Mongolian Named Entity Recognition System with Rich Features

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 505–512, Osaka, Japan, December 11-17 2016. Mongolian Named Entity Recognition System with Rich Features Weihua Wang, Feilong Bao∗, Guanglai Gao College of Computer Science, Inner Mongolia University Huhhot, China, 010021 Email: wangweihuacs@163.com;{csfeilong,csggl}@imu.edu.cn Abstract In this paper, we first build a manually annotated named entity corpus of Mongolian. Then, we propose three morphological processing methods and study comprehensive features, including syllable features, lexical features, context features, morphological features and semantic features in Mongolian named entity recognition. Moreover, we also evaluate the influence of word clus- ter features on the system and combine all features together eventually. The experimental result shows that segmenting each suffix into an individual token achieves better results than delet- ing suffixes or using the suffixes as feature. The system based on segmenting suffixes with all proposed features yields benchmark result of F-measure=84.65 on this corpus. 1 Introduction Named Entity Recognition (NER) is a natural language processing (NLP) task that consists of finding names in an open domain text and classifying them among several predefined categories such as person, organization and location. It is an important tool in almost all NLP application areas, such as Ques- tion Answering, Machine Translation (Chen et al., 2013), Social Media Analysis, Semantic Search or Automatic Summarization. Since the MUC (Sundheim, 1995) and CoNLL (Sang, 2002) conferences, NER has drawn more and more attention in NLP community. Many NER systems have been developed for English and other language (Ratinov and Roth, 2009; Benajiba et al., 2010; Kravalov´a and ˇZabokrtsk`y, 2009). Machine learning based approach have been the predominant in these systems to achieve state-of-the-art results (Radford et al., 2015). As one of them, Conditional Random Fields (CRF) (Lafferty et al., 2001) was proved to an efficient classifier for NER. Recently, there are more and more concern about how to incorporate more latent sematic features into the NER system (Konkol et al., 2015). Therefore, word cluster IDs function as a non local feature to improve the performance of NER system (Turian et al., 2010; Zirikly and Diab, 2015). Mongolian is a widely spread language in the world. It is called classical Mongolian in China and called Cyrillic Mongolian in Mongolia and Russia. The classical Mongolian uses Uighur-script, while Cyrillic Mongolian uses Cyrillic-script. In this paper, we address the problem of NER for classical Mongolian. Compared with other languages, the research on Mongolian NER is still at its initial stage and many issues in Mongolian NER remain unsolved. As far as we know, there has been very little work in the area of NER in Mongolian. Tong (2013) only investigated Mongolian person name recognition. There is still no work publicly reported on recognition of Mongolian location and organization name. Moreover, there are no public available resources and tools for Mongolian NER. However, proper identification and classification of named entities are very crucial in Mongolian in- formation processing. Therefore, we propose a framework to develop resources and several methods for ∗Corresponding author …

Content truncated to 500 words