Julie Sedivy - US grants

A great deal of work in sentence processing over the years has been preoccupied with the question of whether contextual information can influence early parsing decisions. There is by now considerable evidence that suggests that at least some kinds of information from the discourse context do have such effects. However, the question of how these effects occur and where they come from has remained largely ignored. The possible mechanisms for discourse context effects range from very direct ones where classes of linguistic expressions trigger certain discourse representations, to ones involving subtle inferential mechanisms that evaluate the likely discourse functions of alternative linguistic expressions.

The studies in this project focus on the discourse properties of modified definite noun phrases. Work in sentence processing has shown that the resolution of ambiguities in which one of the possible readings involves noun modification is affected by the availability of a discourse model in which the modificational phrase serves to distinguish between two possible referents.

A central question is whether such discourse effects found with modifiers reflect a general, conventionalized property of modification, or whether they are more aptly characterized as a subtle system of expectations regarding typical usages. A series of studies is proposed to investigate the hypothesis that a typical default expression exists for neutral (i.e. non-contrastive) contexts, and that the use of a more marked or informative expression signals a contrastive function in the discourse that has immediate processing consequences. Data from elicited production tasks will be collected and directly compared with on-line comprehension results from eye movement experiments that allow for the monitoring of subjects' eye movements to a visual array in response to spoken linguistic stimuli.

It is suggested here that the discourse effects previously observed with modified NPs are not limited to modified expressions, but extend to other cases involving deviation from a default expression (e.g. the distinction between basic-level and subordinate level expressions as identified in the categorization literature). Experiments testing this hypothesis will be conducted both by means of production and eye monitoring experiments, as well as by building on work in more traditional methodologies such as reading time studies of temporarily ambiguous sentences.

DESCRIPTION (provided by applicant): A great deal of work in sentence processing over the years has dealt with the question of whether contextual information can guide language processing. There is by now considerable evidence that suggests that some kinds of information from the discourse context have immediate effects. However, the question of how these effects occur has remained largely ignored. The studies in this proposal focus on the discourse properties of modified definite noun phrases. Work in sentence processing has shown that the resolution of ambiguities in which one of the possible readings involves noun modification is affected by the availability of a discourse model in which the modificational phrase serves to distinguish between two possible referents. A central question is whether such discourse effects found with modifiers reflect a general, conventionalized property of modification, or whether they are more aptly characterized as a more subtle system based on expectations regarding typical usages. A series of studies is proposed to investigate the hypothesis that a typical default expression exists for neutral (i.e., non-contrastive) contexts, and that the use of a more informative expression signals a contrastive function in the discourse, with immediate processing consequences. Data will come from elicited production tasks, on-line comprehension experiments which monitor subjects' eye movements to a visual array in response to spoken linguistic stimuli, traditional reading time studies, and prosodic analyses in a read-aloud task. The current proposal represents a significant departure from existing work in two salient ways: First, it attempts to provide a detailed investigation into the nature of the referential effects, using a methodology that is especially well-suited for studying referential aspects of language. Second, whereas previous findings have been couched almost exclusively in terms of the mechanisms of sentence processing, the current proposal seeks to integrate experiments from on-line language processing and language production. Results of this project may be useful in developing models for language disorders, for the development of pedagogical tools, and for progress in artificial intelligence.

In order to successfully communicate, speakers and hearers must use more than their knowledge of a common language. Often, understanding a speaker's intended meaning requires "reading between the lines" thereby enriching the linguistic meaning of what the speaker has said. For example, if a speaker says "I like some of my relatives," the hearer will likely understand her to have implied that she does not like all of them, even though her utterance is, strictly speaking, true if she does in fact like them. Previous work has shown that similarly enriched meanings can be computed extremely rapidly by hearers. This project examines in detail how speakers (or readers) arrive at such enriched meanings. What kinds of expectations about the speaker's choice of linguistic expressions does the hearer rely on? Does the hearer take into account the speaker's communicative goal or communicative abilities? Are there individual differences in the extent to which people can integrate more socially-based expectations about typical communicative behavior with linguistic meanings? And are there aspects of linguistic structure or context that make enriched meanings more accessible to the hearer/reader?

This project will address some currently heated debates in theoretical linguistics, philosophy and cognitive psychology about the nature of enriched meanings and their relationship to purely linguistic meanings. It will provide the first study of individual differences in this domain, laying the groundwork for future studies with atypical populations (especially autistic individuals). The results are also likely of interest to computational linguists.