Leverage Lexical Knowledge for Chinese Named Entity Recognition via Collaborative Graph Network

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3830–3840, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3830 Leverage Lexical Knowledge for Chinese Named Entity Recognition via Collaborative Graph Network Dianbo Sui1,2, Yubo Chen1, Kang Liu1,2, Jun Zhao1,2, Shengping Liu3 1 National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2 University of Chinese Academy of Sciences, Beijing, 100049, China 3 Beijing Unisound Information Technology Co., Ltd, Beijing, 100028, China {dianbo.sui, yubo.chen, kliu, jzhao}@nlpr.ia.ac.cn liushengping@unisound.com Abstract The lack of word boundaries information has been seen as one of the main obstacles to develop a high performance Chinese named entity recognition (NER) system. Fortunate- ly, the automatically constructed lexicon con- tains rich word boundaries information and word semantic information. However, inte- grating lexical knowledge in Chinese NER tasks still faces challenges when it comes to self-matched lexical words as well as the near- est contextual lexical words. We present a Col- laborative Graph Network to solve these chal- lenges. Experiments on various datasets show that our model not only outperforms the state- of-the-art (SOTA) results, but also achieves a speed that is six to fifteen times faster than that of the SOTA model.1 1 Introduction Named entity recognition (NER) aims to locate and classify certain occurrences of words or ex- pressions in unstructured text into predefined se- mantic categories such as the person names, lo- cations, organizations, etc. NER is an essen- tial pre-processing step for many natural language processing (NLP) applications, such as relation extraction (Bunescu and Mooney, 2005), even- t extraction (Chen et al., 2015), question answer- ing (Moll´a et al., 2006) etc. In English NER, LSTM-CRF models (Lample et al., 2016; Ma and Hovy, 2016; Chiu and Nichols, 2016; Liu et al., 2018) leveraging word-level representations and character-level representations achieve the state- of-the-art results. In this paper, we focus on Chinese NER. Com- pared with English, Chinese has no obvious word boundaries. Since without word boundaries infor- mation, it is intuitive to use character information 1The code is available at https://github.com/ DianboWork/Graph4CNER 希 尔 顿 北 京 机 场 离 开 离开 leave 北京机场 Beijing airport I-PER or I-ORG ? I-LOC The nearest contextual lexical words information Self matched lexical words information I-PER I-LOC, B-LOC or O? 了 Sentence：希尔顿离开北京机场了。(Mr. Hilton has left Beijing airport.) Matched Lexical Words: 希尔(Hill), 希尔顿(Hilton), 离开(leave), 北京(Beijing), 北京机场(Beijing Airport) Figure 1: An example sentence integrating the near- est contextual lexical words (red line) and self-matched lexical words (green line) only for Chinese NER (He and Wang, 2008; Liu et al., 2010; Li et al., 2014), although such meth- ods could result in the disregard of word informa- tion. However, word information is very useful in Chinese NER, because word boundaries are usu- ally the same as named entity boundaries. For ex- ample, as shown in Figure 1, the boundaries of the word “�¬::” (Beijing airport) are the same as the …

Content truncated to 500 words