A Computational Model of the Spanish Clitic System

In this paper a computational model of the Spanish clitic system is presented. In this model clitics receive a dual analysis in which enclitics are considered inflexions while proclitics are considered independent lexical units, hence proper clitics. The model covers the analysis of simple periphrases that, in addition to auxiliary and modals, have a single content verb (e.g. puede comérselo, se lo ha querido comer) and also the analysis of complex periphrases with more than one content verb (e.g. le hubiera visto comérsela, se la hubiera visto comer). The model introduces three operations on clictis: cancellation, composition and subsumption, and is formalized in Head-driven Phrase Structured Grammar; the standard machinery of this theory is extended with one combination scheme, the head-proclitic rule, and one principle, the clitic principle, that is satisfied by Spanish clitic sentences. A computational implementation of the theory with the Linguistic Knowledge Building (LKB) tool is also reported. Introduction Intuitively, a clitic is an unstressed particle that is attracted to a stressed word, its phonological host, and the resulting object is perceived as lexical unit (see, for instance, the introduction of Nevis, 1991); unlike inflexions and derivations, that are assembled with their stems at the morpho-lexical level of linguistic representation, clitics are combined with their host at the syntactic level. According to Zwicky and Pullum (1983, pp503): “...word-clitic combinality is largely governed by SYNTACTIC considerations. The conditions governing the combinability of stems with affixes are of quite a different sort: they are MORPHOLOGICAL and/or LEXICAL in character, being concerned with the substructure of a finite set of words” However, it is not always clear what is the linguistic level of representation for a given particle; in order to make this distinction Zwicky and Pullum (ibid.) advanced a number of criteria that we summarize as follows: (1) inflexions attach to words of specific syntactic categories while clitics do not exhibit this restriction, so clitics can attach to words of different categories and they often do so, (2) the combination host-clitic is very regular while inflexions show exceptions, (3) the meaning of clitic-host combinations is the same as the meaning of expressions that show no such reduction (e.g. she is gone means the same as she ́s gone), (4) cliticizised forms cannot be affected by syntactic operations, while affixed words can (e.g. no syntactic rule treats I ́ve as a constituent) and (5) clitics can attach to combinations already cliticisized, but inflexions cannot attach to already inflected words. Following these criteria Miller and Sag (1995) and also Abeillé et al. (1996) have classified French clitic pronouns as inflexions (pronominal affixes in Miller and Sag ́s terminology) and Monachesi (1999) has adopted a similar criteria for Italian; however, the case for Spanish is not that clear: according to (1), and perhaps (2), clitic pronouns behave more like inflexions, but according to the other three criteria they behave more like clitics. These criteria reflect a further implicit intuition about the architecture of the grammar and assume that the morpho-lexical and syntactic levels of representation are independent, and that the internal structure of units assembled in the former level (i.e. words) cannot be altered or broken down by syntactic operations. Consequently, if the combination takes place at the syntactic level, the resulting unit is a pseudo-word, or rather a clitic-host combination. From this consideration, a common test to distinguish clitics from affixes is whether the particle can have a wider scope over coordination (point (4) in the list above): if the pronouns are inflexions 1 Although this cannot be ruled out altogether if surface structure and intonation receive an incremental integrated analysis, as in Categorial Grammar (Steedmann, 1991). 2 See also Klavan (1985). assembled with the verb by a morphological operation, they cannot be factored out in coordination operations. However, in Spanish, lo llevó y lo puso sobre la mesa (he/she took it and put it on the table), for instance, can also be expressed as lo llevó y puso sobre la mesa, which is grammatical and has the same meaning. In other cases the grammaticality of the second form is marginal, as shown by te vas o te quedas (you go or you stay) versus te vas o quedas. This kind of examples suggests a dual analysis for Spanish. Further evidence is provided by interruptions and repairs in spontaneous speech; in our corpus, forms like me...muéstrame otra vez los muebles (to-me ... show-me again the furniture) appear often (Villaseñor et al., 2001; Pineda et al., 2002); despite that words can be interrupted in inter-syllable positions, we have observed no cases in which the interruption splits off a stem from its inflexion. Accordingly, if the proclitic were an inflexion it could not be split off after lexical realization. On the basis of these considerations, we propose a dual analysis for clitic constructions: on the one hand enclitics are considered inflexions, but proclitics are considered independent lexical units, which combine with their phonological host in the syntax and are proper clitics, unless the proclitic is assembled with the pronoun at the morphology, either as proclitic (e.g. finite forms) or enclitic, as in reflexives. The basic model Clitic pronouns substitute the direct and indirect object of verbs by accusative and dative pronouns that appear next to verb by its right or left side, forming the enclitic and proclitic constructions respectively. In simple clitic sentences there is only one verb of content, and the clitic pronouns substitute its arguments. Also, in nonperiphrastic constructions the verb is both the cliticized object and the phonological host. This is illustrated in (1). (1) a. Juan muestra [el catálogo]i [a María]j Juan shows the catalogi to Maryj b. muéstraloi [a María]j c. muéstralej [el catálogo]i d. muéstrasejloi e. muéstrasejloi [a María]j f. loi muestra [a María]j g. sej loi muestra h. sej loi muestra [a María]i i. lej muestra [el catálogo]i [a María]j However, when clitics occur in periphrases, the phonological host can be an auxiliary or modal verb different from the cliticized one; in (2) the cliticized verb is mostrar, and it is also the phonological host in (2.b), but in (2.c) and (2.d) the phonological hosts are haber and puede respectively. (2) a. Juan puede haber querido mostrar [el catálogo]i [a María]j Juan could have wanted show the catalogi to Maríaj Juan could have wanted to show the catalogi to Maríaj b. Puede haber querido mostrarsejloi [a María]j c. Puede habérsejloi querido mostrar [a María]j d. Sej loi puede haber querido mostrar [a María]j For this reason we distinguish between the clitic host, the cliticized verb, from the phonological host, and we say that in a well-formed clitic sentence the pronouns attached to the phonological host cancel the corresponding arguments of the clitic host. Following Miller and Sag (ibid.) and Monacheci (ibid.), we consider cliticizised verbs as valence reduced realizations of their basic forms, which require overt complements. We define cliticization as a lexical operation on the basic form of verb; this operation removes the cliticized arguments from its complements list, and places them in a clitic-lists attribute which, in conjunction with the subject and complements attributes, 3 Although the ungrammaticality of this form may be due to considerations related to argument structure, as the action performed by the agent is not received either directly or indirectly by another entity, and this kind of verbs have a reflexive connotation (they are often called reflexive) with the attached particle already built in at the morphology. 4 In our model, auxiliary verbs are subject raising as they are not agentive, and their syntactic subject is the same as the subject of its complement, which is a verbal phrase; similarly, modals, like querer (following Gili Gaya ́s terminology, 1991), are subject control, as they also share their subject with their verbal phrase complements, although these latter forms are agentive (Pineda and Meza, 2004). defines the valence of verbs. Our approach has a strong lexical orientation and we postulate no movement, traces or empty categories, and non-local dependencies are captured through structure sharing, as commonly done in categorical and unification formal approaches to grammar. The model is framed in HPSG (Pollard and Sag, 1994; Sag and Wasow, 1999), and cancellation operations are defined through the standard combination principles of this theory (e.g. head-complement rule, head-specified rule, the GAP principle, etc.). For clictic cancellation to take place, the clitic host must be within the scope of the phonological host, and for this to be the case no pronoun or pronoun sequence can intervene between the phonological and clitic hosts, blocking the binding path between co-indexed argument positions (e.g. pudo verla comerlo versus *lo pudo verla comer). Clitic pronouns sequences present a rigid and idiosyncratic order that poses a severe challenge to the analysis of the phenomenon. In our model we postulate that there is a clitic lexicon which codifies all clitic sequences that occur in a dialect, with the corresponding order and case information, and there is an entry in the clitic lexicon for each sequences of one, two or possible three pronouns; clitic pronouns have a default case (e.g. lo and la are accusative and le and se dative) but they can be used with a different case (e.g. le and se can be accusative given rise to the so-called leísmo) and we define an entry in the clitic lexicon for each sequence of pronouns with a different case assignment. This approach permits to analyze simple clitic sentences in terms of a single cancellation operation. We distinguish three cases: (a) simple lexical cancellation, (b) composite lexical cancellation and (c) syntactic cancellation. Simple lexical cancellation is defined in terms of a lexical rule that implements cliticization and performs the insertion of the pronouns in a single operation, permitting the analysis of (1b-1e) and (2b), for instance. Composite lexical cancellation is defined in terms of two lexical rules: one implements the cliticization operation on the clitic host, and the other performs lexical insertion on the phonological host if structure sharing between the clitic lists of both the clitic and phonological hosts is permitted (i.e. through the head-complement rule), as in (2c). Finally, syntactic cancellation is analyzed in terms of the lexical rule that cliticizes the host, and the head pro-clitic rule that combines an entry in the clitic lexicon with a verbal phrase if the structure of the clitic list attribute of the predicate corresponds with the structure of the sequence in the clitic lexicon (e.g 1f-1h and 2d); this rule captures the intuition that proclitics are proper clitics. Complex periphrasis The model presented so far follows closely Monachesi ́s analysis for Italian, with the exception of the use of the head-proclitic rule whose corresponding effect in Monachesi ́s affixial approach is achieved through lexical rules; however, the analysis of the Spanish complex periphrases with more than one content verb motivates further our dual analysis. Consider sentences in (3): (3) a. Juan pudiera haber visto [a María]i comer [la manzana]j Juan could to-have seen to Mariai to-eat the applej Juan could had seen Maria to eat the apple b. *Pudiera haber vístolei comerlaj c. *pudiera haber vístoseilaj comer d. pudiera habérseilaj visto comer e. *pudiéraseilaj haber visto comer f. Sei laj pudiera haber visto comer (visto + “sei + laj”) (haber + “sei + laj”) (pudiera + “sei + laj”) In this sequence, the subjects of the two content verbs are different (Juan is the one who sees but María the one who eats) and the subject of the composite predicate visto comer, Juan, is shared with the auxiliaries pudiera and haber, and it is also is the subject of the whole periphrasis; in addition, the syntactic object of visto is shared with the subject of comer and the composite verbal phrase visto comer has a composite direct object “se la” occurring as enclitic in (3d) and as proclitic in (3f). In this case, both of the pronouns are in the accusative (i.e. substitute direct objects) and se is used instead of le (with leísmo) or la, as no sequence of two ĺ s pronouns is allowed in Spanish. The sequence shows that two clitic hosts can compose their accusative clitizations if they are next to each other (i.e. accessible), and the result of this operation is an abstract clitic argument. We refer to this operation as clitic composition. This operation is also implement through lexical rules and structure sharing, and clitic sentences of this form are also analyzed in terms of single cancellation. The ungrammaticality of (3b), (3c) and (3e) is due to an idiosyncratic lexical restriction of Spanish for the phonological host, as participles and finite forms cannot have enclitics, while other non-finites forms (infinitive and gerunds) require enclitics always. Next we consider a variant of (3) in which comer has a second dative complement that co-refers with its subject, forming an ethical dative that marks that the subject of this action is also its beneficiary. (4) a. Juan pudiera haber visto [a María]i comer [la manzana]j [para María]i Juan could to-have seen to-Mariai to-eat the applej for-Mariai Juan could had seen Maria to eat the apple for Maria b. Juan pudiera haber visto [a María]i comersei [la manzana]j c. *pudiera haber vístolai comerseilaj d. *pudiera haber vístolai+seilaj comer e. *pudiera haber vístoseilaj comer (i.e. sei = lai+sei) f. pudiera haberseilaj visto comer g. *pudieraseilaj haber visto comer h. Sei laj pudiera haber visto comer i. Lai pudiera haber visto comerseilaj Sentences (4a) does not really occur in the language and it is used only as an aid to illustrate the meaning of (4b) in which comer has already the dative reflexive se as enclitic; the clicitization of the direct objects of visto and comer and its further composition gives rise to the abstract direct object of the predicate visto comer, represented by lai+seilaj. However, in this composition, the object of visto co-refers with the dative se, and the redundant form lai+sei is reduced as se, with the dative case prevailing, and the remaining seilaj form represents the whole of the abstract clitic argument as shown (4f) and (4h) in the enclitic and proclitic forms respectively. We refer to the reduction of this argument, in which an accusative pronoun is subsumed by a co-indexed dative form, as clitic subsumption. If the co-indexed arguments have the same case, they can also be subsumed in a composition. The analysis of sentences with clitic subsumption is carried out with a single cancellation operation, and the ungrammaticality of (4c), (4e) and (4g) is also due to the lexical restriction on participles and finite forms for enclitics. (4d) shows, in addition, that two co-indexed pronouns cannot occur next to each other, and subsumption