Achievement

A team of IGERT participants is developing a provocative new approach to the problem of how children might learn to identify sound alternations and the relations between the surface allophones of individual abstract phonemes by drawing on a combination of techniques from machine learning (Computer Science), early language learning (Psychology/Linguistics) and language diversity (Linguistics). In contrast to standard approaches that first learn surface phonetic categories and then the more abstract phonological categories, the team has arrived at the surprising result that more robust learning is achieved by directly learning the abstract categories from the uncategorized speech corpora.