Two Languages Are Better than One: Bilingual Enhancement for Chinese Named Entity Recognition

Proceedings of the 29th International Conference on Computational Linguistics, pages 2024–2033 October 12–17, 2022. 2024 Two Languages Are Better Than One: Bilingual Enhancement For Chinese Named Entity Recognition Jinzhong Ning1, Zhihao Yang1∗, Zhizheng Wang1, Yuanyuan Sun1, Hongfei LIN1, Jian Wang1 School of Computer Science and Technology, Dalian University of Technology Jinzhong_Ning@mail.dlut.edu.cn, wzz0727@gmail.com {yangzh,syuan,hflin,wangjian}@dlut.edu.cn Abstract Chinese Named Entity Recognition (NER) has continued to attract research attention. How- ever, most existing studies only explore the internal features of the Chinese language but neglect other lingual modal features. Actually, as another modal knowledge of the Chinese lan- guage, English contains rich prompts about en- tities that can potentially be applied to improve the performance of Chinese NER. Therefore, in this study, we explore the bilingual enhance- ment for Chinese NER and propose a unified bilingual interaction module called the Adapted Cross-Transformers with Global Sparse At- tention (ACT-S) to capture the interaction of bilingual information. We utilize a model built upon several different ACT-Ss to integrate the rich English information into the Chinese rep- resentation. Moreover, our model can learn the interaction of information between bilinguals (inter-features) and the dependency information within Chinese (intra-features). Compared with existing Chinese NER methods, our proposed model can better handle entities with complex structures. The English text that enhances the model is automatically generated by machine translation, avoiding high labour costs. Experi- mental results on four well-known benchmark datasets demonstrate the effectiveness and ro- bustness of our proposed model. 1 Introduction “One language sets you in a corridor for life. Two languages open every door along the way.” —Frank Smith, Psycholinguist Named entity recognition (NER) is the task of de- termining spans and semantic categories of named entities such as organization (ORG), person (PER) and location (LOC) in given free text. As the cor- nerstone of a wide range of natural language pro- cessing tasks, NER plays an essential role in many downstream tasks, such as relation extraction (Ze- ∗Corresponding Author Chinese: 美国“亚特兰蒂斯”号航 天飞机上的宇航员… [美国 LOC] [“亚特兰蒂斯”号航天飞机 ORG] [美国“亚特兰蒂斯” 号航天飞机 ORG] English: Astronauts on the US space shuttle Atlantis… Chinese English Single language model Our model 美国 “亚特兰蒂斯”号 航天飞机 上的 宇航员 the US Atlantis space shuttle on astronauts Figure 1: An example of Chinese NER via bilingual enhancement. The entity types of named entities are highlighted. lenko et al., 2003) and question answer (Diefen- bach et al., 2018). Compared with English NER, Chinese NER meets a series of challenges caused by the char- acteristics of Chinese. Aside from the lack of natu- ral word boundary information, Chinese named entities usually vary significantly in length and have complex compositional structures (Dong et al., 2016). At the same time, the annotated data of Chi- nese NER is relatively scarce, and it is difficult and costly to annotate the data manually (Liu et al., 2022). Hence, without more annotated data, it is a promising approach to improve Chinese NER by leveraging external information resources, which has attracted more and more research attention. One way to utilize external resources is to per- …

Content truncated to 500 words