The Transliteration from Alphabet Queries to Japanese Product Names

Copyright 2012 by Rieko Tsuji, Yoshinori Nemoto, Wimvipa Luangpiensamut, Yuji Abe, Takeshi Kimura, Kanako Komiya, Koji Fujimoto, and Yoshiyuki Kotani 26th Pacific Asia Conference on Language,Information and Computation pages 456–462 The Transliteration from Alphabet Queries to Japanese Product Names Rieko Tsujia, Yoshinori Nemotoa, Wimvipa Luangpiensamuta, Yuji Abea, Takeshi Kimuraa, Kanako Komiyaa, Koji Fujimotob, Yoshiyuki Kotania aDepartment of Computer and Information Science, Tokyo University of Agriculture and Technology / 2-24-16 Nakamachi Koganei-shi Tokyo JAPAN bTensor Consulting/ 2-10-1 Koujimachi Chiyoda-ku Tokyo JAPAN {Riekon.m, wimvipa, kittykimura}@gmail.com, 50012646127@st.tuat.ac.jp,wizdomowl@yahoo.co.jp, koji.fujimoto@tensor.co.jp, {kkomiya, kotani}@cc.tuat.ac.jp Abstract There are some cases where the non- Japanese buyers are unable to find products they want through the Japanese shopping Web sites because they require Japanese queries. We propose to transliterate the inputs of the non-Japanese user, i.e., search queries written in English alphabets, into Japanese Katakana to solve this problem. In this research, the pairs of the non- Japanese search query which failed to get the right match obtained from a Japanese shopping website and its transcribed word given by volunteers were used for the training data. Since this corpus includes some noise for transliteration such as the free translation, we used two different filters to filter out the query pairs that are not transliterlated in order to improve the quality of the training data. In addition, we compared three methods, BIGRAM, HMM, and CRF, using these data to investigate which is the best for the query transliteration. The experiment revealed that the HMM was the best. 1 Introduction In recent years, e-commerce is widely used throughout the world and it enables people to purchase products from foreign countries. However, sometimes it is not easy for foreign buyers to find the products they want because of the language difference. In our case, the alphabetic queries that are input by non-Japanese buyers should be translated into Japanese to show product pages which they want to find. There are many cases that non-Japanese people get no or wrong result from their research queries and they are classified into three cases. The first is the case where the non-Japanese people write Japanese product names in alphabets and we expected that this case would be solved by transliteration. The second is the case where non- Japanese people write English product names and this would be solved by translation. The final is the others, for example, the proper nouns such as the names of the animation characters etc., and the misspellings. Among them, we expected that the first case is the most frequent because 53.7% of them could be fully transliterated in the corpus. Hence, we propose the transliteration from the alphabetic queries to Japanese product names cf., from lunchbox to “ランチボックス (translation into English: lunchbox, pronunciation in Japanese: ranchibokkusu)”. Also, many researches about transliteration have been accomplished for clean data, however, as far as we know, there have been no research about transliteration for noisy query data. Thus, we investigated which method is the best for query transliteration, using the parallel data of the alphabetic …

Content truncated to 500 words