27 - Speech segmentation by Japanese listeners: its language-specificity and language-universality

Summary

Introduction

Spoken-word recognition, or how human listeners comprehend the meaning of speech, is one of the main topics of psycholinguistics. More specifically, the study of spoken-word recognition explores how human listeners manage to retrieve the individual words which make up continuous spoken utterances. Speakers do not utter a segregated sequence of isolated words; they run words together without an intervening pause, so that utterances reach listeners as a continuous stream. Given that all human language users use speech as a primary means to communicate, and that no human listeners with normal hearing capacity fail to extract the meaning of utterances, the human speech recognition system must be very efficient at detecting words in continuous speech, i.e. at speech segmentation.

During the past thirty years or so, the study of speech segmentation has progressed remarkably. Many types of phonetic and phonological information have been reported to be segmentation cues. For example, aspiration of word initial stops in English (Lehiste, 1960; Nakatani & Dukes, 1977), duration of segments (Lehiste, 1972; Klatt, 1974, 1975), phonotactics (McQueen, 1998), and silence (Norris et al., 1997) all act as segmentation cues. Although each of these individual cues certainly provides us with salient segmentation information, they will not always be present, so this is not the whole picture of the segmentation process. The ultimate goal of psycholinguistics is to reveal both universal and language-specific aspects of language processing, and the study of speech segmentation is no exception.