Towards Focused and Connected Document-Level Event Extraction

Anonymous

Abstract: Document-level event extraction (DEE) is indispensable when events are naturally described in the form of a document. Although previous methods have made great success on DEE, they are limited by two bottlenecks: losing focus and losing the connection. In this paper, to break through the above bottlenecks, we annotated a new dataset, named WIKIEVENT++, towards focused and connected DEE. Besides, we propose two different models to approach this task: the extractive model and the generative model. Experimental results verify the effectiveness of our proposed methods. We further present a promising case study to explore the performance bottleneck for this task. Data and code will be released at \url{http://anonymized} to advance the research on document-level event extraction.