Acoustic variability and perceptual learning of nonnative-accented speech sounds

The present study seeks to answer the questions of whether and how listeners may come to recognize nonnative-accented speech sounds as a result of formal training, and what impact the acoustic variability of non-native productions has on their learnability, in order to address the larger issue of how humans deal with unpredictability in the instantiations of categories during perceptual learning. In a high-variability, identification-with-feedback training study, it was observed that subjects do not respond to this type of training on accented words in the same way that they have been previously shown to with native-produced words. Trainees demonstrated adaptation to individual Spanish-accented speakers they encountered during training but showed no recognition advantages that were transferable to productions of other Spanish-accented speakers. An acoustic study comparing the accented vowel sounds included in the training study with a parallel set of native-produced sounds revealed a robust increase in variability in nonnatives' use of the perceptual vowel space, involving larger, less regular distributions and greater category overlap. In an additional training study where types and degree of variability in the use of vowel space were controlled artificially, subjects exposed to words with vowel distributions defined by variability typical of that observed in non-native productions did not improve in recognition accuracy over the course of learning as subjects exposed to lower levels of variability did. Post-training recognition of highly-varied sounds was not effected overall by the level of variability of training, though differences were observed in subjects' sensitivity to certain individual vowels. Results are discussed with respect to an exemplar model of perceptual learning, which it is suggested does not entirely account for the perception of accented speech.