In the present study, I compare and contrast the utility of token frequency (including "string" frequency; Krug 1997) versus transitional probablility statistics in an attempt to predict the frequency-of-occurence of word boundary palatalization as it occurs in a large corpus of naturally-occurring data.  Transitional probablility is favored in this debate to the extent that it acts as a more reliable indicator of an underlying chunking process that contributes to the agglutination of lexical representations, reflected by the alternation of surface phonetic forms such as did you [dIdjë], versus did you [dIJë].  Furthermore, I demonstrate that the boundaries of constituent structures, especially as they theoretically enter into traditional, generative palatalization rules (cf. Chomsky and halle 1968, Vogel and Kenesei 1990:340), may be emergent from the non-random distribution of speech segments in discourse.  Statistical notions such as transitional probablity that accurately reflect this non-random distribution have been implicated in the segmentation of speech sounds by both children and adults (Saffran, Newport and Aslin 1996).

Back to the Conference Schedule