Label-Efficient Training for Next Response Selection

Seungtaek Choi, Myeongho Jeong, Jinyoung Yeo, Seung-won Hwang


Abstract
This paper studies label augmentation for training dialogue response selection. The existing model is trained by “observational” annotation, where one observed response is annotated as gold. In this paper, we propose “counterfactual augmentation” of pseudo-positive labels. We validate that the effectiveness of augmented labels are comparable to positives, such that ours outperform state-of-the-arts without augmentation.
Anthology ID:
2020.sustainlp-1.22
Volume:
Proceedings of SustaiNLP: Workshop on Simple and Efficient Natural Language Processing
Month:
November
Year:
2020
Address:
Online
Editors:
Nafise Sadat Moosavi, Angela Fan, Vered Shwartz, Goran Glavaš, Shafiq Joty, Alex Wang, Thomas Wolf
Venue:
sustainlp
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
164–168
Language:
URL:
https://aclanthology.org/2020.sustainlp-1.22
DOI:
10.18653/v1/2020.sustainlp-1.22
Bibkey:
Cite (ACL):
Seungtaek Choi, Myeongho Jeong, Jinyoung Yeo, and Seung-won Hwang. 2020. Label-Efficient Training for Next Response Selection. In Proceedings of SustaiNLP: Workshop on Simple and Efficient Natural Language Processing, pages 164–168, Online. Association for Computational Linguistics.
Cite (Informal):
Label-Efficient Training for Next Response Selection (Choi et al., sustainlp 2020)
Copy Citation:
PDF:
https://aclanthology.org/2020.sustainlp-1.22.pdf
Video:
 https://slideslive.com/38939444