Predicting Chinese Abbreviations with Minimum Semantic Unit and Global Constraints

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1405–1414, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Predicting Chinese Abbreviations with Minimum Semantic Unit and Global Constraints Longkai Zhang Li Li Houfeng Wang Xu Sun Key Laboratory of Computational Linguistics (Peking University) Ministry of Education, China zhlongk@qq.com, {li.l,wanghf,xusun}@pku.edu.cn Abstract We propose a new Chinese abbreviation prediction method which can incorporate rich local information while generating the abbreviation globally. Different to previ- ous character tagging methods, we intro- duce the minimum semantic unit, which is more fine-grained than character but more coarse-grained than word, to capture word level information in the sequence labeling framework. To solve the “character dupli- cation” problem in Chinese abbreviation prediction, we also use a substring tagging strategy to generate local substring tagging candidates. We use an integer linear pro- gramming (ILP) formulation with various constraints to globally decode the final ab- breviation from the generated candidates. Experiments show that our method outper- forms the state-of-the-art systems, without using any extra resource. 1 Introduction Abbreviation is defined as a shortened description of the original fully expanded form. For example, “NLP” is the abbreviation for the corresponding full form “Natural Language Processing”. The ex- istence of abbreviations makes it difficult to iden- tify the terms conveying the same concept in the information retrieval (IR) systems and machine translation (MT) systems. Therefore, it is impor- tant to maintain a dictionary of the prevalent orig- inal full forms and the corresponding abbrevia- tions. Previous works on Chinese abbreviation gen- eration focus on the sequence labeling method, which give each character in the full form an extra label to indicate whether it is kept in the abbre- viation. One drawback of the character tagging strategy is that Chinese characters only contain limited amount of information. Using character- based method alone is not enough for Chinese ab- breviation generation. Intuitively we can think of a word as the basic tagging unit to incorporate more information. However, if the basic tagging unit is word, we need to design lots of tags to repre- sent which characters are kept for each unit. For a word with n characters, we should design at least 2n labels to cover all possible situations. This re- duces the generalization ability of the proposed model. Besides, the Chinese word segmentation errors may also hurt the performance. Therefore we propose the idea of “Minimum Semantic Unit” (MSU) which is the minimum semantic unit in Chinese language. Some of the MSUs are words, while others are more fine-grained than words. The task of selecting representative characters in the full form can be further broken down into se- lecting representative characters in the MSUs. We model this using the MSU-based tagging method, which can both utilize semantic information while keeping the tag set small. Meanwhile, the sequence labeling method per- forms badly when the “character duplication” phe- nomenon exists. Many Chinese long phrases con- tain duplicated characters, which we refer to as the “character …

Content truncated to 500 words