Do you believe what eye believe?

Speaker belief is an essential component how interlocutors interact, according to information structure and other theories of conversational interaction. We tested this assumption by investigating unscripted spontaneous speech patterns in an experimental paradigm that manipulated speakers belief states about shared visual information. We found that interlocutors were more likely to use references based on content (rather than appearance) in mismatched belief conditions. Also, final rising pitch contours (confirmation contours) seemed to be used more when interlocutors share the same visual common ground. These contours seem to elicit more back-channels than final falling pitch contours. Our results provide evidence that situational variables such as visual common ground strongly effect how speakers create their utterances.