A Method of Generating a Dialogue Pattern to Induce Awareness Based on a Reflection Support Agent

Abstract

Patients with acquired brain injuries often present with adverse symptoms, such as attention, memory, and functional disorders, as well as aphasia, which prevent them from effectively executing activities of daily living. Cognitive rehabilitation can be conducted with these patients to mitigate these issues. A crucial component for the efficacy of this rehabilitation is the patient’s self-awareness. To induce “awareness”, patients are exposed to a process called reflection dialogue, wherein they watch a video of themselves interacting with a cognitive rehabilitation specialist. This process allows the patient to objectively observe themselves. However, this reflection dialogue process requires appropriate specialization according to the symptoms and needs of the patient, which depends on the experience of the cognitive rehabilitation specialist. The current study aimed to automatize this process using a system that includes a home robot that acts as an agent during the reflection dialogue. This system included three components: (1) Generation of a reflection dialogue; (2) Generation of agentive behavior, based on reinforcement learning of patients’ symptoms (according to patients’ daily logs); and (3) Generation of an appropriate conversation interval (i.e., timing/tempo of the human-robot conversation). This research used conversation methods in a reflection dialogue to induce awareness. Specifically, the current study proposed a method of generating a dialogue pattern through interactions with a reflection support agent, based on a Bayesian network that utilizes communication history. Results show the possibility of using a robot and web-agent to create dialogue. Further, it was found that the “Personal” type of reflection dialogue was significantly better at noticing remarks than the “General” type of reflection dialogue.