Tone Sandhi as Evidence for Segmenta- tion in Taiwanese

A major question in the study of language acquisition concerns how children learn to segment fluent speech into linguistic constituents such as words and syntactic phrases. In this paper, we show how production data can be used as indirect evidence for children's segmentation of fluent speech. Specifically, this paper examines how developmental changes in segmentation are reflected in Taiwanese tone sandhi (TTS).1 TTS describes a pattern of tone alternations that are sensitive to the boundaries of prosodic phrases built on syntactic constituents such as NP and VP (see Chen 1987, Lin 1994). Infants and young children just beginning to talk may be able to use TTS alternations in adult speech to learn where the constituent boundaries are (see Tsay, in press). Moreover, the errors that children make in producing tone sandhi alternations provide important evidence about how they are segmenting speech into constituents. TTS is thus an as-yet unexplored tool for researchers to learn more about the development of segmentation in production. Evidence about the units used in child language production came from the tone errors of two children (2;1-2;9) acquiring Taiwanese. We found that young children sometimes use smaller units than adults (e.g. splitting