Is there still a TRACE of trace?

Abstract

According to the TRACE model [McClelland and Elman, Cogn. Psychol. 18, 1–86 (1986)], speech recognition is an interactive activation process involving the integrated use of top-down (lexical) and bottom-up (acoustic) information. Although it is widely accepted that there are lexical influences on speech perception, there has been a disagreement over their exact nature. Two contested predictions of TRACE are that (a) lexical influences should delay or inhibit recognition of phonemes not consistent with lexical information and (b) a lexical influence on the identification of one phoneme can trigger compensation for co-articulation, affecting the identification of other phonemes. Others [Norris, McQueen, and Cutler, BBS 23, 299–370 (2000)] have argued that the predicted effects do not occur, taking this to support an alternative to the TRACE model in which lexical influences do not affect perception, but only a post-perceptual identification process. We re-examine the evidence on these points along with the recent finding that lexical information may lead to a lasting adjustment of category boundaries [McQueen, Norris, and Cutler, Psychonomics Abstract 255 (2001)]. Our analysis indicates that the existing evidence is completely consistent with TRACE, and we suggest additional research that will be necessary to resolve unanswered questions.