Universität Bielefeld - Sonderforschungsbereich 360

A Model-Based Approach to Studying Language Processing:
What Eye Movements can Tell Us about Resolving Linguistic Indeterminacies

Julie C. Sedivy, University of Rochester
Much of the research in language comprehension over the last two decades has emphasized the incremental nature of processing language, showing evidence that linguistic information begins to be processed as soon as it becomes available in the speech stream. The incremental property of the human language processor, however, introduces a specific problem for the the processing system: that of temporary indetermincay in the speech stream, such that linguistic information must frequently be processed before there is sufficient information in the input to exclude all but the correct interpretation. A great deal of the literature in language processing has focused on the problem of temporary structural indeterminacy. Implicit in much of this work has been the assumption that the primary task of the human language processor is to recover the appropriate linguistic structure of the input.

In this talk, I will provide evidence from a novel eyetracking methodology that another form of indeterminacy arises that is a result of the task of mapping referential expressions to objects in a model, rather than the result of a structural ambiguity. This methodology uses a head-mounted eyetracking device which monitors eye movements of subjects in response to spoken experimental instructions. Work in this paradign has revealed that this mapping is highly incremental in nature. Evaluation of the objects in the model against the properties denoted in the referential expression begins immediately, and does not depend upon the complete construction of the referential phrase as a syntactic unit. Subjects typically program an eye movement to the target object within several hundred milliseconds of the word in the instruction that makes that object unique with respect to the relevant visual alternatives. This emphasis on the model-based aspect of language processing yields a rich opportunity for examining the resolution of indeterminacies that are referential, rather than structural in nature. I will describe one line of research within this paradigm that examines the effects of semantic and pragmatic information associated with contrastive focus on the resolution of reference.

In addition, I will discuss experiments using this methodology that indicate that the process of mapping expressions onto referents interacts in principled ways with the resolution of structural ambiguities.

Anke Weinberger, 1995-10-23