Scalar timing control in spontaneous speech

Abstract

Phonological phrases were detected and segmented into syllables from the ongoing digitized oscillogram (low pass, 2 kHz) by a PDP-12 computer, utilizing a minimum interphrase silence of 200 msec, vowel onset and offset criteria of ±6 dB within a 20-msec window. Subjects spoke freely on three topics, 5 min per topic. Significant scalar timing (σ/μ=k) for phrase duration was found by stepwise regression analysis across the three topic samples. This timing control was determined to be neither a cumulative syllable count phenomenon, nor a general articulatory rate change (Allen, 1973). Rather, phrase variance within topic remained quite stable regardless of syllabic length, and the scalar changes between samples were unaccompanied by corresponding shifts of stress unit duration. Mean durations of initial, medial, and final stressed vowels in four of the six subjects were well fitted by Lindblom and Rapp's (1973) vowel-length contour equation, which presets vowel lengths in a phrase on the basis of the syllable count. The scalar property has often been found in studies of temporal estimation (Gibbon, 1977). Our results suggest that human phrase length is determined by estimation from a preplanned template, rather than by summing syllabic units of typical durations.