Abstract
Cooperative communication plays a central role in theories of humancognition, language, development, and culture, and is increasingly relevant inhuman-algorithm and robot interaction. Existing models are algorithmic innature and do not shed light on the statistical problem solved in cooperationor on constraints imposed by violations of common ground. We present amathematical theory of cooperative communication that unifies three broadclasses of algorithmic models as approximations of Optimal Transport (OT). Wederive a statistical interpretation for the problem approximated by existingmodels in terms of entropy minimization, or likelihood maximizing, plans. Weshow that some models are provably robust to violations of common ground, evensupporting online, approximate recovery from discovered violations, and deriveconditions under which other models are provably not robust. We do so usinggradient-based methods which introduce novel algorithmic-level perspectives oncooperative communication. Our mathematical approach complements and extendsempirical research, providing strong theoretical tools derivation of a prioriconstraints on models and implications for cooperative communication in theoryand practice.