The process of lexical decision: More words about a parallel distributed processing model.

Abstract

Central to Seidenberg and McClelland's (1989) parallel distributed processing (PDP) model's account of lexical decision is the assumption that performance is determined by the amount of overlap in the distribution of the orthographic error scores between words and nonwords. The present experiment demonstrates lexical decision performance that is independent of the distribution of orthographic error scores between words and nonwords. Furthermore, the orthographic error scores from the model capture no variance in the reaction time data to words, even in a condition in which the orthographic overlap between words and nonwords is minimal. Other issues are discussed.Until very recently there was a remarkable consensus in the visual word recognition literature regarding the nature of the underlying mental representation of words. It was widely assumed that words were stored as lexical entries in a mental lexicon, where each word was represented by an individual node (e.g.,

Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

“…Here we discuss how the remaining three issues-acquired phonological dyslexia, developmental dyslexia, and lexical decision-may be accounted for in light of these findings. We also consider three other empirical findings that have been interpreted as providing evidence against the current approach-pseudohomophone effects (Buchanan & Besner, 1993;Fera & Besner, 1992;McCann & Besner, 1987;Pugh, Rexer, & Katz, 1994), stimulus blocking effects (Baluch & Besner, 1991;Coltheart & Rastle, 1994;Monsell et al, 1992), and the recent finding that naming latencies for exception words are influenced by the position of the exceptional correspondence (Coltheart & Rastle, 1994). Acquired Phonological Dyslexia.…”

“…Two other, somewhat overlapping sets of empirical findings have been viewed as problematic for the current approach: pseudohomophone effects (Buchanan & Besner, 1993;Fera & Besner, 1992;McCann & Besner, 1987;Pugh et al, 1994) and blocking effects (Baluch & Besner, 1991;Coltheart & Rastle, 1994;Monsell et al, 1992). The first set involves demonstrations that, under a variety conditions, pseudohomophones (i.e., nonwords with pronunciations that match that of a word; e.g., BRANE) are processed differently than orthographically-matched nonpseudohomophonic nonwords (e.g., FRANE).…”

“…The final of Coltheart et al's (1993) objections to the SM89 model concerns its ability to perform lexical decisions. While SM89 establish that, under some stimulus conditions, the model can discriminate words from nonwords on the basis of a measure of its accuracy in regenerating the orthographic input, Besner and colleagues (Besner et al, 1990;Fera & Besner, 1992) have demonstrated that its accuracy in doing so is worse than that of human subjects in many conditions. Coltheart et al (1993) mistakenly claim that the SM89 orthographic error scores yield a false-positive rate of over 80% on Waters and Seidenberg's (1985) nonwords when word error rates are equated with subjects' at 6.1%-in fact, these numbers result from using phonological error scores (Besner et al, 1990), which SM89 do not employ (although they do suggest that learning phonological attractors for words might help).…”

“…Again, however, the model is not as accurate at lexical decision under some conditions as are normal subjects (Besner et al, 1990;Fera & Besner, 1992).…”

Semantic and phonological influences on the processing of words and pseudohomophones

“…As evidence that phonology does play a role in the lexical decision task, researchers have shown that nonword response latencies are longer to pseudohomophones (e.g., nale) than to orthographically matched pseudowords (e.g., nalp; Coltheart, Davelaar, Jonasson, & Besner, 1977;Fera & Besner, 1992;McCann, Besner, & Davelaar, 1988;Rubenstein, Lewis, & Rubenstein, 1971;Seidenberg, Petersen, MacDonald, & Plaut, 1996), implying that the pseudohomophone accesses the representation of its base word (e.g., nail ), and that this base word representation slows the processing of the pseudohomophone. Because the pseudohomophone effect has become central to models of word recognition (Ziegler, Jacobs, & Klüppel, 2001), it is critical that we understand fully the processes underlying the effect.…”