Korean-Chinese word translation using Chinese character knowledge

Korean-to-Chinese Word Translation using Chinese Character Knowledge Yuanmei Lu lu@nlp.ist.i.kyoto-u.ac.jp Toshiaki Nakazawa nakazawa@pa.jst.jp Japan Science and Technology Agency, Tokyo, 102-8666, Japan Sadao Kurohashi kuro@nlp.ist.i.kyoto-u.ac.jp Graduate School of Informatics, Kyoto University, Kyoto, 606-8501, Japan Abstract In this paper, we present a way of translating Korean words to Chinese using Chinese charac- ter information. A mapping table of Korean and Chinese characters is constructed and used to obtain possible combinations as translation candidates. The candidates are ranked by the combination score which accounts for the possibility of the character combination and context similarity score, which indicates contextual information among words. Parallel resources like Wikipedia aligned data or Wiktionary data are used for preliminary translation and also used during ranking candidates. 1 Introduction The quality of the statistical machine translation heavily correlates with the amount of parallel corpora used, and improving the lexical coverage of the parallel corpora plays an important role in reducing the number of out-of-vocabulary (OOV) words. However, the vocabularies of languages keep growing over time, especially in the case of technical terms. It is impossible to cover all the newly appearing words by augmenting the parallel corpora and therefore we need to prepare bilingual dictionaries for the new words, or translate them separately, for example, using the transliteration technique. There are some parallel dictionaries available for limited language pairs and limited do- mains. In addition, we can extract parallel resources from Wikipedia. It offers hyper-linked pages of the same topic in different languages, and the title pairs of the linked pages can be used as a parallel dictionary. However, the coverage is not sufficient for both cases especially for technical terms. In some cases, titles aligned by hyper-linked pages can not be seen as exact translations of each other since many of them may just be related to similar concepts. Another limitation in this domain is that although there may exist enough resources between English and other languages, there are very few resources between the non-English languages, such as Korean, Chinese and Japanese. Korean, Chinese and Japanese use Chinese characters. In Japan, they use Kanji, which originated from China, while Korean uses Sino-Korean vocabularies, in which characters (Hangul) can be converted to corresponding Chinese characters (Hanzi). Just as Japanese writ- ing includes both Kana and Kanji, Korean contains two kinds of characters–Hangul and Hanja. Hangul are Korean characters that are most commonly used and seen, and many Sino-Korean words have their Hanja writings–Hanja. For the sake of clarity refer to Table 1 which contains a few examples of Hangul and Hanja. Proceedings of MT Summit XV, vol.1: MT Researchers' Track Miami, Oct 30 - Nov 3, 2015 | p. 256 Hangul (Korean) 㥙㩁 㕤㗴 Hanja (Korean) 愛情 勞動 Kanji (Japanese) 愛情 労働 Hanzi (Chinese) 爱情 劳动 Table 1: Difference of characters written in Hangul, Hanja, Kanji and Hanzi In reality, Hanja writing is seldom used in modern Korean language. It is used in sev- eral technical writings to emphasize the definition of certain words. It is not hard to …

Content truncated to 500 words