Abstract:

A gating experiment addressed the question of how early in the process of recognizing spoken Japanese words pitch--accent information may be exploited. 24 pairs of Japanese words such as nimotsu/nimono, beginning with the same bimoraic CVCV sequence but with the accent pattern of this initial CVCV being HL in one word and LH in the other, were presented, in increasingly large fragments, to 36 native speakers of Tokyo Japanese. After presentation of each fragment, which was incremented in each case by one phoneme transition from the previous fragment, listeners recorded a guess regarding the word's identity and a confidence rating for that guess. The results showed that the accent patterns of the word guesses corresponded to the accent patterns of the actually spoken words with a probability significantly above chance from the second fragment onwards, i.e., from the middle of the vowel in the first mora of the word. Accent correspondence averaged 79.6% at this point, rising to 89% by the fourth fragment (vowel of second mora). This demonstrates that Japanese listeners can exploit pitch--accent information effectively at an early stage in the presentation of a word, and use it to constrain selection of lexical candidates.