Parsing and acquisition: Evidence from Serbo-Croatian.

Abstract

The goal of the thesis was to examine the processing of different types of filler-gap dependencies in Serbo-Croatian. The Active Filler Strategy (Frazier 1987) and its extension, the Minimal Chain Principle (de Vincenzi 1991), formulated as ambiguity resolution strategies in the processing of empty categories, predict that subject extractions are easier than the object ones. Other strategies, such as the Syntactic Prediction Locality Theory (Gibson 1998), although formulated as a more general theory of sentence processing, make very similar predictions about the processing of wh-extractions. The thesis consists of different experiments in Serbo-Croatian, in which a number of predictions based on the AFS, the MCP and potentially the SPLT were tested. Three on-line reading time experiments were carried out: the scarmbling, koji and Sto experiments. In the scrambling experiment the processing of different word orders was tested. In the koji ("who/which") relative clause experiment the processing of filler-gap dependencies in relative clauses was tested. The third experiment, the Sto experiment, presents an innovative contribution to the study of the filler-gap processing. As object binders always precede the subject binders, object Sto relatives are predicted to be easier to process than the subject ones. The results from the scrambling experiment indicate that there is a strong preference to interpret an initial ambiguous NP as the subject. The results from the koji experiment show that the parser prefers to postulate the gap in the first grammatically available position, making subject relatives easier to process than the object ones. In Sto relatives, the locality asymmetry, based on a structural difference, is reflected in processing: object Sto relatives are easier to process than subject relatives. In sum, the results from the on-line experiments indicate that the parser searches actively for the gap/binder in the processing of filler-gap dependencies in Serbo-Croatian. Additionally, the adult study has been extended to testing children's comprehension of different types of relative clauses. Results from act-out tasks and an elicited production task show a trend towards preference for an object interpretation in Sto relatives, as well as a subject interpretation in koji relatives. Results from both adult and child experiments are examined within a model of parsing in which syntactic and semantic information are factored in at different levels.