Seth Kulick (UPenn):
Complex Predicates in Tree Adjoining Grammar

Donnerstag, 10.00 Uhr

In this paper I discuss complex predicate constructions, focusing on Romance,in the framework of Tree Adjoining Grammar (TAG) (Frank and Kroch 1995) and argue that the resulting analysis has several desirable properties. TAG is a formalism in which structural representations are built out of pieces of phrase structure, called elementary trees, which are taken as atomic by the formal system. Inter-clausal movement is captured by using the adjunction operation to insert one tree within another, thus allowing components of a tree to be "stretched" apart. It is an attractive aspect of TAG that well-known constraints on locality, such as subjacency, are derivative from the properties of adjunction, and do not need to be stated as stipulations. As a result, the substantive theory of syntax can be limited to conditions on the well-formedness of elementary trees.

The constrained nature of TAG rules out an analysis of "restructuring" effects in Romance, such as clitic climbing, in terms of movement from one clause to another. Instead, we must have an account in which the restructuring verbs are "defective", and so are small enough that they can adjoin into a clause where a non-defective verb could not. For example, we derive an Italian sentence such as (1) by having a tree for vuole adjoin into a tree for Mario lo leggere, thus stretching Mario lo away from leggere, with no actual "movement" of the clitic. In contrast to movement accounts of restructuring, which often rely on the complement missing some functional projections (e.g., \cite{Martins}), the complement in the TAG analysis has a full projection in its own elementary tree, and it is instead the higher verb which is defective.

Mario lo vuole leggere

Mario wants to read it

Essentially, for these "restructuring" cases, the biclausal/ monoclausal conflict is resolved by allowing the restructuring verbs to adjoin into another verb's elementary tree. This type of adjunction, while formally possible, is usually prohibited because it violates the constraint on a TAG derivation that syntactic composition must be monotonic in its effect on semantic interpretation. But it is exactly in the case of these restructuring verbs that such adjunction is allowed, because of their well-known properties of being semantically weak (\cite{Napoli} and others). The distinction in TAG between the record of the derivational history (used for semantic interpretation) and the corresponding derived phrase structure tree allows for an interesting approach to the syntax/ semantics non-isomorphism.

The TAG analysis has certain desirable results. Two examples are: (1) It predicts, with no stipulations necessary, that restructuring will be impossible with object-control verbs, since the subject of the final sentence is simply the subject of the elementary tree projected from the lower verb. I propose that that all apparent cases of object-control restructuring verbs, such as the {\it permitir} class in Spanish, should be treated as causatives, based both on the semantic properties of these verbs, and the fact that they classify with uncontroversial causatives and against restructuring verbs in crucial ways, such as the placement of the reflexive clitic se (Suņer 1980). (2) The various facts showing that the main predicate in these constructions is the embedded verb, such as the interpretation of plural null pronominals in Catalan (Picallo 1990), follow directly from this analysis.

This type of analysis in TAG, however, is not possible for causative constructions. Instead, a variant of TAG, multi-component TAG, is used for these cases. I suggest that this allows a way to differentiate between the cases of "restructuring" in which the subject of the embedded clause is or is not coreferential with the subject of the matrix clause. I also discuss the locality properties of clitic movement with embedded causatives, a crucial issue for TAG.

I will discuss some issues that are potentially problematic for this analysis (e.g., Italian auxiliary selection), and also how the TAG analysis compares and contrasts with certain aspects of analyses in other frameworks, such as argument composition in HPSG (Monachesi 1995).