Coreference Resolution: Are the Eliminated Spans Totally Worthless?

Up to date, various neural-based methods have been proposed for joint mention span detection and coreference resolution. However, existing studies on resolution mainly depend representations, while the rest spans in text are largely ignored directly eliminated. In this paper, we aim at investigating whether those eliminated totally worthless, or what extent they can help improve performance of To achieve goal, propose refine representation mentions with global including these ones leveraged. On basis, further introduce an additional loss term work encourage diversity between different entity clusters. Experimental results document-level CoNLL-2012 Shared Task English dataset show that indeed useful our approaches promising