The influence of infant-directed speech on vowel categorization and phonological rule learning

Abstract

This MSc thesis investigated the influence of infant-directed speech (IDS) on vowel categorization and phonological rule learning, using both computational modeling and a behavioral experiment. Previous research by Adriaans and Swingley (2012) investigated the effect of using tokens with prosodic emphasis on automatic clustering of vowel sets into categories. We have replicated and extended theses findings by using data from fourteen French speakers, both male and female. The results showed that tokens with prosodic emphasis improve the distributional learning of vowel categories in French, even when data of multiple speakers is used, suggesting that more IDS-like tokens could potentially aid in learning phonetic categories. Next, we investigated whether the learning of phonological rules is facilitated by IDS as compared to adult-directed speech (ADS) with an artificial grammar experiment that could test the learning of two phonological rules at the same time, and independent of each other. The findings suggested that IDS might facilitate phonological rule learning. Finally, we built a computational model using neural networks to mimic the artificial grammar experiment. We found that the model did not learn the phonological rules from the behavioral experiment, in contrast to the participants of the experiment. Therefor, we could not answer the question whether IDS aids the learning of phonological rules by the computational model.