WHO:	John Kingston , University of Massachusetts, Department of Linguistics
TOPIC:	From Ears to Categories: Intermediate Steps in Speech Recognition
ABSTRACT:	What happens between the moment when the acoustic signal arrives at the ear and the moment when the listener recognizes its phonological content?   I present the results of three experiments here that investigate the products of intermediate processes between these two moments.  First, a modification of the Garner paradigm was used to compare the discriminability of orthogonal  combinations of the acoustic properties that differ between intervocalic stops contrasting for [voice], and nonspeech analogues of such stops.  The combinations that corresponded to the observed patterns of covariation of acoustic properties in naturally produced stops were found to be more discriminable than the opposite combinations, in both speech and the matching nonspeech analogues.  This result suggests that the acoustic properties integrate into perceptual properties that mediate between the raw stuff of the acoustic signal and the distinctive feature values the listener recognizes in that signal.  Second, the mutual interactions between the members of an [l-r] continuum and a following [d-g] continuum were studied in an experiment in which listeners identified both the liquid and the stop on every trial.  The results replicated those reported by Mann (1980) et seq. in that listeners responded “g” more often after more [l]-like than more [r]-like liquids.  Although the stop’s acoustics had no complementary effect on liquid judgments, each judgment depended on the other: listeners responded “g” significantly more often when they identified the liquid as “l” than “r” and they responded “l” significantly more often when they identified the stop as “g” than “d”.  Decisions influence context effects as much if not more than the stimuli’s psychoacoustics.  Finally, the mutual perceptual interactions between the members of [b-d-g] continua in coda and onset positions in vowel- stop.stop-vowel sequences were studied in another experiment in which listeners identified both segments on every trial.  The duration of the silent gap between the two vowels was orthogonally varied.  Again, the results replicated and extended earlier ones, in this case those reported by Repp (1983): listeners were more likely to hear the first stop as the same as the second and more likely to hear the second as different from the first.  Moreover, coda vs onset position interacted strongly with place of articulation.  Listeners were more likely to hear the coda consonant as “d” when the onset was the most [d]-like stimulus, but more likely to hear the onset as “d” when the coda was anything but the most [d]-like stimulus. But in both coda and onset, listeners were more likely to hear “b” or “g” when the other consonant was anything but the most [b]-like or [g]-like stimulus, respectively.  (The most [d]-, [b]-, or [g]-like stimuli were also those that listeners identified most often as “d”, “b”, and “g”, respectively, so stimulus and response context effects aren’t distinct in this experiment.)  That is, coda “d” responses are assimilative, while onset “d” responses and both coda and onset “b” and “g” responses are instead contrastive.  This interaction between place markedness and syllable position occurs despite the identical (mirror image) acoustics of coronal place in coda and onset, and the equivalent acoustics of coronal vs labial and dorsal places.  Here, phonological asymmetries between marked and unmarked values of contrasts and more vs less prominent syllable positions  influence the context effects.  Taken together, the results of these experiments show that auditory and decisional processes interact complexly with one another and with linguistic knowledge in speech sound recognition. Mann, V. A. (1980). “Influence of preceding liquid on stop-consonantperception,” Perception &                                            Psychophysics, 28, 407-412. Repp, B. H. (1983). “Bidirectional context effects in the perception of VC-CV sequences,” Perception & Psychophysics, 33, 147-155.
WHEN:	5/28/2003 12:00:00 PM
WHERE:	Lattimore 513

Events Homepage