Fusing Task-Oriented and Open-Domain Dialogues in Conversational Agents

The goal of building intelligent dialogue systems has largely been separately pursued under two paradigms: task-oriented (TOD) systems, which perform task-specific functions, and open-domain (ODD) focus on non-goal-oriented chitchat. modes can potentially be intertwined together seamlessly in the same conversation, as easily done by a friendly human assistant. Such ability is desirable conversational agents, integration makes them more accessible useful. Our paper addresses this problem fusing TODs ODDs multi-turn dialogues. Based popular TOD dataset MultiWOZ, we build new FusedChat, rewriting existing turns adding ODD turns. This procedure constructs conversation sessions containing exchanges from both modes. It features inter-mode contextual dependency, i.e., depend each other. Rich dependency patterns such co-reference ellipsis are included. dataset, with 60k human-written 5k re-written turns, offers benchmark to test model's conversations. challenging task since model determine appropriate mode generate response based context. However, models would better mimic human-level capabilities. We evaluate baseline task, including classification-based two-stage two-in-one fused models. publicly release FusedChat baselines propel future work systems.