Conversation analyses have revealed that information imbalances between speakers and hearers are represented in their speech and drive the epistemic engine to equalize the imbalances. Considering multi-modal communication, however, information can also be conveyed by body movements and reveal the unspoken imbalances in detail. Group reminiscence therapy is used to treat elderly people who are developing dementia, but it is a conversation process in which the epistemic engine involves cultural differences among the participants. In the present study, detailed analyses of conversation during therapy showed that speakers can use gestures to show their epistemic status and the information imbalance between the participants; hearers can imitate the speaker's gestures to show their understanding in the conversational sequence; unspoken epistemic differences can be revealed by the difference between the gestures of the speaker and hearer; and other participants can observe the difference visually and update a gesture to point out the unspoken difference. I discuss the multi-modal structure of an epistemic engine in reminiscence therapy and its implications for the care given to dementia patients.