Semi-Supervised Lexicon Mining from Parenthetical Expressions in Monolingual Web Pages

Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the ACL, pages 424–432, Boulder, Colorado, June 2009. c⃝2009 Association for Computational Linguistics Semi-Supervised Lexicon Mining from Parenthetical Expressions in Monolingual Web Pages Xianchao Wu† Naoaki Okazaki† Jun’ichi Tsujii†‡ †Computer Science, Graduate School of Information Science and Technology, University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan ‡School of Computer Science, University of Manchester National Centre for Text Mining (NaCTeM) Manchester Interdisciplinary Biocentre, 131 Princess Street, Manchester M1 7DN, UK {wxc, okazaki, tsujii}@is.s.u-tokyo.ac.jp Abstract This paper presents a semi-supervised learn- ing framework for mining Chinese-English lexicons from large amount of Chinese Web pages. The issue is motivated by the ob- servation that many Chinese neologisms are accompanied by their English translations in the form of parenthesis. We classify par- enthetical translations into bilingual abbrevi- ations, transliterations, and translations. frequency-based term recognition approach isA applied for extracting bilingual abbreviations. A self-training algorithm is proposed for min- ing transliteration and translation lexicons. In which, we employ available lexicons in terms of morpheme levels, i.e., phoneme correspon- dences in transliteration and grapheme (e.g., suffix, stem, and prefix) correspondences in translation. The experimental results verified the effectiveness of our approaches. 1 Introduction Bilingual lexicons, as lexical or phrasal parallel corpora, are widely used in applications of multi- lingual language processing, such as statistical ma- chine translation (SMT) and cross-lingual informa- tion retrieval. However, it is a time-consuming task for constructing large-scale bilingual lexicons by hand. There are many facts cumber the manual de- velopment of bilingual lexicons, such as the contin- uous emergence of neologisms (e.g., new technical terms, personal names, abbreviations, etc.), the dif- ficulty of keeping up with the neologisms for lexi- cographers, etc. In order to turn the facts to a better way, one of the simplest strategies is to automati- cally mine large-scale lexicons from corpora such as the daily updated Web. Generally, there are two kinds of corpora used for automatic lexicon mining. One is the purely monolingual corpora, wherein frequency-based expectation-maximization (EM, refer to (Dempster et al., 1977)) algorithms and cognate clues play a central role (Koehn and Knight, 2002). Haghighi et al. (2008) presented a generative model based on canonical correlation analysis, in which monolin- gual features such as the context and orthographic substrings of words were taken into account. The other is multilingual parallel and comparable cor- pora (e.g., Wikipedia1), wherein features such as co- occurrence frequency and context are popularly em- ployed (Cheng et al., 2004; Shao and Ng, 2004; Cao et al., 2007; Lin et al., 2008). In this paper, we focus on a special type of com- parable corpus, parenthetical translations. The issue is motivated by the observation that Web pages and technical papers written in Asian languages (e.g., Chinese, Japanese) sometimes annotate named enti- ties or technical terms with their translations in En- glish inside a pair of parentheses. This is considered to be a traditional way to annotate new terms, per- sonal names or other named entities with …

Content truncated to 500 words