LiveTrans: Translation Suggestion for Cross-Language Web Search from Web Anchor Texts and Search Results

LiveTrans: Translation Suggestion for Cross-Language Web Search from Web Anchor Texts and Search Results Wen-Hsiang Lu1,2, Lee-Feng Chien1 and Hsi-Jian Lee2 1. Institute of Information Science, Academia Sinica, Taiwan, ROC 2. Department of Computer Science and Information Engineering, National Chiao Tung University, Taiwan, ROC {whlu, lfchien}@iis.sinica.edu.tw, {whlu, hjlee}@csie.nctu.edu.tw Abstract In this paper we will present a system, called LiveTrans, which can generate translation suggestions for given user queries and provide an English-Chinese cross-language search service for the retrieval of both Web pages and images. The system effectively utilizes two kinds of Web resources: anchor texts and search results. The developed anchor-text-based and search-result-based methods are complementary in the precision and coverage rates and promising in extracting translations of unknown query terms that were not included in general-purpose translation dictionaries. Experimental results demonstrate the feasibility of the system. 1. Introduction To deal with automatic construction of translation lexicons, conventional research on machine translation (MT) [3] and cross-language information retrieval (CLIR) [1, 5, 7, 10, 13, 18] has generally used statistical techniques to automatically extract word translations from domain-specific parallel/comparable bilingual texts, such as bilingual newspapers [4, 11, 12, 20, 21]. However, only a certain set of their translations can be extracted through corpora with limited domains. In our research, we are interested in extracting translations of technical terms and proper names in diverse subjects, which are especially needed in performing CLIR services for Web users, e.g., “Hussein” (海珊/哈珊/侯賽因), “SARS” (嚴重急性呼吸道症候群). Existing CLIR systems usually rely on bilingual dictionaries for query translation [1, 13, 15]. Unfortunately, our analysis of Dreamer query log collected in Taiwan (see Section 3.1) showed that 74% of the 20,000 high frequent Web queries can not be found in general-purpose English-Chinese dictionaries (they are called unknown terms in this paper). How to automatically find translations for unknown terms, therefore, has become a major challenge for cross-language Web search. Different from previous works, we focus on investigating new approaches to mining multilingual Web resources [19]. We have proposed a novel approach to extracting translations of Web queries through the mining of Web anchor texts and link structures [16, 17]. An anchor text is the descriptive part of an out-link of a Web page used to provide a brief description of the linked page. A variety of anchor texts in multiple languages might link to the same pages from all over the world. For example, Figure 1 shows a typical example, in which there are a variety of anchor texts in multiple languages linking to the Yahoo! from all over the world. Such a bundle of anchor texts pointing together to the same page is called an anchor-text set. Web anchor-text sets may contain similar description texts in multiple languages. Thus, for an unknown term appearing in some anchor-text sets, it is likely that its corresponding target translations appear together in the same anchor-text sets. However, discovering translation knowledge from the Web has not been fully explored. In this paper, we intend to investigate another kind of Web resource, search …

Content truncated to 500 words