Chinese Named Entity Abbreviation Generation Using First-Order Logic

International Joint Conference on Natural Language Processing, pages 320–328, Nagoya, Japan, 14-18 October 2013. Chinese Named Entity Abbreviation Generation Using First-Order Logic Huan Chen, Qi Zhang, Jin Qian, Xuanjing Huang School of Computer Science Fudan University Shanghai, P.R. China {12210240054, qz, 12110240030, xjhuang}@fudan.edu.cn Abstract Normalizing named entity abbreviations to their standard forms is an important preprocessing task for question answering, entity retrieval, event detection, microblog processing, and many other applications. Along with the quick expansion of mi- croblogs, this task has received more and more attentions in recent years. In this pa- per, we propose a novel entity abbreviation generation method using first-order logic to model long distance constraints. In order to reduce the human effort of man- ual annotating corpus, we also introduce an automatically training data construction method with simple strategies. Experi- mental results demonstrate that the pro- posed method achieves better performance than state-of-the-art approaches. 1 Introduction Twitter and other social media services have received considerable attentions in recent years. Users provide hundreds of millions microblogs through them everyday. The informative data has been relied on by many applications, such as sentiment analysis (Jiang et al., 2011; Meng et al., 2012), event detection (Sakaki et al., 2010; Lin et al., 2010), stock market predication (Bollen et al., 2011), and so on. However, due to the constraint on the length of characters, abbreviations frequently occur in microblogs. According to a statistic, approximately 20% of sentences in news articles have abbreviated words (Chang and Lai, 2004). The frequency of abbreviation has become even more popular along with the rapid increment of user generated contents. Without pre-normalizing these abbreviations, most of the natural language processing systems may heavily suffer from them. The goal of entity abbreviation generation is to produce abbreviated equivalents of the original en- tities. Table 1 shows several examples of entities and their corresponding abbreviations. A few of approaches have been done on this task. Li and Yarowsky (Li and Yarowsky, 2008b) introduced an unsupervised method used to extract phrases and their abbreviation pair using parallel dataset and monolingual corpora. Xie et al. (2011) proposed to use weighted bipartite graph to extract defi- nition and corresponding abbreviation pairs from anchor texts. Since these methods rely heavily on lexical/phonetic similarity, substitution of charac- ters and portion may not be correctly identified through them. Yang et al. (2009) studied the Chinese entity name abbreviation problem. They formulated the abbreviation task as a sequence labeling problem and used the conditional random fields (CRFs) to model it. However the long distance and global constraint can not be easily modeled thorough CRFs. Entity Abbr. 北京大学 北大 (Peking University) 中国石油天然气集团公司 中石油 (China National Petroleum Corporation) 中国国际航空公司 国航 (Air China) Table 1: Abbreviation examples To overcome these limitations, in this paper, we propose a novel entity abbreviation generation method, which combines first-order logic and rich linguistic features. To the best of our knowledge, our approach is the first work of using first-order logic for this entity abbreviation. Abbreviation generation is converted to character deletion …

Content truncated to 500 words