Phonetic Interpretation Papers in Laboratory Phonology VI: On the factorability of phonological units in speech perception

At least in English, the perception of syllables and words can be largely predicted from the perception of smaller (e.g., segmentor phoneme-sized) symbolic units. Experiments involving the intelligibility of natural speech in noise provide one important source of evidence for the factorability of syllables into segments. Results reviewed by Allen (1994 a, b) show that the correct identification of nonsense CVC syllables in noise can be extremely well predicted from the marginal correct identification of their constituent phonemes. Simulation experiments by Nearey (in press) suggest that this result (as well as the systematic increase in the correct identification of real words over nonsense) can be achieved when stimulus patterns for syllables are simple functions of their phoneme parts. However, stimulus patterns for syllables that include idiosyncratic properties that are not predictable from segmental components apparently cannot produce such results. Several perception experiments with synthetic speech also support the factorability-of-syllables hypothesis. These experiments involve the simultaneous perception of consonants and vowels. They suggest that perception of syllabic wholes can be largely predicted from the perception of their segmental parts. Evidence supporting this hypothesis is reviewed and some remaining problems (where factorability appears to break down) are discussed. The question of whether a similar degree of factorability extends to smaller, feature-sized units and how any such elements may relate to universal distinctive features is briefly explored. Nearey On factorability... OnFactMarch2k1R1Mac.doc 5/17/0112:10 PM 2 X.