Articulatory Motor Events as a Function of Speaking Rate and Stress

Two basic types of explanation have been proposed for the changes in segmental timing that occur when speakers change rate or stress of component syllables. One view is that the segmental "commands" for syllables spoken quickly and for unstressed syllables show more extensive temporal overlap than the same syllables spoken more slowly or with greater syllabic stress. An alternative view is that the temporal relations among articulations remain constant over changes in speaking rate and stress, but that the individual gestures themselves vary. Experiment 1 explored the temporal relations among electromyographic measures of articulatory events, and the pattern of changes in individual muscle actions, over suprasegmental variations in syllable stress and speaking rate. Large variations were found in the magnitude and duration of activity in each muscle; variations accompanying speaking rate change were not equivalent to the variations accompanying a change in stress. The electromyographic activity underlying lip movements for bilabial stop consonants (orbicularis oris) and tongue fronting for the vowels Iii and le/ (genioglossus) appeared to maintain a tight timing pattern. In a second experiment, X-ray microbeam data were collected for the same types of utterances used in the first experiment. Kinematic patterns, like EMG patterns, showed that temporal relations between tongue and lip movements were preserved over changes in speaking rate and syllable stress. Investigations of speech production have often focused on a search for invariant units that correlate with aspects of a speaker/hearer's linguistic competence. Many of these studies share an assumption about linguistic units: namely, that they are discrete, static, and context-invariant entities, selected and ordered prior to their execution by peripheral motor mechanisms. Most experiments have consisted of a search for discrete stretches in the acoustic or physiological output in the hope that they might correlate with linguistic units. However, such studies have met with little success, whether +Also Cornell University Medical College. ++Also The Graduate School, City University of New York. +++Also The University of Connecticut, Storrs. Acknowledgment. We would like to thank Carol Fowler, Michael StuddertKennedy, and Robert Verbrugge for comments. Some of the work reported here is based on a doctoral dissertation by the first author, presented to the University of Connecticut. The X-ray microbeam data were collected at the Insti tute of Logopedics and Phoniatrics, and analyzed at Bell Telephone Laboratories. We are grateful to Dr. Masayuki Sawashima and Dr. Osamu Fujimura. The work was supported by NINCDS grants NS-13617 and NS-13870, and BRS grant RR-05596 to Haskins Laboratories. [HASKINS LABORATORIES: Status Report on Speech Research SR-65 (1981)J 33 looking for invariant units in the acoustic signal (Liberman, Cooper, Shankweiler, & Studdert-Kennedy, 1967; but see stevens, 1973), patterns of muscle activity (Harris, Lysaught, & Schvey, 1965; MacNeilage & DeClerk, 1969), articulatory movements (MacNeilage, 1970), or vocal tract area functions. The repeated failure to find invariant correlates of abstract linguis~ic units has promoted the claim that abstract representations used to describe linguistic competence are obscured when translated into linguistic performance, because the latter are subject to the physical constraints of human speech to which the former are indifferent (cf. Ohman, 1972). The foregoing conception of linguistic units as abstract and discrete is inherent in those current models of speech production which assume that articulatory control of suprasegmental changes is independent of segmental articulation. Articulatory control over variations in speaking rate and syllable stress, for example, is considered as " •.• the consequence of a timing pattern imposed on a group of (invariant) phoneme commands" (Shaffer, 1976, p. 387; parentheses his). Similarly, Lindblom (1963) suggested that each phoneme has an invariant "program" that is unaffected by changes in lexical stress and speaking rate (tempo).1 According to Lindblom, when successive programs are executed, their temporal overlap results in coarticulation between segments. Thus, when a vowel coarticulates with a following consonant, it is because the consonant program begins before the vowel program is finished (see also Stevens & House, 1963). When speaking rate increases or stress decreases, the command for a new segment arrives at the articulators before the preceding segment is fully realized. As a consequence, there is temporal shortening and articulatory undershoot, both of which characterize unstressed syllables and fast speaking rates (see also Kozhevnikov & Chistovich, 1965). In such models, therefore, increases in speaking rate and decreases in syllable stress are accomplished with comparable strategies and hence have similar acoustic consequences. They predict that the "commands" for some aspects of articulation of a given phoneme stand in a fixed relation to commands for other aspects of the same phoneme, but that the relative temporal alignment of control signals for successive segments, and their kinematic realizations, vary with stress and speaking rate. The models discussed above suggest that changes in speaking rate and syllable stress are both characterized by invariant segments with variable temporal relations between them. One prediction of this view is that the relation between target formant frequency and duration is fixed; that is, when the duration of a vowel shortens, it will undershoot the articulatory "target," resul ting in more centralized formant frequencies than occur with longer vowel durations. However, Harris (1978) performed a spectrographic analysis of a small set of nonsense utterances produced at two speaking rates and with two levels of stress, and found that changes in vowel formant frequencies were not fixed in relation to changes in vowel duration. Her results suggest that extant models for suprasegmental changes cannot be supported at an acoustic level. A similar conclusion follows from a small body of electromyographic (EMG) data showing that segmental articulation varies considerably with speaking rate (Gay & Hirose, 1973; Gay & Ushijima, 1974; Gay, Ushijima, Hirose, & Cooper, 1974) and syllable stress (Harris, 1971, 1973; Harris, Gay, Sholes, & Lieberman, 1968; Sussman & MacNeilage, 1978). However, these studies have not