Accounting for the fine structure of syntactic working memory: Similarity-based interference as a unifying principle

Abstract

A promising approach to more refined models consistent with the Caplan & Waters hypothesis is based on similarity-based interference, a general principle that applies across working memory domains. This may explain both the fine details of syntactic working memory phenomena and the gross fractionation for which Caplan & Waters have found evidence. Detailed models of syntactic processing that embody similarity-based interference fare well cross-linguistically. Caplan & Waters (CW 1996b; 1998). C&W start their discussion of memory demands in sentence comprehension with the notoriously difficult English double center-embedded relative clause (their sentence [1]) and go on to cite a few of the computational models and linguistic metrics that have been developed to account for the problem with such embeddings. What they did not mention is how surprisingly difficult it has been to produce models and metrics that are empirically adequate. Though it is true that there is a “remarkable degree of similarity in the measurements” that some of the models produce (sect. 1), the vast majority of these measurements do not fare well when considered against a broad range of embeddings cross-linguistically. Furthermore, there has been little independent psychological motivation for the proposed memory structures (e.g., stacks, lookahead buffers) and their associated limitations (see Lewis 1996 for a review). As an example of the kind of empirical hurdle faced by any theory of syntactic working memory, consider a fact established by Cowper (1976) and Gibson (1991) in their seminal work: a metric based on the amount of center-embedding does not account for many difficulty contrasts in English and other languages. Consider sentences (1a) and (1b): 1a. That the food that John ordered tasted good pleased him (Cowper