Two Approaches to Building Collaborative, Task-Oriented Dialog Agents through Self-Play
This addresses the data scarcity issue for researchers and developers building task-oriented dialog systems, though it appears incremental as it builds on existing self-play and dialog training methods.
The paper tackles the problem of insufficient human/human dialog data for training task-oriented dialog systems by exploring two self-play approaches where agent-bots and user-bots autonomously interact in an API environment to develop communication strategies, with empirical results provided for reinforcement learning and game-theoretic equilibrium finding.
Task-oriented dialog systems are often trained on human/human dialogs, such as collected from Wizard-of-Oz interfaces. However, human/human corpora are frequently too small for supervised training to be effective. This paper investigates two approaches to training agent-bots and user-bots through self-play, in which they autonomously explore an API environment, discovering communication strategies that enable them to solve the task. We give empirical results for both reinforcement learning and game-theoretic equilibrium finding.