Effects of training procedures and vowel set on learning non-native vowel categories

Abstract

Results of four new experiments are reported which examine American English listeners’ perception of German front rounded vowels. They differ from previous experiments [Kingston et al., J. Acoust. Soc. Am. 99, 2602–2603(A) (1996)] in that listeners were trained with classification of all two-stimulus subsets of four-member sets of German vowels as well as with the complete four-stimulus identification task used by Logan et al. [J. Acoust. Soc. Am. 89, 874–876 (1991)] in training. Despite the addition of these easier tasks, as in the previous experiments, both learning and generalization were modest in the four new experiments. One new experiment replicated the result obtained earlier with the front rounded vowel set, /y,Y,o/,œ/, whose members contrast for [tense] and [high]: that any contrast involving mid lax /œ/ is particularly easy. Two other experiments examined the lax front vowels contrasting for [high] and [round], /i,Y,ε,œ/; both found correlated feature contrasts to be easier than single feature contrasts, which did not differ in difficulty. The fourth experiment examined the lax round vowels contrasting for [high] and [back], /u,Y,■,œ/, and found all contrasts to be equally easy except the backness contrast between the high vowels, /u:Y/, which was markedly harder. Listeners’ accuracy on a particular contrast thus varies with the set of vowels they have heard. [Work supported by NIH and NSF.]