Language that describes people in a concise manner may con- flict with social norms (e.g., referring to people by their race), presenting a conflict between transferring information effi- ciently and avoiding offensive language. When a speaker is describing others, we propose that listeners consider the speaker’s use or absence of potentially offensive language to reason about the speaker’s goals. We formalize this hypothe- sis in a probabilistic model of polite pragmatic language un- derstanding, and use it to generate predictions about interpre- tations of utterances in ambiguous contexts, which we test empirically. We find that participants are sensitive to poten- tially offensive language when resolving ambiguity in refer- ence. These results support the idea that listeners represent conflicts in speakers’ goals and use that uncertainty to inter- pret otherwise underspecified utterances.