Haskins Labs., 270 Crown St., New Haven, CT 06511
Catherine T. Best
Wesleyan Univ. and Haskins Labs.
Marianna Di Paolo
Univ. of Utah
Analysis based on phonemes assumes that ``u,'' for example, in whatever context is phonologically the same, and that deviations---sometimes extreme---imposed by following approximants do not affect the category membership of the vowel. Informed by Best's perceptual assimilation model, which predicts perception of non-native speech contrasts from their relationship to native segment inventories, this study investigates the contrasting possibility that American English contrasts like pool-pull and heel-hill are perceived independently of the contrasts between /u/--/(invverted closed omega)/ and /i/--/(small capital eye)/ in other contexts. Connecticut and Utah listeners heard productions of both contrasts by speakers from both states; the contrasts are acoustically less different for Utah than for CT speakers. Listener groups completed Keyword identification, forced-choice labeling, and AXB discrimination tasks. There were systematic listener dialect differences on the keyword and AXB tasks, and both groups did better on the CT than on the UT contrasts. Heel and hill were identified with the vowels in HID and/or HEED; however pool and Npull were identified with AWED or HOED more than with FOOD or HOOD. Even when pool and pull were identified with the same keyword, one class often contained better exemplars of the keyword than the other, and discrimination was still above chance. These results suggest that pool and pull, and heel and hill, minimally distinct in Utah, may differ primarily in their distance from /(open oh)/--/o/ and /(small capital eye)/, respectively. [Work supported by NIH Grant Nos. HD-01994 and DC-00403.]