A Trigger-Free Method Enhanced by Coreference Information for Document-Level Event Extraction

Fumin Chen, Xu Wang, Xiaohui Liu, Dezhong Peng

Abstract: Document-level event extraction (DEE) aims to extract structured event information from a document. Previous document-level event extraction methods relied on trigger annotation, which was very expensive and time-consuming. In addition, entity representation plays an important role in the overall event extraction task, but we found that the previous work could not effectively use entity abbreviations and coreference information, which limited the representation ability of event-related entities in documents. Based on the above two aspects, we propose a new model named TFECI. In our model, we propose an efficient and effective method to make full use of abbreviations and coreference information to better exert the representational ability of event-related entities. Besides, triggers play an extremely central role in an event, but trigger annotation is often very difficult, so we propose a new strategy to select pseudo-trigger automatically. Experiments show that, compared with the previous system, our system can extract events without trigger annotation and achieve competitive results.