Efficient Dialogue State Tracking by Masked Hierarchical Transformer

TL;DR

This paper introduces a masked hierarchical transformer model for cross-lingual multi-domain dialogue state tracking, achieving high accuracy by jointly learning slot operations and state tracking with a novel fusion mechanism.

Contribution

It proposes a novel masked mechanism and joint learning approach for cross-lingual dialogue state tracking, improving performance on DSTC challenge datasets.

Findings

Achieved 62.37% joint accuracy on MultiWOZ dataset.

Achieved 23.96% joint accuracy on CrossWOZ dataset.

Demonstrated effectiveness of the proposed model in cross-lingual settings.

Abstract

This paper describes our approach to DSTC 9 Track 2: Cross-lingual Multi-domain Dialog State Tracking, the task goal is to build a Cross-lingual dialog state tracker with a training set in rich resource language and a testing set in low resource language. We formulate a method for joint learning of slot operation classification task and state tracking task respectively. Furthermore, we design a novel mask mechanism for fusing contextual information about dialogue, the results show the proposed model achieves excellent performance on DSTC Challenge II with a joint accuracy of 62.37% and 23.96% in MultiWOZ(en - zh) dataset and CrossWOZ(zh - en) dataset, respectively.