Agents to assist with rescue, surgery, and similar activities could collaborate better with humans if they could learn new strategic behaviors through communication. We introduce a novel POMDP dialogue policy for learning from people. The policy has 3-way grounding of language in the shared physical context, the dialogue context, and persistent knowledge. It can learn distinct but related games, and can continue learning across dialogues for complex games. A novel sensing component supports adaptation to information-sharing differences across people. The single policy performs better than oracle policies customized to specific games and information behavior.

An assistive robot that could communicate with dementia patients would have great social benefit. An assistive robot Pepper has been designed to administer Referential Communication Tasks (RCTs) to human subjects without dementia as a step towards an agent to administer RCTs to dementia patients, potentially for earlier diagnosis. Currently, Pepper follows a rigid RCT script, which affects the user experience. We aim to replace Pepper’s RCT script with a dialogue management approach, to generate more natural interactions with RCT subjects. A Partially Observable Markov Decision Process (POMDP) dialogue policy will be trained using reinforcement learning, using simulated dialogue partners. This paper describes two RCT datasets and a methodology for their use in creating a database that the simulators can access for training the POMDP policies.

This paper presents MDP policy learning for agents to learn strategic behavior–how to play board games–during multimodal dialogues. Policies are trained offline in simulation, with dialogues carried out in a formal language. The agent has a temporary belief state for the dialogue, and a persistent knowledge store represented as an extensive-form game tree. How well the agent learns a new game from a dialogue with a simulated partner is evaluated by how well it plays the game, given its dialogue-final knowledge state. During policy training, we control for the simulated dialogue partner’s level of informativeness in responding to questions. The agent learns best when its trained policy matches the current dialogue partner’s informativeness. We also present a novel data collection for training natural language modules. Human subjects who engaged in dialogues with a baseline system rated the system’s language skills as above average. Further, results confirm that human dialogue partners also vary in their informativeness.