Communication as moving target tracking

We view communication partners as moving targets. Achieving the goal of communication, thus, requires tracking the conversational trajectory of the partner in real time. We formalize this as dynamic inference with an action-perception-learning cycle and use sequential Bayesian estimation to do that. Our information-theoretic, dynamic Bayesian formulation suggests to understand communication as a Markov decision process where one participant tries to simultaneously improve predictions about its partner’s future state, manipulate the partner into states that maximize predictive information and minimize decision costs. The dynamic inference cycle model offers an overarching framework in which the mathematical tools developed in different fields can be used for modelling communication. It also helps develop technologies for multimodal embodied interaction and human-like cognitive agents.