Coordinating on shared meaning for communication and collaboration
Robert Hawkins, Princeton Neuroscience Institute
Communication is the engine of human culture. When we’re able to communicate successfully, there’s nothing we can’t accomplish together. But when communication breaks down, we’re stopped in our tracks, stuck in incompatible worlds. My work investigates the cognitive mechanisms that allow people to successfully communicate, coordinate, and collaborate with one another, and what obstacles stand in their way. In this talk, I’ll argue that the central computational problem of communication is not simply transmission over a noisy channel, as in classical theories, but continual learning to coordinate on shared meaning over multiple timescales. Partner-specific common ground quickly emerges from social inferences within an interaction, while community-wide conventions are stable priors that have been abstracted away across interactions with different partners. I’ll show how this account may be formalized in a hierarchical Bayesian model called CHAI (Continual Hierarchical Adaptation through Inference), and present two natural-language communication experiments testing its predictions. In these experiments, participants are grouped into small communities and take turns referring to ambiguous tangram shapes. CHAI explains several empirical phenomena that have posed difficulties for prior models, including the convergence to more efficient referring expressions across repeated interaction with the same partner and the gradual generalization of partner-specific common ground to strangers. Finally, I’ll discuss four areas of ongoing work exploring broader implications of this theory, including (1) code-switching and the relationship between language and social identity, (2) neural mechanisms of continual learning in language use, (3) mental representations of norms and conventions across development, and (4) artificial agents that can flexibly form natural-language conventions with human partners. Taken together, this line of work aims to build a computational foundation for a more dynamic and socially-grounded view of meaning in communication.