Vowel and sonorant lengthening as cues to phonological phrase boundaries

Abstract

Five sentences per speaker were selected from six individuals who participated in simulations of computer interactions. The utterances were distorted by spectral inversion and presented to five listeners who marked stressed syllables and the locations of phonological phrase boundaries, using only the prosodic cues remaining in the signal. Vowel and sonorant durations (with and without aspiration) were measured from spectrograms, and then declared stressed or unstressed based on the perceptions. Exploring the hypothesis that large increases in duration are syntactically determined [Klatt, Speech Communication Seminar Proceedings, Stockholm 277–289 (1974)], perceived boundary locations were compared with preceding segments which were 20%, 30%, and 40% above the median length for their categories. Of all the perceived phrase boundaries, those involving silences longer than 200 msec (pauses) were less reliably preceded by duration increases than perceived boundaries not at pauses. A possible reason is that the pause is enough of a cue to the boundary, and a duration increase is not needed. Of the lengthened segments not at perceived boundaries, some mark major syntactic boundaries, but most are at minor syntactic breaks, notably between modifiers and nouns, and after prepositions. The results suggest that speaker differences and style variations may be important. [Research supported by the Advanced Research Projects Agency, Department of Defense, and the Phonetics Laboratory, University of Michigan.]