Proceedings of the Annual Meeting of the Cognitive Science Society

Theories of the pragmatic use of language in the Gricean tradition, and related formalizations such as the Rational Speech Act (RSA) modeling framework, have focused on information exchange between cooperative interlocutors. More recent work has emphasized the importance of other, potentially less-cooperative functions of language, such as persuasion, argumentation, and manipulation. In this work, we bridge these two traditions by developing a model of argumentative utterance choice in the RSA framework. We propose several formalizations of the argumentative strength of a signal. We then measure how well they capture behaviour in a simple utterance production task in which participants were assigned an explicit argumentative goal. We show that the most popular formalization of argumentative strength fails to capture some of the participants' utterance choices, and discuss some alternatives. Our models constitute a substantial improvement over a baseline model, and remaining unexplained patterns are discussed.