Aligning Cross-Lingual Entities with Multi-Aspect Information

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4431–4441, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4431 Aligning Cross-Lingual Entities with Multi-Aspect Information Hsiu-Wei Yang1∗, Yanyan Zou2∗, Peng Shi1∗, Wei Lu2, Jimmy Lin1, and Xu Sun3,4 1 David R. Cheriton School of Computer Science, University of Waterloo 2 StatNLP Research Group, Singapore University of Technology and Design 3 MOE Key Lab of Computational Linguistics, School of EECS, Peking University 4 Center for Data Science, Beijing Institute of Big Data Research, Peking University {h324yang,p8shi,jimmylin}@uwaterloo.ca yanyan zou@mymail.sutd.edu.sg luwei@sutd.edu.sg, xusun@pku.edu.cn Abstract Multilingual knowledge graphs (KGs), such as YAGO and DBpedia, represent entities in dif- ferent languages. The task of cross-lingual entity alignment is to match entities in a source language with their counterparts in tar- get languages. In this work, we investigate embedding-based approaches to encode enti- ties from multilingual KGs into the same vec- tor space, where equivalent entities are close to each other. Specifically, we apply graph con- volutional networks (GCNs) to combine multi- aspect information of entities, including topo- logical connections, relations, and attributes of entities, to learn entity embeddings. To exploit the literal descriptions of entities expressed in different languages, we propose two uses of a pretrained multilingual BERT model to bridge cross-lingual gaps. We further propose two strategies to integrate GCN-based and BERT- based modules to boost performance. Exten- sive experiments on two benchmark datasets demonstrate that our method significantly out- performs existing systems. 1 Introduction A growing number of multilingual knowledge graphs (KGs) have been built, such as DBpedia (Bizer et al., 2009), YAGO (Suchanek et al., 2008; Rebele et al., 2016), and BabelNet (Navigli and Ponzetto, 2012), which typically represent real- world knowledge as separately-structured mono- lingual KGs. Such KGs are connected via inter- lingual links (ILLs) that align entities with their counterparts in different languages, exemplified by Figure 1 (top). Highly-integrated multilingual KGs contain useful knowledge that can benefit many knowledge-driven cross-lingual NLP tasks, such as machine translation (Moussallem et al., 2018) and cross-lingual named entity recognition ∗Equal contribution. Inter-Lingual Link Missing Inter-Lingual Link KG-English KG-Japanese e1 e3 u2 u3 almaMater country almaMater country e2 u1 English: University of Toronto Japanese: トロント大学 Attribute Value Attribute Value Name University of Toronto 大学名 トロント大学 Type Public University 学校種別 州立 Found Date 1827-03-15 創立年 1827 Campus Ontario キャンパス セントジョージ （トロント） Former Name King’s College 旧名 キングスカレッ ジ ... ... Descriptions The University of Toronto is a pub- lic research university in Toronto, Ontario, Canada · · · トロント大学 は、オンタリオ 州、トロントに本部を置くカナ ダの州立大学である · · · Figure 1: An example fragment of two KGs (in English and Japanese) connected by an inter-lingual link (ILL). In addition to the graph structures (top) consisting of entity nodes and typed relation edges, KGs also provide attributes and literal descriptions of entities (bottom). (Darwish, 2013). However, the coverage of ILLs among existing KGs is quite low (Chen et al., 2018): for example, less than 20% …

Content truncated to 500 words