A Simple Language Model for Task-Oriented Dialogue
This work addresses the problem of building efficient dialogue systems for task-oriented applications, offering a simple yet effective approach that improves over prior methods.
The paper tackles task-oriented dialogue by proposing SimpleTOD, a unified language model that recasts sub-tasks as sequence prediction, achieving state-of-the-art performance on the MultiWOZ dataset with improvements such as an 8.1-point increase in inform rate and a 9.7-point increase in success rate.
Task-oriented dialogue is often decomposed into three tasks: understanding user input, deciding actions, and generating a response. While such decomposition might suggest a dedicated model for each sub-task, we find a simple, unified approach leads to state-of-the-art performance on the MultiWOZ dataset. SimpleTOD is a simple approach to task-oriented dialogue that uses a single, causal language model trained on all sub-tasks recast as a single sequence prediction problem. This allows SimpleTOD to fully leverage transfer learning from pre-trained, open domain, causal language models such as GPT-2. SimpleTOD improves over the prior state-of-the-art in joint goal accuracy for dialogue state tracking, and our analysis reveals robustness to noisy annotations in this setting. SimpleTOD also improves the main metrics used to evaluate action decisions and response generation in an end-to-end setting: inform rate by 8.1 points, success rate by 9.7 points, and combined score by 7.2 points.