Department of Psychological and Brain Sciences
Indiana University Bloomington
Statistical Learning from the Infant’s Point of View
The world presents learners with many statistical regularities. Considerable evidence indicates that humans are adept at discovering those regularities across many different domains including language, vision, and social behavior. In this talk, I will present a set of experimental and computational studies to examine how young learners employ a statistical learning solution to solve several challenging tasks in early language learning, such as learning to recognize visual objects and learning to build word-referent mappings. We argue that the relevant data for infant statistical learning are not the statistics of the physical and social world but only the samples that emerge within the learners’ own experiences and through their daily interactions with social partners. Using head-mounted eye tracking and computer vision techniques, we analyzed infants’ egocentric video and parent speech collected from free-flowing parent-infant interactions, and we also designed lab experiments to study real-time cognitive processes that accumulate co-occurring statistics of heard words and seen objects. Our results show that young learners create their own curriculum by sampling and selecting certain aspects of statistical information from the environment. Such statistical data through active information selection have unique properties and distributions that facilitate early object learning and early word learning. I will conclude by discussing how infant learning can teach us about how machines can learn.
A pizza lunch will be served at 11:45am. The seminar will run from 12:00pm – 1:30pm.