ABSTRACT
A connectionist approach to processing in quasi-regular domains, as exemplified by English word reading, is developed. Networks using appropriately structured orthographic and phonological representations were trained to read both regular and exception words, and yet were also able to read pronounceable nonwords as well as skilled readers. The simulation was put forward in support of a more general framework for lexical processing in which orthographic, phonological, and semantic information interact in gradually settling on the best representations for a given input. Lexical tasks involve transformations among these representations — for example, oral reading requires the orthographic pattern for a word to generate the appropriate phonological pattern. The SM89 framework’s second major break with tradition concerns the degree of uniformity in the mechanism(s) by which orthographic, phonological, and semantic representations interact. The relevant assumptions about the phonological representations are that they are segmental and that they are strongly constrained by phonotactics.
