A Rate-Distortion view of human pragmatic reasoning

  • 2020-05-13 22:04:27
  • Noga Zaslavsky, Jennifer Hu, Roger P. Levy
  • 41


What computational principles underlie human pragmatic reasoning? A prominentapproach to pragmatics is the Rational Speech Act (RSA) framework, whichformulates pragmatic reasoning as probabilistic speakers and listenersrecursively reasoning about each other. While RSA enjoys broad empiricalsupport, it is not yet clear whether the dynamics of such recursive reasoningmay be governed by a general optimization principle. Here, we present a novelanalysis of the RSA framework that addresses this question. First, we show thatRSA recursion implements an alternating maximization for optimizing a tradeoffbetween expected utility and communicative effort. On that basis, we study thedynamics of RSA recursion and disconfirm the conjecture that expected utilityis guaranteed to improve with recursion depth. Second, we show that RSA can begrounded in Rate-Distortion theory, while maintaining a similar ability toaccount for human behavior and avoiding a bias of RSA toward random utteranceproduction. This work furthers the mathematical understanding of RSA models,and suggests that general information-theoretic principles may give rise tohuman pragmatic reasoning.


Quick Read (beta)

loading the full paper ...