Description:

This dissertation explores how items encountered in the comprehension of language are stored in memory and subsequently accessed. Processing and comprehending language frequently requires the retrieval of items in memory so that a current linguistic element can be assigned an interpretation. For example, in a sentence such as "Miles loved his father", the pronoun "his" has no intrinsic value but instead receives its interpretation from the noun "Miles". Another similar example concerns ellipsis, where phonologically silent material must select an antecedent from memory to be successfully interpreted.

In this presentation, I examine dependent elements such as pronouns and ellipsis in light of popular cue-based models of memory retrieval. While these models can accommodate a certain amount of syntactic information, maintaining large hierarchical structures and the relation between items in such a structure is problematic. To probe whether such structures and relations are available to the parser during dependency resolution, I employ a series of both offline tasks and eye tracking tasks. The first set of experiments uses the concept of 'parallelism' to explore whether the relation created by a pronoun and its antecedent is stored in memory, and whether such a representation can affect subsequent processing. The second set of experiments looks at sluicing type ellipsis, and challenges the claims of cue-based models by providing observations that antecedent size and complexity have an effect on processing. Finally, the last set of experiments employs verb phrase ellipsis to examine cases in which the antecedent and the interpretation of ellipsis is non-isomorphic. The overall results of these experiments provide evidence that highly detailed syntactic structure is both stored in memory and re-examined during dependency resolution, and that the content of retrieval is remarkably faithful.