The aid of prosodic emphasis for distributional learning of vowel categories in French

It has been shown that for English the automatic classification of vowels into categories is aided by using tokens with prosodic emphasis (Adriaanse & Swingley, 2012). This study investigates whether the findings also hold in French. Using a corpus of conversational French speech, prosodically prominent vowels were automatically extracted. Using Gaussian mixture models to represent the vowel categories, two models were trained; one on the prominent vowels and one on all vowels. The results of the clustering show that the model trained only on the prominent vowels performs better than the model trained on all vowels. This suggests that infants learning the sounds of their first language will have an easier task if they pay attention to only the stressed vowels