Generating Abbreviations for Chinese Named Entities Using Recurrent Neural Network with Dynamic Dictionary

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 721–730, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Generating Abbreviations for Chinese Named Entities Using Recurrent Neural Network with Dynamic Dictionary Qi Zhang, Jin Qian, Ya Guo, Yaqian Zhou, Xuanjing Huang Shanghai Key Laboratory of Data Science School of Computer Science, Fudan University Shanghai, P.R. China {qz, jqian12, yguo13, zhouyaqian, xjhuang}@fudan.edu.cn Abstract Chinese named entities occur frequently in formal and informal environments. Various approaches have been formalized the problem as a sequence labelling task and utilize a character-based methodology, in which char- acter is treated as the basic classification unit. One of the main drawbacks of these methods is that some of the generated abbreviations may not follow the conventional wisdom of Chinese. To address this problem, we pro- pose a novel neural network architecture to perform task. It combines recurrent neural network (RNN) with an architecture determin- ing whether a given sequence of characters can be a word or not. For demonstrating the effectiveness of the proposed method, we evaluate it on Chinese named entity generation and opinion target extraction tasks. Experi- mental results show that the proposed method can achieve better performance than state-of- the-art methods. 1 Introduction Abbreviations of Chinese named entities are fre- quently used on different kinds of environments. Along with the development of social media, this kinds of circumstance occurs more frequently. Un- like western languages such as English, Chinese does not insert spaces between words or word forms that undergo morphological alternations. Hence, most of the Chinese natural language processing methods assume a Chinese word segmenter is used in a pre-processing step to produce word-segmented Chinese sentences as input. However, if the Chinese word segmenter produces erroneous output, the quality of these methods will be degraded as a direct result. Moreover, since the word segmenter may split the targets into two individual words, many methods adopted character-based methodologies, such as methods for named entity recognition (Wu et al., 2005), aspect-based opinion mining (Xu et al., 2014), and so on. Through character-based methodology, most of the previous abbreviation generation approaches have been formalized as sequence labelling prob- lem. Chinese characters are treated as the basic classification unit and are classified one by one. In these methods, dictionaries play important effect in constructing features and avoiding meaningless outputs. Various previous works have demonstrated the significant positive effectiveness of the external dictionary (Zhang et al., 2010). However, because these external dictionaries are usually static and pre- constructed, one of the main drawbacks of these methods is that the words which are not included in the dictionaries cannot be well processed. This issue has also been mentioned by numerous previous works (Peng et al., 2004; Liu et al., 2012). Hence, understanding how Chinese words are constructed can benefit a variety of Chinese NLP tasks to avoid meaningless output. For example, to generate the abbreviation for a named entity, we can use a binary classifier to determine whether a character should be …

Content truncated to 500 words