Chinese Event Causality Identification Based on Retrieval Enhancement.

Event causality identification (ECI) is a critical and challenging information extraction task, which aims to identify whether there is a causal relationship between the two events. To address the current problem of insufficient annotated data in Chinese event causality identification, and the rich semantics of the event itself is not exploited. We proposed a prompt-learning based retrieval enhancement Chinese events causality identification framework, named RE-CECI, which improves the few-shot learning capability of the model. Moreover, it enables the pretrained language model to better learn event information by adding retrieved examples to the input text, after which the model can learn more event information. In addition, we construct a retrieval store of retrieved examples to serve as clues for reasoning about causality. Our experimental results on the only available Chinese causality dataset, show that our proposed method significantly improves the performance of Chinese event causality identification.