Abstract
In order to refer using a name, speakers must believe that their addressee knows about the link between the name and the intended referent. In cases where speakers and addressees learned a subset of names together, speakers are adept at using only the names their partner knows. But speakers do not always share such learning experience with their conversational partners. In these situations, what information guides speakers' choice of referring expression? A speaker who is uncertain about a names' common ground (CG) status often uses a name and description together. This N+D form allows speakers to demonstrate knowledge of a name, and could provide, even in the absence of miscommunication, useful evidence to the addressee regarding the speaker's knowledge. In cases where knowledge of one name is associated with knowledge of other names, this could provide indirect evidence regarding knowledge of other names that could support generalizations used to update beliefs about CG. Using Bayesian approaches to language processing as a guiding framework, we predict that interlocutors can use their partner's choice of referring expression, in particular their use of an N+D form, to generate more accurate beliefs regarding their partner's knowledge of other names. In Experiment 1, we find that domain experts are able to use their partner's referring expression choices to generate more accurate estimates of CG. In Experiment 2, we find that interlocutors are able to infer from a partner's use of an N+D form which other names that partner is likely to know or not know. Our results suggest that interlocutors can use the information conveyed in their partner's choice of referring expression to make generalizations that contribute to more accurate beliefs about what is shared with their partner, and further, that models of CG for reference need to account not just for the status of referents, but the status of means of referring to those referents.