Statistics and Abstraction in Today's Computational Models

Longstanding questions in cognitive science have centered on the role of statistical learning versus rules and abstraction. Today's computational language models, which emerged largely from the connectionist tradition, are often taken as a triumph of the statistical approach. But is that conclusion too fast? In a series of talks and a panel discussion, we consider how modern models can inform --- and be informed by --- connectionism, Bayesianism, generative linguistics, and other traditions in psychology and cognition.