PIMS - CSC Seminar: Bob McMurray

Developmental Systems Theory (Gottlieb, 2007; Johnston & Edwards, 2002; Oyama, 2000) offers a compelling approach to development. Stressing bidirectional flow of information between genes, biological structures like the brain, and behavior, it offers a powerful counterpoint to tired debates between nature and nurture. However it was largely developed in response to animal models and with exceptions, it has not made deep in roads into human cognitive development (though see, Spencer et al., 2009). The broad goal of this talk is to highlight the role that developmental computational modeling can play in bridging this divide. While developmental systems theory points out the important role of non-obvious causes of developmental phenomena, computational modeling can help us understand the non-obvious consequences of simple mechanisms like real-time competition, and statistical or associative learning. That is, it can reveal opportunities for emergence embedded within our theories, and highlight deeper developmental principles that are not always obvious from measures of brain or behavior. These are illustrated with a series of case studies illustrating a range of models from my lab that examine the development of word learning and speech perception. Across these case studies – which employ connectionist, dynamic systems, formal statistical learning, and even old-school associative paradigms – I highlight how parallel learning, interactions across time scales, and the statistics of the input can lead to unexpected solutions to developmental puzzles. More importantly, they point to principles of development, which can often transcend many specific modeling paradigms to help us understand development without our complex computational models.