Unsupervised Japanese-Chinese Opinion Word Translation using Dependency Distance and Feature-Opinion Association Weight

Proceedings of COLING 2012: Technical Papers, pages 1503–1518, COLING 2012, Mumbai, December 2012. Unsupervised Japanese-Chinese Opinion Word Translation Using Dependency Distance and Feature-Opinion Association Weight Guo-Hau Lai, Ying-Mei Guo and Richard Tzong-Han Tsai* Department of Computer Science and Engineering, Yuan Ze University 135 Yuan-Tung Road, Chungli, Taoyuan, Taiwan, R.O.C. *Corresponding Author {s986005, s996148}@mail.yzu.edu.tw, thtsai@saturn.yzu.edu.tw ABSTRACT Online shoppers depend on customer reviews when evaluating products or services. However, in the international online marketplace, reviews in a user’s language may not be available. Translation of online customer reviews is therefore an important service. A crucial aspect of this task is translating opinion words, key words that capture the reviewers’ sentiments. This is challenging because opinion words often have multiple translations. We propose an unsupervised opinion word translation disambiguation scoring method using dependency distance and feature-opinion association as weighting factors. The scores of an opinion word’s translation and its surrounding words’ translations are estimated using Google snippets. We focus on Japanese-Chinese translation of hotel reviews from Rakutan Travel, using the 10 most common polysemous Japanese opinion words to evaluate system performance. Results show our weighting factors significantly improve translation accuracy compared to Google and Excite. KEYWORDS : Opinion Word Translation Disambiguation, Dependency Distance, Feature-Opinion Association 1503 1 Introduction The development of Web 2.0 has made it easier for internet users to post their reviews or comments about products or services on structured websites. Online shoppers are increasingly likely to look at these reviews before deciding on a purchase. In recent years, the research field of sentiment analysis has focused on analyzing this form of textual-information, particularly opinions or sentiments expressed by internet users. However, given the international nature of the web and online shopping, opinions in a user’s mother language may not be available. Translation of online customer reviews is therefore an important service sought after in many markets. A crucial aspect of customer review translation is translating opinion words, key words that capture the sentiments. At present, machine translation (MT) systems can translate whole sentences or even complete paragraphs. This not a trivial task, however, because opinion words usually have multiple possible translations and the MT systems have low accuracy on polysemous words (Carpuat & Wu, 2005). This paper proposes an unsupervised method of selecting the most appropriate Chinese translation for an opinion word in a given Japanese sentence. Candidate translations are retrieved from a bilingual dictionary. Consider the following Japanese sentence: 綺麗な夜景とともに食事を楽しむことができました。 (I was able to enjoy a nice meal with a beautiful night view.) The target opinion word 綺麗 has three candidate translations: 漂亮 (beautiful), 乾淨 (clean), and 清楚 (clear) in Chinese. In this example, the most appropriate translation is 漂亮 (beautiful). This disambiguation problem is known as Word Translation Disambiguation (WTD). One way to solve the WTD problem is to calculate the sum of association scores of pairs among translation of the target word and all its surrounding words’ translations, and then select the one with the highest score. However, since the different surrounding words have different amounts of …

Content truncated to 500 words