LearnDA: Learnable Knowledge-Guided Data Augmentation for Event Causality Identification

Abstract

Modern models for event causality identification (ECI) are mainly based on supervised learning, which are prone to the data lacking problem. Unfortunately, the existing NLPrelated augmentation methods cannot directly produce available data required for this task. To solve the data lacking problem, we introduce a new approach to augment training data for event causality identification, by iteratively generating new examples and classifying event causality in a dual learning framework. On the one hand, our approach is knowledge guided, which can leverage existing knowledge bases to generate well-formed new sentences. On the other hand, our approach employs a dual mechanism, which is a learnable augmentation framework, and can interactively adjust the generation process to generate task-related sentences. Experimental results on two benchmarks EventStoryLine and Causal-TimeBank show that 1) our method can augment suitable task-related training data for ECI; 2) our method outperforms previous methods on EventStoryLine and Causal-TimeBank (+2.5 and +2.1 points on F1 value respectively).

Event Causality Identification via Generation of Important Context Words

“…Proposed by (Caselli and Vossen, 2017), EventStoryLine (i.e., version 0.9) involves 258 documents, 22 topics, 4316 sentences, 5334 event mentions, and 1770 of 7805 event mention pairs with causal relation in a sentence. Following the same data split in previous work Zuo et al, 2021), we utilize the last two topics in EventStoryLine for the development data while the remaining 20 topics are used for 5-fold cross-validation evaluation. For Causal-TimeBank (Mirza, 2014a), there are 184 documents, 6813 event mentions, and 318 of 7608 event mention pairs annotated with causal relation.…”

“…For Causal-TimeBank (Mirza, 2014a), there are 184 documents, 6813 event mentions, and 318 of 7608 event mention pairs annotated with causal relation. Using the same setting and data split as previous work Zuo et al, 2021), we perform 10-fold cross-validation evaluation.…”

“…Compared to the feature-based methods (Do et al, 2011;Ning et al, 2018), recent deep learning models have demonstrated their state-of-the-art performance for ECI (Kadowaki et al, 2019;Zuo et al, 2021). As such, prior work has mainly treated ECI as a classification problem where the only output from the models is a label to indicate causal or non-causal relation between input events.…”

ERGO: Event Relational Graph Transformer for Document-level Event Causality Identification

“…Zuo et al (2021a) learn contextspecific causal patterns from external causal statements and incorporate them into a target ECI model. Zuo et al (2021b) propose a data augmentation method to further solve the data lacking problem.…”

“…(3) LSIN (Cao et al, 2021), which constructs a descriptive graph to leverage external knowledge and has the current SOTA performance for intra-sentence ECI. ( 4) LearnDA (Zuo et al, 2021b), which uses knowledge bases to augment training data. ( 5) CauSeRL (Zuo et al, 2021a), which learns context-specific causal patterns from external causal statements for ECI.…”