Supervised Relation Classification as Two-way Span-Prediction

Anonymous

Abstract: Most of the current supervised relation classification (RC) algorithms use a single embedding to represent the relation between a pair of entities. We argue that a better approach is to treat the RC task as a Span-Prediction (SP) problem, similar to Question Answering (QA). We present an SP-based system for RC and evaluate its performance compared to the embedding-based system. We demonstrate that by adding a few improvements, the supervised SP objective works significantly better than the standard classification-based objective. We achieve state-of-the-art results on the TACRED, SemEval task 8, and the CRE datasets.