Abstract:

Just as drivers depend on shared behavioral conventions to safely navigate traffic, successful communication depends on a set of shared linguistic conventions. All members of a language community share stable knowledge of global conventions, many of which evolved over long timescales. Yet communication partners are also remarkably flexible in their ability to dynamically coordinate on local, ad hoc conventions across a relatively small number of repeated interactions. In this talk, I’ll present results from a number of interactive reference games that investigate the conditions under which such conventions form, and the cognitive mechanisms that support this flexibility. These results motivate a theory of convention-formation where agents, though initially uncertain about word meanings in context, learn from a partner who is assumed to be using language with such knowledge. We formalize this theory in a computational model of language understanding as recursive social inference. 

Contact:

Robert Hawkins, Stanford University