TY - JOUR
AU - 
AB - Kimani Gray, a young man who likes football, was killed in a police attack shortly after a tight match. S1: Modern models for event causality identifica- S2: In the week following the fatal violence, several protests have erupted because of the official statement. tion (ECI) are mainly based on supervised EDA deletion learning, which are prone to the data lack- S3: Kimani Gray, a young man who likes football, was killed in a police attack shortly after a tight match. ing problem. Unfortunately, the existing NLP- related augmentation methods cannot directly Figure 1: S1 and S2 are causal sentences that contain produce available data required for this task. causal events. S3 is produced by EDA based on S1. To solve the data lacking problem, we intro- The dotted line indicates the causal relation. duce a new approach to augment training data for event causality identification, by iteratively generating new examples and classifying event Riaz and Girju, 2014b; Hashimoto et al., 2014; Hu causality in a dual learning framework. On the and Walker, 2017; Gao et al., 2019). However, ex- one hand, our approach is knowledge guided, isting datasets are relatively small, which impede which can leverage existing knowledge bases 
TI - LearnDA: Learnable Knowledge-Guided Data Augmentation for Event Causality Identification
JF - Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
DO - 10.18653/v1/2021.acl-long.276
DA - 2021-01-01
UR - https://www.deepdyve.com/lp/unpaywall/learnda-learnable-knowledge-guided-data-augmentation-for-event-EcZCXzJmWf
DP - DeepDyve
ER -