Machine translation using bilingual term entries extracted from parallel texts

Machine Translation Using Bilingual Term Entries Extracted from Parallel Texts Tatsuya Izuha Corporate Research & Development Center, Toshiba Corporation 1, Komukai Toshiba-cho, Saiwai-ku, Kawasaki, 210-8582 JAPAN tatsuya.izuha@toshiba.co.jp Abstract Patent summaries are machine-translated using bilingual term entries extracted from parallel texts for evaluation. The result shows that bilingual term entries extracted from 2,000 pairs of parallel texts which share a specific domain with the input texts introduce more improvements than a technical term dictionary with 38,000 entries which covers a broader domain. The result also shows that only 10 pairs of parallel texts found by similar document retrieval have comparable effects to the technical term dictionary, suggesting that parallel texts to be used do not need to be classified into fields prior to term extraction. Keywords machine translation, parallel text, bilingual lexicon, technical term, similar document retrieval 1. Introduction Bilingual technical term lexicons are indispensable for better translation of technical documents. Since technical term dictionaries developed for machine translation systems generally cover broad domains such as "business" and "medicine," they often fail to contain those terms used only in some specific domains, and sometimes incorrectly translate those words according to more specific domains. However, technical term dictionaries for more specific domains are difficult to build and maintain because there exist so many technical domains and the number is constantly increasing. In recent years, various methods have been proposed to build bilingual lexicons (semi-)automatically from parallel texts (Kumano & Hirakawa 1992; Melamed 1996; Smaja et al. 1996; Haruno et al. 1996; Kitamura & Matsumoto 1996). Many of these methods, however, require humans with expertise in the domain to check the final output entailing considerable cost on the part of humans. In this paper, bilingual term entries automatically extracted from parallel texts are directly used by a machine translation system to reduce such cost dramatically. This paper is organized as follows: In the next section, the overview of our approach is presented. In Sections 3, 4 and 5, each component is described in detail. In Sections 6 and 7, evaluations of patent summaries and discussions on the results are given. Section 8 concludes the paper with final remarks. 2. Overview of our approach Figure 1 shows the flow of Japanese-English translation using bilingual term entries extracted from parallel texts. When a Japanese text to be translated into English is given to the system, the system retrieves parallel texts similar to the input text from the parallel text database. Then, it extracts bilingual term entries consulting the Japanese-English bilingual dictionary and chooses those entries that should be registered into the user dictionary used for machine translation. The system then translates the input text with the newly registered entries in the user dictionary. The resulting translation is compared with manual translation for evaluation. 3. Similar document retrieval Ideally, bilingual term entries should be extracted from texts which belong to the same domain as the input text. However, parallel texts classified into specific domains are not always available to the user. To deal with this problem, …

Content truncated to 500 words