Assessment of perceptual assimilation before and after training.

Abstract

Many non‐native speech sounds are challenging to perceive and, ultimately, to acquire. The development of a model of cross‐language speech perception and learning has been hampered by, among other issues, the high variability commonly observed in the perceptual assimilation of non‐native contrasts due to phonetic context, talker, and a variety of indexical properties of speech. This variability may reflect persistent patterns in perception that influence learning, or it may represent only an early sensitivity to low‐level phonetic detail that is attenuated as experience increases. To examine this issue, the perceptual assimilation by native speakers of American English of a variety of voicing and place contrasts from Hindi was assessed before and after training in a paired‐associate word learning task. Training utilized tokens from four of the six talkers used in perceptual assimilation tasks. The purpose of the study was to determine whether or not assimilation patterns could be greatly modified and/or simplified over the course of word learning in the laboratory. If limited training resulted in large changes in perceptual assimilations, then models of cross‐language speech perception and learning will need to rely on assimilation data elicited via procedures that result in stable and repeatable patterns.