The reference argument of epistemic must

Epistemic must is used to present a conclusion. In this paper, I explore the hypothesis that this should be modeled computationally using the notion of argument presented by Simari and Loui [16]. An utterance of must p in conversational context is interpreted as asserting that the argument hA; pi is justi ed in . The parameter A provides a set of reasoning rules which, along with factual premises from which they derive p, must be salient in for the utterance to be felicitous. Simari and Loui's formulation describes a relationship of defeat between arguments. Thus, in this account as in previous ones, the conclusions presented by epistemic must may be defeasible. This proposal improves on previous accounts in three key respects. First, the criterion that the argument be justi ed ensures that the speaker believes p when uttering must p. Second, the requirement that the speaker intend the hearer to recover the argument helps to explain the distribution and ofmust in discourse and the accommodation sometimes involved in understanding uses of must. Third, the link between the claim made by must and a speci c argument correctly predicts the variation in apparent force of the modal in di erent contexts: it varies according to the strength of the argument and the speaker's intentions in providing the argument. Because this interpretation for must incorporates restrictions based on salience into a framework designed to be relatively tractable, it may be uniquely suited for implementation.