Spoken word recognition including prosodic processing in Japanese

While conventional spoken word recognition models have focused on how identification of sequential segments contributes to lexical access, recent empirical studies have also shown the crucial role of lexical prosody in spoken word recognition. In Japanese, lexical pitch accent that contrasts homophonic words is utilized for narrowing down candidates incrementally, helps resolution of acoustic ambiguity in segments as lexical bias, and nevertheless is not utilized when it is inappropriately pronounced. These flexible behaviors of lexical prosody in spoken word recognition in Japanese are realized by parallelism of two linguistic cues, i.e., segments and lexical prosody, in speech signals. The two linguistic cues are processed simultaneously and contribute to lexical access interactively, which makes the spoken word recognition a more robust process. We propose that this is theoretically accounted for a modified version of the TRACE model of spoken word recognition (McClelland & Elman, 1986) including lexical prosodic processing, which is tuned specifically for Japanese lexical pitch accent. Evidence from Japanese emphasizes how lexical prosody is important in spoken word recognition.