Leila Wehbe: University of California at Berkeley

Abstract: The advent of machine learning has allowed us to supplement
hypothesis-driven science with data-driven science. In neuroscience. Most
language experiments to date have studied the brain by crafting conditions
designed to isolate a single hypothesis. My research focuses on naturalistic
language experiments in which subjects process a rich text, and use machine
learning and natural language processing techniques to discover and test
multiple hypotheses. In this talk, I will describe a framework for making
inferences about what the brain represents along three levels. Level 1 focuses
on correct inference for a single naturalistic task (reading). Level 2 is concerned with combining data across subjects and tasks (reading, writing, speaking).Level 3 addresses the reproducibility of inferences drawn across subjects and tasks in entirely different experimental paradigms (controlled vs. naturalistic). My framework consists of a promising collaboration between machine learning, natural language processing and cognitive neuroscience that could help us better understand language processing in the brain.