A memory-based explanation of antecedent-ellipsis mismatches: New insights from computational modeling

Abstract

An active question in psycholinguistics is whether or not the parser and grammar reflect distinct cognitive systems. Recent evidence for a distinct-systems view comes from cases of ungrammatical but acceptable antecedent-ellipsis mismatches (e.g., *Tom kicked Bill, and Matt was kicked by Tom too.). The finding that these mismatches show varying degrees of acceptability has been presented as evidence for the use of extra-grammatical parsing strategies that restructure a mismatched antecedent to satisfy the syntactic constraints on ellipsis (Arregui et al. 2006; Kim et al. 2011). In this paper, I argue that it is unnecessary to posit a special class of parser-specific rules to capture the observed profiles, and that acceptable mismatches do not reflect a parser-grammar misalignment. Rather, such effects are a natural consequence of a single structure-building system (i.e., the grammar) that relies on noisy, domain-general memory access mechanisms to retrieve an antecedent from memory. In Experiment 1, I confirm the acceptability profiles reported in previous work. Then in Experiment 2, as proof-of-concept, I show using an established computational model of memory retrieval that the observed acceptability profiles follow from independently motivated principles of working memory, without invoking multiple representational systems. These results contribute to a uniform memory-based account of acceptable ungrammaticalities for a wide range of dependencies.