Perceptual learning of vowel identity

Abstract

In our recent experiments on vowel perception, subjects appeared able to improve their ability to identify vowels without the aid of training procedures or feedback. Two experiments explored this effect. In experiment I, error rates in the identification of vowels in b-vowel-b, vowel-b, and b-vowel syllables and isolated vowels were obtained at several points during an extensive test series. Improvement was shown for all syllable types throughout testing and ordinal relationships among them were maintained. In experiment II, a transfer paradigm was used to identify factors accounting for this improvement. Subjects received practice on d-vowel-d syllables or on isolated vowels produced by five talkers; practice consisted of a series of identification trails without feedback. An identification test on isolated vowels produced by a different set of talkers was then administered. Practice on both d-vowel-d syllables and on isolated vowels produced positive transfer relative to a control group. Also, a lower error rate on the test syllables was obtained after practice on the isolated vowels than was obtained after practice on the d-vowel-d syllables. Three factors are implicated in these results: (i) vocal tract normalization (ii) task learning, and (iii) perceptual learning. An interpretation emphasizing perceptual learning is offered. [Supported by NIMH.]