Development of an Intelligent Dialogue Agent for Older Adults : Evaluation of Functions to Control Spontaneous Talk and Coordinate Speech Content

This paper aims to develop an agent, the Intelligent Dialogue Agent (IDA), which is able to engage in appropriate dialogue with a variety of older adults depending on their situation. Incorporating the IDA into preventive care frameworks is expected to encourage older adults to habitually engage in preventive care behaviors (e.g., exercise and cognitive training) due to their familiarity with the frameworks. We implemented a function into the IDA to spontaneously talk to older adults. Reinforcement learning based on the policy gradient method was also employed to develop a function to coordinate the IDA's speech content. In preparation for application to older adults, we carried out experiments with students to evaluate their impression of the IDA's spontaneous talking function using a prototype IDA with a smart speaker and also conducted the experiments using a text-based chat-bot with a speech-contents-coordinating function to discuss basic characteristics and performance of the function.