A signal delection theory-based analysis of American English vowel identification and production performance by native speakers of Japanese

Abstract

The identification and production performance by two groups of native Japanese of the American English (AE) vowels /æ/, /a/, /■/, /■/, /■/ was measured before and after a six-week, identification training program. A signal detection theory (SDT) analysis of the confusion data, as measured by d′, revealed that all five AE vowels were more identifiable by the experimental trained group than the control untrained group. The d′ results showed that /■/ was less identifiable than /■/ in the pretest, even though the percentage identification rate for /■/ was slightly greater than that for /■/. Both groups productions of a list of CVCs, each containing one of the target AE vowels, were presented to a group of native AE listeners in a series of identification tasks. The d′ results revealed that the AE listeners could more sensitively identify the experimental groups post-test vowel productions than they could the control groups. SDT analysis also clarified an additional potentially confusing result: /■/ was somewhat less identifiable than /■/, despite the fact that the percentage identification rate for /■/ was higher. Overall, the SDT-based analysis served to change the pattern of results observed for L2 vowel identification and influenced the interpretation of the data.