Dual processing versus response-limitation accounts of categorical perception: A response to Macmillan, Kaplan, and Creelman

Abstract

At the last meeting of the Society, Macmillan, Kaplan, and Creelman argued that measures of discrimination typically obtained in speech perception experiments employing single-interval absolute-identification tasks do not reveal the “true sensitivity” of the observer because only a very small number of response categories are permitted and identification into these putative phonological categories is often without error. In the present paper we present the results of two experiments, one employing synthetic stop consonants and the other employing synthetic steady-state vowels, in which observers were permitted to use a relatively large number of response categories in an absolute-identification (AI) procedure similar to that used by Braida and Durlach. Discrimination performance in the identification task was then compared to discrimination obtained in the traditional ABX format in order to assess the correspondence between the two measures. The results revealed systematic differences in discrimination between consonants and vowels and between AI and ABX procedures. While improved discrimination performance was observed in the AI task for vowels by permitting a larger number of response categories in identification, the same effects were not observed for stop consonants. Our findings suggest that the account of categorical perception offered by Macmillan et al., in terms of response limitation in identification, is not a sufficient and viable alternative to previous accounts of categorical perception in terms of dual coding models which postulate separate levels of coding information in speech perception. [Supported by grants from NIH and NIMH.]