Defining the Knowledge Units of a Synthetic Language : Comment on Vokey and Brooks (1992)

Vokey and Brooks (1992) reported a set of experiments intended to demonstrate that judgments of grammaticality are determined by two characteristics of the test items: their similarity with a specific study item and their conformity with an abstract representation of the generative grammar. I argue that both effects may be encompassed within a unified account, which requires neither a specific-item retrieval process nor an abstractive capacity. My basic assumption is that the primary knowledge units are not whole strings of letters, as postulated in models relying on specific similarity or abstraction, but rather fragments of 2 or 3 letters. Partial memorization of these small units provides a convenient account of the whole pattern of Vokey and Brooks's findings because study items have more units in common with similar than with dissimilar test items, and likewise with grammatical than with ungrammatical ones.