Job Talk: Dr. Andrew Martin. Fake it 'til you make it: Learning phonemes with a proto-lexicon.

Wednesday, February 12, 2014 3:00 PM to 4:00 PM EST
447

Before the end of the first year of life, infants begin to lose the ability to perceive distinctions between sounds that are not phonemic in their native language. It is typically assumed that this developmental change reflects the construction of language-specific phoneme categories, but how these categories are learned largely remains a mystery. In this talk I will consider two traditional approaches to this problem - one in which the contexts in which a sound appears are used to group sounds into phonemes, and another in which knowledge of words is used to learn the phoneme categories. I will show that although both approaches work on highly simplified example data, neither is sufficient when applied to language data of realistic complexity. I will propose a third approach, in which infants construct a crude proto-lexicon which approximates the set of words in their language, and then use it to extract information about phoneme categories. Using large corpora of naturalistic Japanese and Dutch data, I will show that this approach retains the advantages of earlier models of phoneme learning while avoiding many of their shortcomings.