Perceptual assimilation of American English vowels by Japanese listeners

To assess cross language patterns of perceptual assimilation directly, 24 Japanese (J) listeners were presented with American English (AE) vowels produced in citation form /hVba/ bisyllables and in a carrier sentence by 4 adult male speakers. They selected the J katakana character(s) which contained the vowel most similar to the AE vowel and rated the goodness of fit on a 7 point scale. Results showed that, as expected, AE vowels were judged as most similar to J vowels which were adjacent in acoustic articulatory "vowel space". However, the consistency of assimilation and goodness of fit of AE vowels to J categories varied considerably with speech style (citation vs sentence). Assimilation of long AE vowels to two mora response categories was much more consistent for target syllables produced in sentences. Acoustical analysis of stimuli suggested that listeners judged the duration of target vowels in citation bisyllables with respect to the following (constant) vowel. Other differences in perceptual assimilation patterns as a function of speech style could not easily be attributed to differences in speakers' productions. These results have implications for theories of L2 speech learning and for training studies of non native speech sounds.