Visual Word Recognition: Theories and Findings

chology and, therefore, a chapter on the topic must be selective. This chapter will first place the relevant issues in a historical context and then review the basic visual word recognition phenomena within the context of current models. It will then be argued that any successful model of visual word recognition needs to incorporate the assumption of “interactivity,” that is, that the various components of the visual word recognition system (i.e., orthographic, phonological, semantic) mutually activate and inhibit each other while a word is being processed (see also Van Orden & Kloos, this volume). (Hereafter, the term “word recognition” will be used as shorthand for the term “visual word recognition.”) What is “word recognition”? At least until the appearance of Seidenberg and McClelland’s (1989) connectionist model of reading, word recognition was typically thought of as the process of going from a printed letter string to the selection of a single item stored in lexical memory. Lexical memory, or the “lexicon,” is a mental dictionary containing entries for all the words a reader knows. Thus, word recognition was essentially synonymous with the terms “lexical access” or “lexical selection.” Such a definition, of course, assumes that words are represented as lexical entries in memory. Seidenberg and McClelland’s model explicitly denied the existence of such representations, arguing instead that representations were distributed across sets of simple subsymbolic processing units. To the extent that models of this sort have been successful, they have forced theorists to contemplate the possibility that some of the standard assumptions about the architecture of the word recognition system should be altered. What appears to be an equally important aspect of Seidenberg and McClelland’s (1989) model was that it contained a straightforward outline for how semantics should be integrated into the word recognition system. That is, semantic information was assumed to be represented no differently than other types of information (i.e., orthographic and phonological) and all of these mental representations were assumed to follow 3