave? c. *What did Lee insist that, for Kim, Robin should buy? c. *Who did Robin say that, this present, t gave Lee? Browning 1996 observes that there seems to be only one systematic exception to the facts in (23), constituted by the absence of blocking effects on wh-movement when the topic is one of the adverbials which have been independently identified as mitigating the that-trace effects occurring when a wh-phrase is extracted from a subject position (adverb effect adverbials). This is shown by the perfect grammatical status of the sentences in (24) ((25) shows that both adverbials involved in (24) are adverb effect adverbials): (24) a. Who did Leslie say that, for all intents and purposes, t was the mayor of the city? b. What did Lee insist that, under normal circumstances, Robin would give to us? (25) a. Robin met the man Leslie said that for all intents and purposes t was the mayor of the city. b. Lee forgot which dishes Leslie had said that under normal circumstances t should be put on the table. Let us consider first the behavior of adverb effect adverbials. Browning 1996 proposes an analysis of the correlation between the absence of that-trace effects and the On the Semantics of Pronominal Clitics CJL 1, 2002 57 Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 57 absence of blocking effects on wh-movement in terms of CP-recursion. The sentences in (24) are associated with the syntactic structure exemplified in (26), where t is the trace of the displaced wh-phrase: (26) …[CP ti’ [C’ thati[CP [C’ ti[IP ti… The wh-trace in the spec of the higher CP-layer (ti’) induces agreement on C, which is sufficient to license the wh-trace in subject position according to the elegant analysis of that-trace effects developed in Rizzi 1990. In effect, it is CP-recursion that ensures that the empty subject be formally licensed by Agr-on-C: although the complementizer that, being lexically incompatible with the realization of Agrfeatures, is expected to block Agr-on-C in the higher C-head, the Agr-features can be successfully transmitted to the foot of the C-chain (the C-head of the lower CP), which is phonologically empty. At the same time, the presence of the C-chain ensures that the wh-phrase be moved over the topicalized adverbial without producing any Relativized Minimality violation (due to the so-called equidistance effect proposed in Chomsky 1995). Technically, we may assume that CP-recursion is performed by allowing the C-head to target its own maximal projection (cf. Watanabe 1992)12. Moreover, I want to propose that adverb effect adverbials, being non-referential linguistic expressions (in the sense that they do not denote discourse entities), undergo a kind of default realization in the left periphery: the C-head to which they are related contains neither non-interpretable features (as is the case with the Top-head involved in the standard cases of English topicalization) nor interpretable features (as I will argue to be the case in CLLD). My hypothesis is that CP-recursion is limited to this kind of ‘minimal’ C-heads, which do not contain any of the syntactic features (relevant either to the syntactic computation or to the interpretive interface) that are typically realized in the left-periphery. In other words, the fact that non-referential adverbials do not induce intervention effects on wh-movement can be traced back to the distinct syntactic status of these adverbials in the left-periphery and more particularly to the fact that their interpretation does not involve the activation of any formal (either interpretable or non-interpretable) feature within the Comp-system. Given this analysis of the absence of potential RM-violations with adverb effect adverbials, the question arises whether it can be legitimately extended to the absence of RM-effects with CLLD-constructions. Empirically, it is easy to see that all violations which have been analyzed in terms of intervention effects brought about by an intervening topic simply do not arise with Italian CLLD.13 This is shortly 58 CJL 1, 2002 Denis Delfitto 12. Subtle theoretical questions arise concerning the justification of intermediate movement of the wh-phrase to the C-layer that undergoes recursion. The answer might be quite different depending on whether one assumes that wh-movement is motivated by feature-checking (Chomsky 1995) or by Antisymmetry requirements at the phonological interface (Kayne 1994, Moro 2000). I will not discuss this important issue here. 13. The situation is arguably different in other Romance languages, such as Catalan, where an intervening topic may give rise to strong island effects. The contrast is shown in (i) (Catalan) and (ii) Italian: Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 58 exemplified by the well-formedness of the Italian counterparts of (19) (multiple topicalization), (20) (intervention effects of a topic on subsequent topicalization) and (23) (blocking effect of an intervening topic on wh-movement), which are given below: (27) a. Ho detto che, quel libro, a me Marcello lo ha dato. I have said that that book to me Marcello it has given ‘I said that, that book, to me Marcello has given.’ b. Il denaro, io penso che, a Marcello, Teo lo abbia dato. the money I think that to Marcello Teo it has-SUBJ given ‘The money I think that, to Marcello, Teo has given.’ c. Che cosa pensi che, per Marcello, Teo debba comprare? what do you think that for Marcello Teo should-SUBJ buy ‘What do you think that, for Marcello, Teo should buy?’ Since topic-phrases in Italian CLLD appear to behave as topicalized adverb effect adverbials in English, one might propose that Romance topics constitute a generalization of the syntactic structure that has been argued to be relevant for aparticular case of adverb-preposing in English. The reason why the sentences in (27) are perfectly acceptable would be that the dislocated topics are generated in positions related to default C-heads in the left-periphery. More precisely, the fact that C is not endowed with non-interpretable PRED-features which trigger movement to spec-of-CP and turn spec-of-CP into a non-L-related checking position (that is, an A’-position) paves the way to CP-recursion, yielding an equidistance effect and voiding RM-violations. In principle, this analysis seems to be compatible with the semantics of pronominal clitics for which I argued in the previous sections: it is the presence of the λ-abstracts induced by the syntax of cliticization that makes recourse to formal features such as PRED, activating the C-system, completely redundant, and allows the dislocated topic to be realized in a syntactically inert CP-layer (the only requirement being the realization of the topic-phrase in a convenient structural position in order for it to count as the argument of the λ-abstract). Although conceptually feasible, this analysis is clearly wrong. It predicts that dislocated topics in CLLD not only allow subsequent wh-movement but also void that-trace effects, as a consequence of CP-recursion. The prediction cannot be checked for Italian (where that-trace effects are notoriously absent) but can be checked for French (cf. Rizzi 1997, from which the following examples are drawn): On the Semantics of Pronominal Clitics CJL 1, 2002 59 (i) *Com creus que, de Cuba, en parla Chomsky? ‘How do you believe that Chomsky talks about Cuba?’ (ii) Come credi che, di Cuba, Chomsky parli, nel suo libro? ‘How do you believe that Chomsky talks about Cuba (in his book)?’ Here, I will concentrate on Italian CLLD, postponing a detailed investigation of the comparative facts to a future occasion. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 59 (28) a. *?Voici l’homme que je crois qui, ton livre, t pourrait l’acheter. here is the man that I believe that your book could it buy ‘Here is the man that I believe that, your book, could buy.’ (28) b. Voici l’homme que je crois qui, l’année prochaine, t pourra ‘Here is the man that I believe that next year could nous aider. help us.’ If dislocated topics in CLLD were uniformly analyzed as English adverb effect adverbials, the contrast between (28a) and (28b) would be completely unpredicted. In fact, the grammaticality of (28b) suggests that the proposed parallelism partially goes through, since Romance adverb-preposing actually dissolves that-trace effects14. However, the ungrammaticality of (28a) also clearly indicates that objectreferring topics in CLLD cannot be assigned the same analysis as English preposed adverbials15. To summarize the analysis so far: topics in Italian CLLD exhibit a dissociation of the cluster of properties that I have argued to be relevant for adverb effect adverbials in English topicalization. Italian topics do not block wh-movement over the topic (they do not trigger any RM-violation) but clearly give rise to that-trace effects. The presence of that-trace effects suggests that there is no CP-recursion in CLLD. A way to derive this result consists in hypothesizing that the feature-endowment proper to the C-head to which the topic is related extends beyond the presence of mere categorial features. Let us thus suppose that Italian topics are related to a Topic-head that involves the realization of some interpretable feature. This is sufficient to exclude the presence of a suitable landing site for intermediate whmovement: even if the Top-head undergoes recursion, its spec is not allowed to 60 CJL 1, 2002 Denis Delfitto 14. Of course, temporal adverbs such as next year in (28b) contrast with adverb effect adverbials in that the former, contrary to the latter, are easily analysed as object-referring expressions. However, this does not constitute a problem for a uniform syntactic analysis of topics. Below, we will see that even object-referring expressions can undergo default syntactic realization in the left-periphery (triggering CP-recursion), provided they are syntactically unrelated to the argument position of the predicate in which they are canonically realized. The possibility of realizing a topic in a default CP-layer depends on whether the semantic role associated with the topic has undergone canonical syntactic realization. 15. One might try to avoid this conclusion by proposing that the ungrammaticality of (28a) is not due to a proper government violation concerning the subject-trace, that is, that it does not represent a that-trace effect. However, I do not see how this line of analysis could be successful, given the minimal pair in (i)-(ii), that is, the substantial grammatical improvement detected when the whtrace is not in subject-position (cf. Rizzi 1997): (i) *?Voici l’homme que je crois que, ton livre, t pourrait l’acheter here is the man that I believe that your book could buy (ii) ?Je ne sais pas à qui, ton livre, je pourrais le donner I do not know to whom your book I could give Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 60 host a phrase that encodes formal features incompatible with those realized in Top (wh-features are in fact related to the Focus-position, cf. Rizzi 1997). The correct result follows: the absence of a legitimate intermediate landing site for the wh-phrase brings about the absence of Agr-on-C; as a consequence, the subject wh-trace will not be properly governed (explaining the degraded status of 28a). However, we do not expect the topic to block wh-movement. Being interpretable, the formal feature realized in the Topic-head needs not be erased. As a consequence, spec-of-Top will not qualify as an A’-position (under the fairly plausible identification of A’-positions with non-L-related checking positions) and the topic-phrase filling it will not interfere with wh-movement under Relativized Minimality. To put it more effectively, the absence of intervention effects follows from the fact that the realization of Italian topics does not involve syntactic operations of feature-checking. In English, topics are linked to Topic-heads which involve the realization of the non-interpretable feature PRED, triggering checking, to the effect that the interference between wh-phrases and topics is predicted under Relativized Minimality (both topics and wh-phrases fill A’-positions). Although topics are interpreted in essentially the same way in English topicalization and Italian CLLD (they are argument of a λ-abstract), they deeply differ in syntactic behavior as a consequence of the different way in which λ-abstraction is syntactically encoded. In English, the λ-operator corresponds to a formal feature – PRED – which enters checking configurations and yields A’-interveners, whereas in Romance the λ-abstracts are created – without syntactic activation of the Comp-system – as the result of the proposed semantics of cliticization. The next natural question concerns the status of the interpretable feature – let us call it TOP – which is arguably realized in the Top-head related to topic-phrases in CLLD. If there is an interpretable feature in Top, we expect it to affect the interpretation of the topic-phrase in its spec. I would like to propose that there is in fact a largely disregarded interpretive generalization concerning the realization of topics in Italian CLLD: the legitimate topics are exactly those quantified phrases which warrant a collective interpretation of the pronoun contained in the same clause (cf. Kamp and Reyle 1993: 345-356). In order to see this, consider the paradigm in (29) and (30) below: (29) a. *Pochi avvocati, (li) ho incontrati ieri. few lawyers (them) I have met yesterday b. Pochi avvocati hanno assunto una segretaria che apprezzano. (*collective) few lawyers hired a secretary who they like (30) a. Alcuni avvocati, li ho incontrati ieri. some lawyers them I have met yesterday b. Alcuniavvocati hanno assunto una segretaria che apprezzano. (ok collective) some lawyers hired a secretary who they like On the Semantics of Pronominal Clitics CJL 1, 2002 61 Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 61 As can be seen, there is an interesting correlation between the acceptability of a generalized quantifier as a topic in CLLD and the acceptability of collective intrasentential anaphora with the very same generalized quantifier. In (29b) the pronoun cannot easily refer to a small set of lawyers as its antecedent, it can only be interpreted as an individual variable distributing over this small set of lawyers: significantly, few lawyers is not admitted as a dislocated topic in CLLD, as shown in (29a). The correlation goes the other way round in (30): some lawyers is a legitimate topic in CLLD, as shown in (30a), and is able to provide a collective antecedent for the pronoun in (30b). To put it shortly, it seems that a quantified phrase can be correctly licensed as a topic only when it is able to provide a (non-atomic) discourse-entity. This is exactly how the contrast between (29b) and (30b) would be represented in Discourse Representation Theory: some lawyers, contrary to few lawyers, introduces a (non-atomic) discourse referent in the main Discourse Representation Structure. This analysis of the facts is corroborated by the behavior of bare quantifiers, whose inability to provide discourse antecedents is witnessed by the fact that they are not only incompatible with collective intrasentential anaphora, but are also incompatible with intersentential anaphora, as shown by cases as (11) above, repeated here as (31): (31) #Non é ancora arrivato nessuno. Eppure gli avevo detto (loro) di essere puntuale/i ‘Nobody has arrived, yet. And I had told him/them to be on time!’ Bare quantifiers are completely excluded as topics in Italian CLLD, as shown in (32), with the important exception of the existential bare quantifier qualcuno/qualcosa, which will be discussed in the final section of this paper (cf. Cinque 1990): (32) a. *Nessuno, (lo) ho visto. nobody (him) I have seen b. *Tutto, (lo) ho fatto. everything (it) I have done If the conditions governing topicalization in CLLD were essentially those governing semantic composition (that is, the combination of the λ-abstract with a linguistic expression of a suitable logical type), we should not expect the sentences in (32) to be ungrammatical. In fact, the generalized quantifiers nobody and everything (type <<e,t>,t>) should successfully combine with the λ-abstract (type <e,t>). The fact that they are unsuited as topics confirms that there is more to topichood than mere semantic composition: the topic must refer to a discourse entity.16 Let us thus 62 CJL 1, 2002 Denis Delfitto 16. In Catalan bare quantifiers are regularly licensed with a resumptive clitic. In fact, Catalan requires the resumptive clitic in virtually all instances of CLLD. Although I intend to postpone a full discussion of these comparative issues to a future occasion, I would like to emphasize that what these Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 62 hypothesize that this is the content of the interpretable feature TOP realized on the Top-head. In purely semantic terms, TOP constrains semantic composition: the ‘argument’ of the λ-abstract cannot be a second-order predicate (a generalized quantifier of type <<e,t>,t>).17 In terms of discourse-theory and pragmatics, TOP paves the way for the interpretation of topics in terms of the aboutness relation (cf. Reinhart 1981). Last but not least, it is worth emphasizing that this interpretation of TOP as enforcing the presence of discourse-referring constituents is fully compatible with, and in fact supports, the analysis of the diachronic phenomena discussed in section 2.2. above: we have seen that object clitics developed from pronominal forms that were originally used to mark syntactic constituents as referring to familiar discourse entities. Let me briefly summarize the content of this section. In English topicalization, topic positions are non-L-related checking positions. As such, they are expected to give rise to interference effects with wh-movement. In Italian CLLD, the topic is simply the argument of the λ-abstract encoded by means of cliticization: the topic-position is inert for the sake of the syntactic computation and is thus not expected to interfere with wh-movement under Relativized Minimality. In this way, we have provided an answer to question (17c) of section 3.1 (why topics give rise to different intervention effects in different languages?). We have also introduced a principled bipartition, within the Comp-system, between syntactically active positions (where non-interpretable features are realized) and syntactically inert positions (where interpretable features are realized). Given this analysis, we may simply propose that only syntactically active positions have to fill designated slots between Force and Finiteness, whereas inert positions are in a way extra-syntactic (this provides an answer to question (17b)). Finally, we have seen that recursion is a rather pervasive property of the C-system: the problem is rather to find a principled account for the cases where recursion is impossible. This we have done: the C-heads which cannot undergo recursion are those endowed with non-interpretable features that trigger checking, that is, the syntactically active heads in the left-periphery (this provides an answer to question (17a)). On the Semantics of Pronominal Clitics CJL 1, 2002 63 facts seem to suggest is that the feature endowment of TOP does not constrain semantic composition in Catalan, contrary to what happens in Italian. This clearly entails that there is nothing special to be said about qualcuno/qualcosa in Catalan (cf. Section 3.3): qualcuno/qualcosa are simply allowed as generalized quantifiers, independently of the definite or indefinite interpretation they may receive. 17. The argument of the λ-abstract can be an object-referring expression, as in (i), or a property-referring expression, as in (ii): (i) [TOP e] [lo vedo] (ii) [TOP e] [Gianni lo è] As pointed out by A. Moro, the analysis endorsed in the present contribution can effectively deal with the cases in (ii): if lo simply encodes λ-abstraction, all one has to do is allow λ-abstraction over properties. Conversely, if lo is a saturating expression, (ii) is left unaccounted for and has to be dealt with as a distinct case, since one does not want to say that the verb BE is saturated by lo. Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 63 3.3. Some speculations on the distribution of clitic resumption As is well-known, there is an important generalization concerning the distribution of resumptive clitics in Italian CLLD that should be accounted for: resumptive clitics are optional, except with direct objects. As exemplification, consider the following minimal pair (drawn from Cinque 1990): (33) a. Di questa faccenda, non (ne) voglio più parlare. of this matter not (of-it) I want anymore to speak ‘About this matter, I don’t want to speak anymore.’ (33) b. Gianni, *(lo) vedrò domani. Gianni (him) I will see tomorrow ‘Gianni, I will see tomorrow.’ The traditional explanation for this distribution (Cinque 1990, Rizzi 1997) adopts the view that resumptive clitics are inserted whenever there is no empty category available. More precisely, it is assumed that the empty category in (33b) cannot qualify as a variable, since the condition in (34) is not satisfied (given that this condition applies only to empty NP-types, there is an empty variable in (33a)): (34) Variable =def [NP e] in A-position locally A’-bound and operator bound Operator =def bare quantifiers, wh-phrases and null NPs in spec-CP This analysis is clearly incompatible with the view of Italian CLLD that I have proposed in the present contribution, whose basic insight is that there is no syntactic dependency between the dislocated topic and the argument position to which the topic is related. Moreover, the traditional analysis has conceptual and empirical problems. Conceptually, it has a highly stipulative flavor and it is based on an old-fashioned functional theory of empty categories which is too theory-bound to satisfy the present standards of explanatory adequacy. Empirically, it leads to wrong predictions. Both sentences in (35) should be grammatical, given that the empty category in object position qualifies as a variable (nessuno and tutto are bare quantifiers):18 64 CJL 1, 2002 Denis Delfitto 18. That the sentences in (35) should be grammatical is clearly confirmed by the fact that the theory explains the ungrammaticality of their counterparts with a resumptive object clitics, given below as (i) and (ii), by means of the assumption that clitics do not qualify as variables (we should then expect that (i) and (ii) license an empty category in object position as a variable): (i) *Nessuno, lo ho visto (ii) *Tutto, lo ho fatto The reason why the data in (35) have not been interpreted as decisive evidence against the traditional analysis of clitic resumption has probably to do with the fact that the prediction is fulfilled with existential bare quantifiers, that is, indefinites like qualcuno ‘someone’ and qualcosa ‘something’, as shown in (iii)-(iv) below. The dubious idealization that was implicitly proposed is that we Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 64 (35) a. *Nessuno, ho visto. nobody I have seen (35) b. *Tutto, ho fatto. everything I have done Moreover, the theory does not predict any contrast between (36) and (37): since the empty category is a PP, for which the proposed typology of empty categories and definition of variable are taken to be irrelevant, we expect the empty category to qualify as a variable both in (36) and in (37). It is a fact, however, that (37) is quite more degraded than (36): (36) A Gianni, ho parlato ieri. to Gianni I have spoken yesterday (37) *?A nessuno, ho parlato ieri. to nobody I have spoken yesterday Given these shortcomings of the traditional theory, I will try, in the last few pages of this contribution, to sketch the main lines of a radically alternative analysis of clitic resumption, based on the results of the previous sections. My starting point is a set of data reported in Rizzi 1997 as an observation due to Laezingler, showing that the realization of the resumptive clitic in contexts where the clitic is optional triggers a that-trace effect (which is virtually absent when there is no resumptive clitic). The relevant minimal pair is given in (38) below: (38) a. ?*Je me demande qui, au Pape, t oserait lui parler Iwonder who to the pope would dare to him to talk ainsi. in this way (38) b. ?Je me demande qui, au Pape, t oserait parler ainsi. Iwonder who to the pope would dare to talk in this way Under the analysis developed in section 3.2, the absence of that-trace effects in (38b) should be related to the possibility of CP-recursion: what is required in order to license the subject wh-trace is an extra CP-layer which provides a suitOn the Semantics of Pronominal Clitics CJL 1, 2002 65 should concentrate on the well-formedness of (iii)-(iv) and put (35) into the background, so to speak. I will come back to (iii)-(iv) later in this section. (iii) Qualcuno, ho visto Someone, (I) have seen (iv) Qualcosa, ho fatto Something, (I) have done Cat.Jour.Ling. 1 001-259 26/2/03 15:51 Página 65