BORT: Back and Denoising Reconstruction for End-to-End Task-Oriented Dialog

Abstract

A typical end-to-end task-oriented dialog system transfers context into dialog state, and upon which generates a response, which usually faces the problem of error propagation from both previously generated inaccurate dialog states and responses, especially in low-resource scenarios. To alleviate these issues, we propose BORT, a back and denoising reconstruction approach for end-to-end task-oriented dialog system. To improve the accuracy of dialog state that is essential for the task completion of dialog system, back reconstruction is used to reconstruct the original input context from the generated dialogue state since the inaccurate dialog state cannot recover its corresponding input context. To enhance the antinoise capability of the model, denosing reconstruction is used to reconstruct the corrupted dialog state and response. Extensive experiments conducted on MultiWOZ 2.0 and CamRest676 show the effectiveness of BORT which achieves state-of-the-art performance. Furthermore, BORT demonstrates its advanced capabilities in zero-shot domain scenarios and in low-resource scenarios.

Tasks

Reproductions