Quantitative Evaluation of Competing Syllable Parses

This paper develops computational tools for evaluating competing syllabic parses of a phonological string on the basis of temporal patterns in speech production data. This is done by constructing models linking syllable parses to patterns of coordination between articulatory events. Data simulated from different syllabic parses are evaluated against experimental data from American English and Moroccan Arabic, two languages claimed to parse similar strings of segments into different syllabic structures. Results implicate a tautosyllabic parse of initial consonant clusters in English and a heterosyllabic parse of initial clusters in Arabic, in accordance with theoretical work on the syllable structure of these languages. It is further demonstrated that the model can correctly diagnose syllable structure even when previously proposed phonetic heuristics for such structure do not clearly point to the correct diagnosis.