The analysis which I will present is rather "purely" syntactic and only involves a few well-motivated adjustments to the framework.
On the one hand, argument extraction is not considered as a lexical property but instead as a structural one and it is then taken in charge by the Valence Principle. This principle not only verifies if expected arguments of a head are present in a local tree, but also memorizes the information relative to those that are absent in the [INHER|SLASH] feature value. In this way, subcategorization fulfillment and gap introduction are uniformly handled by the same universal syntactic mecanism in charge of valence checking.
On the other hand, pronominal clitics, which I view as of subtype "specifier", select their lexical host via the [SPEC] attribute. That is, the fact that (pronoun) clitics can appear in a structure is not a lexical property of the host but of the clitics themselves. As they appear as sisters of a head in a "spec-head-schema", weak pronouns then introduce the information (modelized as a [TO-BIND|SLASH] feature value) relative to the particularities of the full argument that they can represent and replace. In this view, pronominal clitics are then lexicaly cross-classified on different partitions that not only distinguish gender, number, case and referentiality but also their "binding" properties.
Finally, because "to-bind" dependencies are now directly introduced in structures by lexical items that are "binders" (for example, relative, interrogative and weak pronouns), the Non-Local Principle needs to be reformulated. This principle will then constrain the value of the "inherited" and "to-bind" [SLASH] attributes of the projection of the head daughter to be the set difference of the "respective union of these attributes on all daughters in a local tree".
The features that I consider decisive for determining inclusion in this class (whose members will be called either `ergative constructions' or `ergative forms' depending whether unaccusativity emerges at the phrasal or lexical level, respectively) are the following:
The basic idea of the present approach is that the features determining an unaccusative behavior are syntactic in nature, even though they can be semantically derived, as in the case of unaccusatives(Levin & Rapaport, 1995). The justification for such a choice is that whereas a semantic generalization grouping together the verbs traditionally labeled `unaccusatives' can be found, convincing semantic generalizations grouping together all verbs and verbal constructions which manifest an `unaccusative behavior' are still missing.
Contrary to the previous GB literature, I reject the assumption that movement of constituents is at the base of an explanation for the unaccusative behavior. Indeed, I will show that an analysis which refuses movements and relies on the more `economical' hypothesis that unaccusative behaviors is always lexically determined, can fully account for the data in question.
The central claim of such an analysis, which is casted in the version of the HPSG theory described in Chapter 9 of Pollard and Sag (1994), is that while unergative forms are characterized by the presence of the subject within the SUBJ list, ergative forms have an empty SUBJ list, their surface subject being the first member of the COMPS list. Unaccusative behaviours follow from this assumption.
Beth Levin and Malka Rappaport, 1995, "Unaccusativity", MIT Press, Cambridge, Mass.
Carl Pollard and Ivan Sag, 1994, "Head Driven Phrase Structure Grammar", University of Chicago Press and CSLI Publications, Stanford: Center for the Study of Language and Information
Note, for instance, the differences between the German verb versprochen and the Spanish verb prometer both meaning promise in terms of their possible participation in structures that have been assumed to be flat in recent HPSG analyses. The German data in (1) illustrates the phenomenon described as 'clause union' and has been analyzed in terms of 'argument composition' by some authors (e.g. Pollard, Kasper and Levine 1994).
(1) Dass es ihm jemand zu lesen versprochen hat
Dass ihm es jemand zu lesen versprochen hat
Dass jemand es ihm zu lesen versprochen hat
Dass jemand ihm es zu lesen versprochen hat
Dass es jemand ihm zu lesen versprochen hat
Dass ihm jemand es zu lesen versprochen hat
'That someone has promised him to read it'
However, in Catalan, Spanish, and most Romance languages characterized by exhibiting clitic climbing phenomena, the equivalent verb does not typically participate in structures analyzed in terms of 'argument composition' (by e.g. Monachesi, to appear), in contrast with other predicates such as querer ( want ) or permitir ( allow ) which characteristically permit clitic climbing in most languages. This is illustrated by the contrasts in (2) and (3) obtaining in most Spanish dialects.
(2) a. Lo quiso comer
Cl.it wanted.3Sg eat
Cl.it wanted.3Sg eat
b. Quiso comerlo
'S/he wanted to eat it'
c. *Lo prometio' comer
Cl.it promised.3Sg eat
d. Prometio' comerlo
'S/he promised to eat it'
(3) a. Se lo permitio' dejar en mi casa
Cl.him.Dat.Cl.it.Acc wanted.3Sg leave in my house
b. Le permitio' dejarlo en mi casa
Cl.him.Dat allow leave.Cl.it in my house
'S/he allowed him to leave it in my house'
c. *Se lo prometio' dejar en mi casa
Cl.her .Cl.it promised.3Sg leave in my house
d. Le prometio' dejarlo en mi casa
Cl.her promised.3Sg leave.Cl.it in my house
'S/he promised to read it to her after dinner'
Assuming the argument composition approach to treat these phenomena is in the right direction, if the kinds of differences observed above turn out to be systematic between different languages or language families, it would seem appropriate to try to establish which factors, syntactic and/or semantic, might determine what can count as a verbal complex in a given language.
The task of finding an appropriate unified characterization of trigger verbs in Romance appears to be enormously complicated because there is some variation in the particular predicates that can license clitic climbing and in the specific types of clitic climbing allowed among different dialects and across the various standard varieties in this language family.
There are a number of possible ways to tackle this variation. In the tradition of Transformational Grammar and its subsequent developments, the most popular type of solution has been to propose highly abstract and complex syntactic analyses which derive all these differences in terms of parametrized constraints on possible tree configurations. Within unification-based, lexically oriented frameworks such as HPSG or LFG, one might be tempted to account for this variation by simply listing in the lexicon those predicates that can act as triggers and those that don't. The differences observed across different languages and dialects would then be seen as a result of the different memberships in the class of 'restructuring predicates'. I would like to suggest, however, that such solutions, while doing the job more or less elegantly, miss some important and potentially useful generalizations.
While it might turn out to be impossible to avoid having to state some of the observed idiosyncracies in particular lexicons of particular languages, I believe that a general characterization of trigger verbs in the Romance languages is possible. After a brief discussion of some relevant data from contemporary Spanish and Catalan dialects, I will attempt to sketch out some possible ways in which such a characterization could be implemented within HPSG. This will serve as a basis to extend the basic analysis to other Romance varieties. Within this discussion I will comment on the results of an extensive investigation of the evolution of the clitic system in Spanish, carried out in my dissertation, which will help make sense out of some of the most obvious differences observed between the different Romance languages with respect to clitic climbing.
The analysis of "si", which I will propose, will distinguish between an argument and a non-argument usage of "si", capturing intuitions similar to those of Cinque (1988). In the reflexive/reciprocal and in the impersonal interpretation, the clitic "si" acts as an argument; it fills the relevant slot in the subcategorization requirements of the verb and it contributes to the semantics with agreement information. In the middle, ergative and inherent-reflexive interpretation, clitic "si" is instead a grammatical marker of the relevant verb form.
I will show how the introduction of a mechanism of unselective binding within the semantic organization of HPSG and the addition of a new type of index (`arbitrary') to the standard index hyerarchy of HPSG can uniformily accont for the syntax (masculine plural inflection) and semantics (genericity) of arbitrary null objects (1), generic subjects of `si'-constructions (2) and non controlled subjects of infinitival clauses (3):
It will be proven that the treatment provides the correct results also w.r.t. other romance languages, such as French, Spanish and Catalan. In particular, I will show how a parametric variation of the constraints associated to the `arbitrary' index is responsible for the fact that both plural and singular inflection can be associated to unespressd generic participants in Spanish and Catalan: