Reasoning with Latent Structure Refinement for Document-Level Relation Extraction

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1546–1557 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 1546 Reasoning with Latent Structure Refinement for Document-Level Relation Extraction Guoshun Nan1∗, Zhijiang Guo1∗, Ivan Sekuli´c2† and Wei Lu1 1StatNLP Research Group, Singapore University of Technology and Design 2Università della Svizzera italiana guoshun_nan@sutd.edu.sg, zhijiang_guo@mymail.sutd.edu.sg ivan.sekulic@usi.ch, luwei@sutd.edu.sg Abstract Document-level relation extraction requires in- tegrating information within and across mul- tiple sentences of a document and capturing complex interactions between inter-sentence entities. However, effective aggregation of relevant information in the document remains a challenging research question. Existing approaches construct static document-level graphs based on syntactic trees, co-references or heuristics from the unstructured text to model the dependencies. Unlike previous methods that may not be able to capture rich non-local interactions for inference, we pro- pose a novel model that empowers the re- lational reasoning across sentences by auto- matically inducing the latent document-level graph. We further develop a refinement strat- egy, which enables the model to incrementally aggregate relevant information for multi-hop reasoning. Specifically, our model achieves an F1 score of 59.05 on a large-scale document- level dataset (DocRED), significantly improv- ing over the previous results, and also yields new state-of-the-art results on the CDR and GDA dataset. Furthermore, extensive analyses show that the model is able to discover more accurate inter-sentence relations. 1 Introduction Relation extraction aims to detect relations among entities in the text and plays a significant role in a variety of natural language processing applica- tions. Early research efforts focus on predicting re- lations between entities within the sentence (Zeng et al., 2014; Xu et al., 2015a,b). However, valu- able relational information between entities, such as biomedical findings, is expressed by multiple mentions across sentence boundaries in real-world scenarios (Peng et al., 2017). Therefore, the scope ∗ Equally Contributed. † Work done during internship at SUTD. Lutsenko is a former minister of internal affairs. He occupied this post in the cabinets of Yulia Tymoshenko. The ministry of internal affairs is the Ukrainian police authority. Subject: Yulia Tymoshenko Object:Ukrainian Relation: country of citizenship Figure 1: An example adapted from the DocRED dataset. The example has four entities: Lutsenko, inter- nal affairs, Yulia Tymoshenko and Ukrainian. Here en- tity Lutsenko has two mentions: Lutsenko and He. Men- tions corresponding to the same entity are highlighted with the same color. The solid and dotted lines repre- sent intra- and inter-sentence relations, respectively. of extraction in biomedical domain has recently been expanded to cross-sentence level (Quirk and Poon, 2017; Gupta et al., 2018; Song et al., 2019). A more challenging, yet practical extension, is the document-level relation extraction, where a sys- tem needs to comprehend multiple sentences to infer the relations among entities by synthesizing relevant information from the entire document (Jia et al., 2019; Yao et al., 2019). Figure 1 shows an example adapted from the recently proposed document-level dataset DocRED (Yao et al., 2019). In order to infer the inter-sentence relation (i.e., country of citizenship) between …

Content truncated to 500 words