Abstract:

In a pre-test, 20 Japanese listeners' identification of American English vowels /(cursive beta), (ae ligature), (open aye), (inverted vee), (open oh)/ was examined, using syllables /bVb, bVp, dVd, dVt, gVg, gVk/ embedded in sentences produced by four men (CVCs) and real words produced by two men and one woman (words). Experimental subjects (N=10) completed nine training sessions on /(ae ligature), (open aye), (inverted vee)/, using CVCs (three speakers). Pre-test to Post-test comparisons showed that identification of trained vowels in CVCs improved for the experimental group (42% to 85% correct), but not for the control group (44% to 44%). The training effect generalized to the speaker not used in training. Experimental subjects also improved significantly on trained vowels in words (43% to 63%); the control group did not (44% to 47%). Again, the training effect transferred to novel speakers. Accuracy on untrained vowels /(cursive beta), (open oh)/ decreased somewhat for the experimental group, while it improved for the control group, suggesting the presence of response biases. Nevertheless, overall greater performance gains for the experimental group (23% for CVCs, 10% for words) versus the control group (6% for CVCs, 6% for words) suggesting that robust changes in perception of non-native vowels are possible with identification training. [Work supported by NIDCD and ATR.]