Generating Paired Transliterated-cognates Using Multiple Pronunciation Characteristics from Web corpora

Generating Paired Transliterated-cognates Using Multiple Pronunciation Characteristics from Web Corpora Jin-Shea KUO 1, 2 Ying-Kuei YANG 2 1Chung-Hwa Telecommunication Laboratories, 12, Lane 551, Sec. 5, Min-Tsu Rd., Yang-Mei, 326, Taoyuan, Taiwan 2Electrical Engineering Dept., National Taiwan University of Science and Technology, 43, Sec. 4, Keelung Rd., 106, Taipei, Taiwan jskuo@cht.com.tw ykyang@mouse.ee.ntust.edu.tw Abstract A novel approach to automatically extracting paired transliterated-cognates from Web corpora is proposed in this paper. One of the most important issues addressed is that of taking multiple pronunciation characteristics into account. Terms from various languages may pronounce very differently. Incorporating the knowledge of word origin may improve the pronunciation accuracy of terms. The accuracy of generated phonetic information has an important impact on term transliteration and hence transliterated-term extraction. Transliterated-term extraction is a fundamental task in natural language processing to extract paired transliterated-terms in studying term transliteration. An experiment on transliterated-term extraction from two kinds of Web resources, Web pages and anchored texts, has been conducted and evaluated. The experimental results show that many transliterated-term pairs, which cannot be extracted using the approach only exploiting English pronunciation characteristics, have been successfully extracted using the proposed approach in this paper. By taking multiple language-specific pronunciation transformations into account may further improve the output of the transliterated-term extraction. 1. Introduction M e dia (Inte rne t/Bo o ks /N E W S/...) in E nglis h/C hine se /F re nc h/J apane s e/ G e rm an/K o re an/Italian.... M e dia C hanne l ( Trans latio n/Trans lite ratio n) Te xts In O riginal L angua ge s M ixe d- L anguage Te xt C o rpora P us an P us an 부 산 ( K o re an) 釜 山 Fans 粉 絲 S fo rza (Italian) Sfo rza 史 佛 拉 Sfo rza 三 和 S anw a S anw a サ ン ワ (J apane s e) 湖 北 H ube i H ube i 湖 北 (C hines e) Fans Fans (E nglis h) Figure 1. A conceptual flow for transliterating terms in multi-languages into Chinese. Machine transliteration is one part of machine translation. Term transliteration in machine transliteration addresses the problem of converting terms in one language into their phonetic equivalents in the other language via spoken form. It is especially concerned with proper nouns, such as personal names, place names and organization names. Transliterated-term extraction, which is a fundamental task in studying term transliteration, has been focused on producing a large quantity of paired transliterated-cognates in order to observe various relations between cognate pairs. A transliteration lexicon, which is composed of many transliterated-term pairs, is important to the researches on term transliteration. Virga (2003) and Meng (2001) have explored relationships trained from a transliteration corpus collected manually to cross-language information retrieval and spoken document retrieval, respectively. However, it is time- and labor-consuming to prepare a transliteration PACLIC 18, December 8th-10th, 2004, Waseda University, Tokyo - 275 - lexicon manually. It will be helpful if a …

Content truncated to 500 words