FriSem

Gedeon Deák, Professor of Cognitive Science and Human Development, at UCSD who leads the Cognitive Development Lab

Title: Hidden Figures: Finding Statistical Regularities That Teach Infants Language and Social Skills

Abstract: Human infants' capacity to learn language and communication has fascinated thinkers for millenia. In the past century psychological research has tended to address this mystery by attempting to specify species-specialized learning phenotypes. That enterprise addresses only half of a "dual modelling problem" (Deák, Bartlett, & Jebara, 2007): the need to understand not only learning phenotypes but also the event- and information-structures in learners' environments that support learning. I will describe two lines of research that contribute to broader efforts to reveal information-structures that are often non-obvious even to experts in infant behavior. First, I will summarize results suggesting that infants' attention-following (often considered a cornerstone of intersubjective communication and social learning) is a skill learnable by simple agents from infant-typical social patterns. Additionally, infants' slow acquisition of the skill is likely due to a sparse learning signal. Second, I will summarize newer findings that infants can learn to classify meaning content in infant-directed speech from distributional contextual statistics. Moreover, a full account of the 'richness of the stimuli' should consider co-occurrences of both non-linguistic and linguistic cues, and linguistic content cues across multiple scopes. Through these examples I hope to encourage consideration of the potentially learnable information remaining to be discovered in infants' social environments, as well as more objective consideration of infants' general and specialized learning phenotypes.