From partners to populations: Balancing specificity and generalizability in the emergence of linguistic conventions

One of the most important properties of social conventions is their generalizability. They amortize costly on-the-fly coordination into priors that allow us to interact flexibly with new social partners in new situations. But how do generalizable conventions emerge in the first place when so much of social interaction is situation-specific? In this talk, I will present a model that aims to explain the balance between specificity and generalization via three basic components: structured uncertainty about which conventions will hold for which people, social inference to adapt to specific partners given past interactions, and hierarchical learning to abstract away "communal lexicons" that may apply more broadly across entire communities. I will test the model's predictions using data from experiments where we connected participants over the web for a series of communication games, allowing us to measure the emergence of signaling conventions in social groups. Together, this line of work aims to bridge the "micro" level of individual cognition with the "macro" level of collective behavior.

Speaker: Robert Hawkins, Linguistics Department, Stanford University

See weblink for entry instructions

Room 126