15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Can Adult Phonetic Categories be Predicted Based on Statistical Distribution Alone?

Megha Sundara

McGill University, Canada

Recently there has been support for learning of phonetic categories based on sensitivity to distributional differences. Distribution-based approaches are powerful in that they make a priori predictions of discrimination performance for certain non-native contrasts. In the present set of studies we examined the limits of a distribution-based approach to predict phonetic categories of adults. We tested Canadian English, Canadian French, native Hindi and simultaneous bilingual Canadian English-Canadian French listeners on discrimination of a dental-alveolar contrast. These groups of listeners differ on whether they hear a unimodal or bimodal distribution of dental and alveolar stops during acquisition.
   We find that contrary to prediction, listeners exposed to a unimodal distribution show discrimination performance ranging from poor through intermediate to good. Listeners exposed to bimodal distribution, following prediction, show good discrimination performance. We discuss specific problems of applying distribution approaches to acquisition of categories in a natural context.

Full Paper

Bibliographic reference.  Sundara, Megha (2003): "Can adult phonetic categories be predicted based on statistical distribution alone?", In ICPhS-15, 2329-2332.