Memorizing All for Implicit Discourse Relation Recognition

Implicit discourse relation recognition is a challenging task due to the absence of necessary informative clues from explicit connectives. An implicit recognizer has carefully tackle semantic similarity sentence pairs and severe data sparsity issue. In this article, we learn token embeddings encode structure dependency point view in their representations use them initialize baseline model make it really strong. Then, propose novel memory component issue by allowing master entire training set, which helps achieving further performance improvement. The mechanism adequately memorizes information pairing relations all instances, thus filling slot data-hungry current recognizer. proposed component, if attached with any suitable baseline, can help enhancement. experiments show that our full memorizing provides excellent results on PDTB CDTB datasets, outperforming baselines fair margin.