Meaning & inference in case of conflict

This paper applies a model of boundedly rational "level-k thinking" (c.f. Stahl and Wilson, 1995; Crawford, 2003; Camerer, Ho and Chong, 2004) to a classical concern of game theory: when is information credible and what shall I do with it if it is not? The model presented here extends and generalizes recent work in game-theoretic pragmatics (Stalnaker, 2006; Jäger, 2007; Benz and van Rooij, 2007). Pragmatic inference is modeled as a sequence of iterated best responses, defined here in terms of the interlocutors' epistemic states. Credibility considerations are a special case of a more general pragmatic inference procedure at each iteration step. The resulting analysis of message credibility improves on previous game-theoretic analyses, is more general and places credibility in the linguistic context where it, arguably, belongs.