Although phrasal dependencies, grammatical agreement and other elements of sentence structure are inherently local in their formal character, from a production perspective, these dependencies commonly require tracking across unbounded amounts of lexical and structural planning. This talk explores the mechanisms that provide sentence planning and production with a reliable system for recording the elements of (often as yet incomplete) dependencies. Specifically, I present cross-linguistic experimental evidence focusing on the components of sentence production that calculate and execute Subject-Verb agreement.
Although seemingly flawless, the agreement generation process systematically gives rise to one category of errorÑthe agreement attraction error (e.g., "The response from all the actors were terrific"). The specific contexts which elicit agreement attraction errors provide important clues to how grammatical agreement is calculated and tracked. One current theory explains agreement attraction by positing a feature percolation mechanism analogous to the one found in some Syntactic Theories. On this account, attraction occurs when subsequent copying (or checking) procedures accommodate features of a local/non-root noun that have erroneously migrated to the root node of a subject phrase. However, findings from a number of our recent studies in English and Slovak indicate that this formally transparent interpretation of grammatical performance fails to capture the important facts about agreement attraction and, by extension, about the mechanisms that underlie normal fluent production.
Synthesizing our most recent findings with previously reported work, I develop an alternative model of sentence production which captures the broader range of facts about agreement processing by embedding the representational underpinnings of grammatical agreement (and sentence planning in general) within an activation-based, computational architecture. .