Attention Modulation for Zero-Shot Cross-Domain Dialogue State Tracking

Mathilde Veron, Olivier Galibert, Guillaume Bernard, Sophie Rosset


Abstract
Dialog state tracking (DST) is a core step for task-oriented dialogue systems aiming to track the user’s current goal during a dialogue. Recently a special focus has been put on applying existing DST models to new domains, in other words performing zero-shot cross-domain transfer. While recent state-of-the-art models leverage large pre-trained language models, no work has been made on understanding and improving the results of first developed zero-shot models like SUMBT. In this paper, we thus propose to improve SUMBT zero-shot results on MultiWOZ by using attention modulation during inference. This method improves SUMBT zero-shot results significantly on two domains and does not worsen the initial performance with the great advantage of needing no additional training.
Anthology ID:
2022.codi-1.11
Volume:
Proceedings of the 3rd Workshop on Computational Approaches to Discourse
Month:
October
Year:
2022
Address:
Gyeongju, Republic of Korea and Online
Editors:
Chloe Braud, Christian Hardmeier, Junyi Jessy Li, Sharid Loaiciga, Michael Strube, Amir Zeldes
Venue:
CODI
SIG:
Publisher:
International Conference on Computational Linguistics
Note:
Pages:
86–91
Language:
URL:
https://aclanthology.org/2022.codi-1.11
DOI:
Bibkey:
Cite (ACL):
Mathilde Veron, Olivier Galibert, Guillaume Bernard, and Sophie Rosset. 2022. Attention Modulation for Zero-Shot Cross-Domain Dialogue State Tracking. In Proceedings of the 3rd Workshop on Computational Approaches to Discourse, pages 86–91, Gyeongju, Republic of Korea and Online. International Conference on Computational Linguistics.
Cite (Informal):
Attention Modulation for Zero-Shot Cross-Domain Dialogue State Tracking (Veron et al., CODI 2022)
Copy Citation:
PDF:
https://preview.aclanthology.org/nschneid-patch-4/2022.codi-1.11.pdf
Code
 mathilde-veron/attention-modulation-zero-dst