Extending Bilingual WordNet via Hierarchical Word Translation Classification

Extending Bilingual WordNet via Hierarchical Word Translation Classification* Tzu-yi Nien a, Tsun Ku b, Chung-chi Huang a, Mei-hua Chen a, and Jason S. Chang a aInstitute of Information Systems and Applications, National Tsing Hua University, HsinChu, Taiwan 300, R.O.C. {zinien, u901571, chen.meihua, jason.jschang}@gmail.com bInstitute for Information Industry, 11F, No. 106, Section 2, Heping East Road, Taipei, Taiwan 106, R.O.C. cujing@iii.org.tw Abstract. We introduce a method for learning to assign word senses to translation pairs. In our approach, this sense assignment or disambiguation problem is transformed into one on how to navigate through a sense network like WordNet aimed at distinguishing the more adequate senses from others. The method involves automatically constructing classification models for branching nodes in the network, and automatically learning to reject less probable senses, based on the translation characteristics of word senses and semantically- related word groups (e.g., lexicographer files) respectively. At run-time, translation pairs are expanded with their synonyms and sense ambiguity is resolved using a greedy algorithm choosing the most likely branches based on the trained classification models. Evaluation shows that our method significantly outperforms the strong baseline of assigning most frequent sense to the translation pairs and effectively determines suitable word senses for given translation pairs, suggesting the possibility of employing our method as a computer- assisted tool for speeding up the process of lexicography or of using our method to assist machine translation systems in word selection. Keywords: Word sense disambiguation, word translation classification, WordNet, machine-learning technique, maximum entropy model. Copyright 2009 by Tzu-yi Nien, Tsun Ku, Chung-chi Huang, Mei-hua Chen, and Jason S. Chang 1 Introduction Many words (e.g., plant) have different senses in different contexts (e.g., green plant and nuclear plant), usually leading to different translations in another language (e.g., 植物 and 工廠 respectively). On the other hand, different words (e.g., plant and factory) may express very similar meanings (e.g., the working place for industrial labors). Therefore, WordNet (Miller et al., 1990), a sense inventory encoding with semantic relatedness of words, has been a valuable resource in the field of natural language processing since its introduction. In WordNet, nominal, verbal, adjective, and adverbial words are grouped into synonym sets, or so-called synsets and synsets are interlinked with various semantic relations (e.g., hypernym, hyponym and etc). Its rich and well-defined lexical semantic relations have made WordNet an important knowledge source for various research areas: word sense disambiguation; computer-assisted language learning; information retrieval. The well-established lexical hierarchy residing in English WordNet has prompted researchers to construct WordNet-like sense inventory for other languages. Take the Chinese language for example. Efforts have been made on automatic construction and on manual 375 23rd Pacific Asia Conference on Language, Information and Computation, pages 375–384 translation from English WordNet into Chinese. Chinese translations in the latter case, however, may not sufficiently cover the scope. It would be more efficient and cost-effective if translations of various word senses could be automatically integrated from bilingual sources (e.g., dictionaries and phrase tables in machine translation systems). Consider the word “plant” with …

Content truncated to 500 words