Neural Cross-Lingual Relation Extraction Based on Bilingual Word Embedding Mapping

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 399–409, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 399 Neural Cross-Lingual Relation Extraction Based on Bilingual Word Embedding Mapping Jian Ni and Radu Florian IBM Research AI 1101 Kitchawan Road, Yorktown Heights, NY 10598, USA {nij, raduf}@us.ibm.com Abstract Relation extraction (RE) seeks to detect and classify semantic relationships between enti- ties, which provides useful information for many NLP applications. Since the state-of- the-art RE models require large amounts of manually annotated data and language-specific resources to achieve high accuracy, it is very challenging to transfer an RE model of a resource-rich language to a resource-poor lan- guage. In this paper, we propose a new ap- proach for cross-lingual RE model transfer based on bilingual word embedding mapping. It projects word embeddings from a target lan- guage to a source language, so that a well- trained source-language neural network RE model can be directly applied to the target language. Experiment results show that the proposed approach achieves very good perfor- mance for a number of target languages on both in-house and open datasets, using a small bilingual dictionary with only 1K word pairs. 1 Introduction Relation extraction (RE) is an important informa- tion extraction task that seeks to detect and clas- sify semantic relationships between entities like persons, organizations, geo-political entities, lo- cations, and events. It provides useful informa- tion for many NLP applications such as knowl- edge base construction, text mining and question answering. For example, the entity Washington, D.C. and the entity United States have a CapitalOf relationship, and extraction of such relationships can help answer questions like “What is the capi- tal city of the United States?” Traditional RE models (e.g., Zelenko et al. (2003); Kambhatla (2004); Li and Ji (2014)) re- quire careful feature engineering to derive and combine various lexical, syntactic and seman- tic features. Recently, neural network RE mod- els (e.g., Zeng et al. (2014); dos Santos et al. (2015); Miwa and Bansal (2016); Nguyen and Grishman (2016)) have become very successful. These models employ a certain level of auto- matic feature learning by using word embeddings, which significantly simplifies the feature engineer- ing task while considerably improving the accu- racy, achieving the state-of-the-art performance for relation extraction. All the above RE models are supervised ma- chine learning models that need to be trained with large amounts of manually annotated RE data to achieve high accuracy. However, anno- tating RE data by human is expensive and time- consuming, and can be quite difficult for a new language. Moreover, most RE models require language-specific resources such as dependency parsers and part-of-speech (POS) taggers, which also makes it very challenging to transfer an RE model of a resource-rich language to a resource- poor language. There are a few existing weakly supervised cross-lingual RE approaches that require no hu- man annotation in the target languages, e.g., Kim et al. (2010); Kim …

Content truncated to 500 words