Signal to Act: Game Theory in Pragmatics Michael Franke Abstract: This thesis offers a general game theoretic model of language use and interpretation and applies it to linguistic pragmatics in the vein of (Grice 1989). The model presented here ---called the IBR model--- explains pragmatic phenomena, such as conversational implicatures, as arising from a sequence of iterated best responses: starting from the literal, semantic meaning as a psychologically salient attractor of attention, speaker and hearer initially compute the rational best responses to a literal use or interpretation of expressions; subsequently, agents continue computing best responses to best responses, for as long as this is reasonable and their cognitive resources permit. This algorithmic solution procedure is simple and intuitively appealing. But more importantly, it has a clear epistemic interpretation as modelling so-called ``level-k thinking'' which has gained recent popularity in behavioral game theory (see Stahl and Wilson 1995, Ho et al. 1998, Camerer et al. 2004). Laboratory data supports the assumption that human reasoners are cognitively biased and possibly resource-bounded in the sense that they are susceptible to focal framing effects and perform theory of mind reasoning to possibly only a given depth k. Thus conceived, the IBR model formally implements a number of empirically attested assumptions about the cognitive architecture of human reasoners. The IBR model then effectively provides a novel non-equilibrium solution concept as a form of strong rationalizability (Battigalli 2006) in which these psychological assumptions have been implemented. The thesis aims to show how this turn towards psychological realism solves outstanding conceptual problems with game theoretic approaches to communication and moreover improves on predictions in linguistic applications. Firstly, by implementing semantic meaning as a focal attractor of attention, the IBR model singles out those strategies that conform to our intuitions about credible communication without altogether precluding the possibility, and even occasional optimality of lying, misleading and distrust (see Farrell and Rabin 1996, Stalnaker 2006). Secondly, the model explicitly represents agents with absent or only limited capacity of taking opponent behavior and reasoning into account. This sheds light on higher-order theory of mind reasoning in language use and especially in the pattern of acquisition of pragmatic competence by young children (see Noveck 2001, Papafragou and Musolino 2003). An in-depth comparison of the IBR model with bidirectional optimality theory (Blutner 2000) suggests that the former is the better tool for modelling limitations in theory of mind reasoning in interpretation and acquisition. Finally, the IBR model unifies and extends a collection of recent work in game theoretic pragmatics (see especially Benz 2005, Stalnaker 2006, Benz and van Rooij 2007, Jaeger 2007). It yields formidable predictions for, among others, complex and nested cases of scalar implicatures, generalized M-implicatures and free-choice readings. The model also backs up natural accounts of conditional perfection, and unconditional readings of conditionals. Keywords: pragmatics, game theory, conversational implicatures, iterated best response, credibility, optimality theory, free choice disjunction, conditional perfection, biscuit conditionals