Ambiguity and connectionist networks: still settling into a solution - comment on Joordens and Besner (1994)

Empirical results (e.g., see G. Kellas, F. R. Ferraro, & G. B. Simpson, 1988; M. L. Millis & S. B. Button, 1989) suggest that readers perform lexical decisions faster for words with many meanings than for words with only one meaning. S. Joordens and D. Besner (1994) attempted to simulate this effect with a connectionist model but found that their network only produced an ambiguity advantage when its error rate was unacceptably high. In this commentary 3 issues relevant to an understanding of the implications of Joordens and Besner's findings are discussed: the reliability of the empirical data, the manner in which readers (and networks) make lexical decisions, and the effects of certain algorithmic and architectural factors on network performance.