Testing the interactive two-step model of lexical access: How we do it and why

Abstract

Computational models of aphasia must, first, characterize the actions of the unimpaired system, and then explain how that system is damaged in aphasia. Our model of lexical deficits in aphasic speakers is based on the interactive two-step theory, an account of normal lexical access in production that associates the process with two distinct steps, word retrieval and phonological retrieval. Each step is achieved by the interactive, or bidirectional, spread of activation through a layered network of units for semantic features, words, and phonemes (Fig. 1). Deficits are created by altering parameters that affect activation. In our first model of this type, the weight-decay model (Dell, Schwartz, Martin, Saffran, & Gagnon, 1997), lesions corresponded to global decreases in the weight of connections among all units in the network (parameter w) and global increases in the rate with which activation decays (parameter d). The more recent semantic-phonological model (Foygel & Dell, 2000) also had two lesionable parameters, s, the weight of the connections between semantic and lexical units, and p, the weight of the connections between the phonological and lexical units. Developing and testing these models involves four stages: specifying the normal model, defining the possible lesions, assigning parameters to patients based on their errors in a picture naming task, and using those parameters to test predictions about other aspects of patient behavior.