For successful language use, interlocutors must be able to accurately assess their shared knowledge (“common ground”). Such knowledge can be accumulated through linguistic and non-linguistic context, but the same context can be associated with different patterns of knowledge, depending on the interlocutor’s participant role (Wilkes-Gibbs and Clark, 1992). Although there is substantial evidence that children’s ability to model partners’ knowledge develops gradually, most such evidence focuses on non-linguistic context. We investigated the extent to which 8- to 10-year-old children can assess common ground developed through prior linguistic context, and whether this is sensitive to variations in participant role. Children repeatedly described tangram figures to another child, and then described the same figures to a third child who had been a side-participant, an overhearer, or absent during the initial conversation. Children showed evidence of partner modeling, producing shorter referential expressions with repeated mention to the same partner. Moreover, they demonstrated sensitivity to differences in common ground with the third child based on participant role on some but not all measures (e.g., description length, but not definiteness). Our results suggest that by ten, children make distinctions about common ground accumulated through prior linguistic context but do not yet consistently deploy this knowledge in an adult-like way.