Anticipatory Coarticulation Effects Research Articles

Coarticulatory dynamics in Greek disyllables produced by young adults with and without hearing loss

ABSTRACT Hearing loss affects both speech perception and production with detrimental effects on various speech characteristics including coarticulatory dynamics. The aim of the present study is to explore consonant-to-vowel (C-to-V) and vowel-to-vowel (V-to-V) coarticulation in magnitude, direction and temporal extent in the speech of young adult male and female speakers of Greek with normal hearing (NH) and hearing impairment (HI). Nine intelligible speakers with profound HI, using conventional hearing aids, and five speakers with NH produced /pV1CV2/ disyllables, with the point vowels /i, a, u/ and the consonants /p, t, s/, stressed either on the first or the second syllable. Formant frequencies F1 and F2 were measured in order to examine C-to-V effects at vowel midpoint and V-to-V effects at vowel onset, midpoint and offset. The acoustic and statistical analyses revealed similarities but also significant differences regarding coarticulatory patterns of the two groups. Interestingly, prevalence of anticipatory coarticulation effects in alveolar contexts was observed for speakers with HI. Findings are interpreted on account of possible differences in articulation strategies between the two groups and with reference to current coarticulatory models.

Misparsing coarticulation

Considerable evidence shows that listeners often successfully compensate for coarticulation, and parse the speech signal's acoustic properties into their articulatory sources. Our experiments show pervasive misparsing of the acoustic effects of anticipatory coarticulation. Listeners categorized more of a front-back vowel continuum as back before [p] than [t]. A following [p] causes F2 and F3 to fall at the end of preceding vowels, and these listeners treated these falls as information that the vowel was back because F2 and F3 are also lower in back vowels. In experiment 1, misparsing was more extensive in a lax front-back continuum [ɛ-ْ] than in a tense one [e-o], but the lax vowels were also shorter. In experiment 2, the durations of lax and tense vowels were equalized, and the steady-state:transition ratios were varied from the naturally occurring 70:30 to 50:50 and 30:70. Listeners still misparsed consonantal formant transitions more for lax than tense vowels, likely because the acoustic differences between [ɛ] and [ْ] are smaller than those between [e] and [o]. They also misparsed more as transitions lengthened relative to the steady-state, which suggests that they did not adjust for differences in speaking rate that would affect the steady-state:transition ratios.

Immediate effects of anticipatory coarticulation in spoken-word recognition

Two visual-world experiments examined listeners' use of pre word-onset anticipatory coarticulation in spoken-word recognition. Experiment 1 established the shortest lag with which information in the speech signal influences eye-movement control, using stimuli such as "The … ladder is the target". With a neutral token of the definite article preceding the target word, saccades to the referent were not more likely than saccades to an unrelated distractor until 200-240 ms after the onset of the target word. In Experiment 2, utterances contained definite articles which contained natural anticipatory coarticulation pertaining to the onset of the target word (" The ladder … is the target"). A simple Gaussian classifier was able to predict the initial sound of the upcoming target word from formant information from the first few pitch periods of the article's vowel. With these stimuli, effects of speech on eye-movement control began about 70 ms earlier than in Experiment 1, suggesting rapid use of anticipatory coarticulation. The results are interpreted as support for "data explanation" approaches to spoken-word recognition. Methodological implications for visual-world studies are also discussed.

Temporal organization of interarticulator muscle activity in American‐English monosyllables

The activity patterns of various muscles were investigated during speech production for the purpose of determining the muscular coordination of the labial, lingual, and mandibular systems in relation to anticipatory coarticulation effects. Intramuscular wire electrodes were placed in three muscles of the lips (upper and lower lip), three muscles of the tongue, and one muscle of the jaw. The activity of these muscles was recorded along with the acoustic signal from one adult American‐English speaking subject. A number of CVC syllables, where C = /p − b − w − t − d − s − z/ and V = /i − a/, were produced at two different speaking rates (conversational and fast). Results indicated anticipatory coarticulatory effects and constant intermuscular activity patterns across the two speech rates. [Research supported by N.I.H. Grant NS‐10424.]

Perceptual and acoustic evidence for tradeoffs in anticipatory coarticulation effects

The stimuli were /stri, stru/ pairs produced by four talkers. A part of /s/, /t/, or /r/ was exchanged between the pairs so that any coarticulatory information within the exchanged interval incorrectly predicted the remainder of the sequence. Recognition time (RT) was observed to final vowel targets /i/ or /u/. Results showed generally slower RT (interference) in exchanged compared to intact (as spoken) sequences. Within a given pair, however, when there was large interference from exchange of /t/ there was little interference from exchange of /r/, and vice versa. The tradeoff in perceptual effects between /t/ and /r/ was mirrored in acoustic differences, e.g., rate of F2 rise in /r/ toward /i/. Apparently the effects of anticipatory coarticulation are not necessarily always manifested to the greatest degree in the nearest segment. Articulation appears to be a more or less closed system of mutual constraints among segments in a sequence such that the various tokens of a given sequence may be seen as members of a set or class of utterances. Discussion concerns some theoretical implications of this notion for production and perception. [Work supported by NIH.]

Perception of anticipatory coarticulation effects in vowel-stop consonant-bowel sequences.

Articulatory and acoustic studies of speech production have shown that the effects of anticipatory coarticulation may extend across several segments of an utterance. The present experiments show that such effects have perceptual significance. In two experiments, a talker produced consonant (C) and vowel (V) sequences in a sentence frame (e.g., "I say pookee") of the form "I say / C V1 C V2/" in which V1 was /u, ae/ and V2 was /i, a/. Each /i, a/ sentence pair was cross-spliced by exchanging the final syllable /C V2/ so that coarticulatory information prior to the crosspoint was inappropriate for te final vowel (V2) in crossed sentences. Recognition time (RT) for V2 in crossed and intact (as spoken) sentences was obtained from practiced listeners. In both experiments RT was slower in crossed sentences; crossed sentences also attracted more false alarms. The pattern of perceptual results was mirrored in the pattern of precross acoustic differences in experimental sentences (e.g., formants F2 and F3 were higher preceding /i/ than preceding /a/). Pretarget variation in the formants jointly predicted amount of RT interference in crossed sentences. A third experiment found interference (slower RT) and also facilitation (faster RT) from exchanges of pretarget coarticulatory information in sentences. Two final experiments showed that previous results were not dependent on the use of practiced listeners.

Perception of anticipatory coarticulation effects in vowel–stop consonant–vowel sequences.

Perception of anticipatory coarticulation effects in vowel–stop consonant–vowel sequences.

Perception of anticipatory coarticulation effects in /stri, stru/ sequences

These experiments assess the identity and temporal reach of anticipatory coarticulation effects in /stri, stru/ sequences which can influence perception. Spoken /stri, stru/ pairs were cross‐spliced after /s/ and after /t/ so that any coarticulatory information preceding the crosspoint incorrectly predicted the remainder of the sequence. Results were that recognition time (RT) to final vowel targets /i/ or /u/ was slower in crossed compared to intact (as spoken) sequences. Differences in RT between crossed and intact sequences were much smaller when crossing after /s/ than after /t/. Further work examines the relative contributions of acoustic information throughout the /str/ region. These experiments and others reported earlier on vowel‐to‐vowel effects show that the perceptual effects of anticipatory coarticulation are not limited to information in adjacent segments. Theories may treat context effects from nonadjacent segments as unwanted or unaccounted‐for variance, but there is little reason to believe listeners do so. [Work supported by NIH.]

Perception of anticipatory coarticulation effects.

Articulatory and acoustic studies have shown that the effects of anticipatory coarticulation may extend across several segments in an utterance. But previous perceptual studies suggest that only the information carried by immediately adjacent segments is used in perception. To show that perception is not so limited, we persuaded ten talkers each to produce 12 sentences (e.g., "I say poozee") of the form "I say /C V1 z V2/" in which C was /p, t, k/, V1 was /u, ae/, and V2 was /i, a/. Each /i, a/ sentence pair was cross spliced by exchanging the final syllable /z V2/ so that coarticulatory information prior to the crosspoint was inappropriate for the final vowel V2 in crossed sentences. Recognition time (RT) for V2 in crossed and intact (as spoken) sentences was obtained from practiced listeners. The results were slower RT in crossed sentences, and amount of interference depended on both V1 and C context. Another experiment varied the location of crosspoints across /C V1/ and found that RT interference increased directly with amount and proximity to target of inappropriate precross acoustics. LPC analysis of the experimental sentences showed pretarget variations in F2 frequency which were jointly dependent on identity of C, V1, and V2. Pretarget F2 variations and C and V1 identity jointly predicted amount of RT interference in crossed sentences. Finally, experiments with pretarget F2 variations in synthetic speech repeated and extended the results with real speech. These studies lead to the conclusion that the perceptual significance of coarticulation is not limited to effects on immediately adjacent segments. Listeners appear to be sensitive to many acoustic effects of the mutual influence among the segments in a sequence.

Perceptual contributions of anticipatory coarticulation

In experiments reported by us at earlier meetings, the stimuli were sentences (e.g., “I say poozee”) of the form “I say/C V1 z V2/” in which C was /p, t, k/, V1 was /u, ae/, and V2 was /i, a/. Each /i, a/ sentence pair was cross‐spliced by exchanging the final syllable /z V2/and recognition time (RT) obtained to V2 in crossed and intact (as spoken) sentence versions. The results were slower RT in crossed sentences, suggesting misleading coarticulatory information prior to the cross. LPC analysis showed pretarget variations in formant movement were jointly dependent on identity of C, V1, and V2 and were correlated with RT interference in crossed sentences. The present paper reports acoustic and perceptual results of effects of anticipatory coarticulation between other V1‐V2 combinations and across other consonant classes. [Supported by NINCDS.]