The effect of distributional training on the production of non-native vowel contrasts

Abstract

The current study tests production accuracy of a non-native vowel contrast following a modified perceptual learning paradigm. Previous research has found that adults can learn non-native sound categories with sensitivity to distributional properties of their input. In the current study, 34 native-English adults heard stimuli from /o/-/oe/ continuum. Half of the participants heard stimuli drawn from a bimodal distribution and half from a unimodal distribution. To support learning, we incorporated active learning with feedback, lexical support in the form of two images, and overnight consolidation. Production of this contrast was assessed through a repetition task at baseline and at the end of the experiment. Production accuracy is measured as the Euclidean distance between /o/ and /oe/ at baseline and post-training. Preliminary analyses suggest that the distance between these two vowels increased for both groups, indicating that listeners in both conditions were able to transfer perceptual learning to production. These results suggest a way to mitigate the disadvantages previously found for participants in the unimodal condition, by incorporating active engagement with the target stimuli using lexical support, accuracy feedback, and overnight consolidation.