While speakers tailor referring expressions to the knowledge of their addressees, they do so imperfectly. Our goal here is to provide an explanation for this type of pattern by extending a probabilistic model introduced to explain perspective-taking behavior in comprehension. Using novel production data from a type of knowledge mismatch not previously investigated in production, we show that production patterns can also be explained as arising from the probabilistic combination of the speaker’s and the addressee’s perspectives. These results show the applicability of the multiple-perspectives approach to language production, and to different types of knowledge mismatch between conversational partners.