Honey or Poison? Solving the Trigger Curse in Few-shot Event Detection via Causal Intervention

Jiawei Chen, Hongyu Lin, Xianpei Han, Le Sun


Abstract
Event detection has long been troubled by the trigger curse: overfitting the trigger will harm the generalization ability while underfitting it will hurt the detection performance. This problem is even more severe in few-shot scenario. In this paper, we identify and solve the trigger curse problem in few-shot event detection (FSED) from a causal view. By formulating FSED with a structural causal model (SCM), we found that the trigger is a confounder of the context and the result, which makes previous FSED methods much easier to overfit triggers. To resolve this problem, we propose to intervene on the context via backdoor adjustment during training. Experiments show that our method significantly improves the FSED on both ACE05 and MAVEN datasets.
Anthology ID:
2021.emnlp-main.637
Volume:
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
Month:
November
Year:
2021
Address:
Online and Punta Cana, Dominican Republic
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
8078–8088
Language:
URL:
https://aclanthology.org/2021.emnlp-main.637
DOI:
10.18653/v1/2021.emnlp-main.637
Bibkey:
Cite (ACL):
Jiawei Chen, Hongyu Lin, Xianpei Han, and Le Sun. 2021. Honey or Poison? Solving the Trigger Curse in Few-shot Event Detection via Causal Intervention. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8078–8088, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
Honey or Poison? Solving the Trigger Curse in Few-shot Event Detection via Causal Intervention (Chen et al., EMNLP 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.emnlp-main.637.pdf
Video:
 https://aclanthology.org/2021.emnlp-main.637.mp4
Code
 chen700564/causalfsed
Data
MAVEN