| Task-oriented dialogue focuses on conversational agents that participate in user-initiated dialogues on domain-specific topics. Traditionally, the task-oriented dialogue community has often been hindered by a lack of sufficiently large and diverse datasets for training models across a variety of different domains. In an effort to help alleviate this problem, we release a corpus of 3,031 multi-turn dialogues in three distinct domains appropriate for an in-car assistant: calendar scheduling, weather information retrieval, and point-of-interest navigation. Our dialogues are grounded through knowledge bases ensuring that they are versatile in their natural language without being completely free form. |