The success of high-variability phonetic training techniques (HVPT) for second-language learners (e.g., Japanese adults learning English /r/-/l/) demonstrates that plasticity in speech perception is retained through adulthood. HVPT has been thought to be effective because it exposes listeners to naturalistic phonetic variability, from many talkers and phonetic contexts [e.g., Logan et al., J. Acoust. Soc. Am. 89, 874-886 (1991)]. The present investigation examined: (a) whether the learning of new phonetic categories is best promoted by exposing the learner to natural token variability or by enhancing key acoustic cues and (b) whether exposure to natural tokens at some stage of the auditory training is essential for category learning.
Groups of Japanese listeners were trained on the /r/- /l/ distinction in initial position using one of four training methods. In all training methods, listeners were exposed to tokens produced by 10 talkers during ten training sessions. In the HVPT training method, they were trained using natural stimuli only. In the All Enhanced training method, F3 was set to extreme values and listeners were never exposed to natural stimuli. In the Perceptual Fading training method, the amount of F3 enhancement progressively decreased during training; in the Secondary-Cue Variability training method, variability in cues that are not informative for the /r/-/l/ distinction (F2 and closure duration) was progressively increased during training.
All four training methods were similarly effective in improving /r/-/l/ identification by an average of 15- 20 percentage points for Japanese adults. The presence of naturalistic phonetic variability may not therefore be critical to the effectiveness of HVPT. The implications of these findings for exemplar models of speech perception will be discussed. There appeared to be no advantage in enhancing acoustic cues that are critical to the phonetic contrast. [Work supported by ESRC grant RES-000-22-0445]