Abstract: Conveying information in a false or indirect manner in consideration of listeners’ wants (i.e. being polite) seemingly contradicts an important goal of a cooperative speaker: information transfer. In the current work, we propose that people reason about polite speech as a tradeoff between speaker’s epistemic goals (conveying information in the most accurate, efficient, conserving way possible) and social goals (maintaining or boosting interactants’ self-image). We formalize this tradeoff within a probabilistic model of language understanding with an extension of Rational Speech Act framework: a cooperative speaker considers both epistemic utility (convey the truth) and social utility (make the listener feel good). We test these model predictions with empirical data on people’s inferences about the relation between speakers’ goals, utterances, and true states of the world. In an ongoing work looking at different types of polite speech (e.g. white lies, indirect speech) and participants from different cultural backgrounds (US, India and Korea), we begin to explore how the model can shed light on divergent conceptions of polite speech in various contexts.