Mechanisms of selective feature adaptation

Abstract

Eimas and Corbit (1973) have postulated the existence of feature detectors as part of the human speech recognition mechanism. They have suggested that the existence of such detectors can be demon­ strated by repeated stimulation so that they become fatigued and less sensitive to further stimulation. In particular, Eimas and Corbit have postulated that there are two feature detectors tuned to different values of voice onset time (VOT), the time between the release burst in a stop consonant and the start of laryngeal pulsing. Lisker and Abramson (1967) have shown that this dimension is appropriate for classify­ ing stop consonants as (negative or short values of VOT) or (longer values of VOT). In the normal situation, it is supposed that the fea­ ture detector that is tuned to the value of VOT closest to that of the incoming stimulus responds more vigorously than the other, signaling to the higher region of the analyzing system the presence of the feature that this detector represents. If stimuli having VOT values intermediate between those appropriate for and those-appropriate for stop consonants are presented for identification, the position of the perceptual bound­ ary along the VOT dimension can be determined. Eimas and Corbit (1973) were able to demonstrate that if the voiced detector was fatigued by many repetitions of a stop consonant, the position of the boundary shifted away from the voiceless end of the VOT dimension towards the voiced end. It was found that if the adapting stimulus was Ibl the Ibl-/pl boundary shifted towards Ib/, whereas if it was Ipl the boundary shifted in the opposite direction. Similarly, the Id/-It! boundary could be shifted with Idl or It!. Furthermore, it was found that Idl shifted the Ibl-/pl boundary towards Ib/, and Ibl shifted the IdI-It! boundary towards Id/. These cross-series experiments showed that the detectors must respond to of the sounds rather than to complete syllables. Eimas, Cooper, and Corbit (1973) have taken the view that the detectors respond to linguistic 365 of the stimuli. These linguistic correspond approximately to the features of Jacobson, Fant, and Halle (1963). Further evidence for the existence of linguistic feature detectors has been provided by Cooper (1974), who demonstrated perceptual boundary shifts along a dimension related to place of articulation. Bailey (1974) has suggested that the adaptation takes place at the level of acoustic feature analysis, rather than at the higher linguistic levels. Acoustic are of the sound itself, such as formant frequencies or the time intervals between events, whereas linguistic are more abstract and refer to the language to which the syllable be­ longs rather than to physical attributes of the sound. Bailey has shown that it is only when the relevant of the sounds occupy the same region of the spectrum that adaptation takes place. Ainsworth (1975), Diehl (1975), and Ganong (1975) have attempted to distinguish between acoustic fea­ ture adaptation and linguistic feature adaptation. In each of these experiments, an attempt was made to devise adaptor syllables which had linguistic fea­ tures in common with the test syllables but did not share acoustic with them. Significant shifts were found when the adaptor and test syllables shared acoustic features, but smaller shifts occurred when the adaptor and test syllables shared linguistic, but not acoustic, features, suggesting that adapta­ tion might take place at both levels. Pisoni (1975), however, has demonstrated that the linguistic feature analyzers are not as general as required by distinctive feature theory. Consonant-vowel syllables do not adapt vowel-consonant syllables, and vice versa. Ades (1974) has attempted to determine the site of adaptation in the brain. By stimulating one ear with the adapter syllable and the other with the test syllable, he was able to show that adaptation occurred centrally. On the other hand, he also showed, by simultaneous adaptation of different ears, that each hemisphere of the brain could be adapted in a differ­ ent direction. This suggests that the specialized