Disambiguation of Single Noun Translations Extracted from Bilingual Comparable Corpora

s of papers of four academic societies, namely Japan Architecture Society (JAS), Institute of Electric Engineering (IEE), Institute of Electronics and Communication Engineering (IECE), and Information Processing Society of Japan (IPSJ), published in Japan. Numbers of abstracts of each of these corpora are shown in Table 1. Parts of these bilingual corpora are parallel. The percentages of parallel text in the four corpora are also shown in Table 1 where “parallel texts ratio” is defined as follows. Parallel texts ratio = (ParaJ × ParaE) where ParaJ=(Number of parallel abstracts)/(Number of the whole Japanese abstracts) ParaE=((Number of parallel abstracts)/(Number of the whole English abstracts) Table 1: Corpora used in our experiment and parallel texts ratio Academic society name Number of Japanese articles Number of English articles Parallel texts ratio JAS 55,715 50,236 0.95 IEE 18,008 399 0.12 IECE 86,346 33,076 0.59 IPSJ 26,815 11,860 0.65 5.3. Morphological Analysis and POS Tagging We use morphological analyzer Chasen (Matsumoto 1997) for Japanese corpora, and Brill's tagger (Brill 1994) for English corpora to extract terms. In order to process Japanese corpora, Chasen segments out words from Japanese sentences, and at the same time it assigns a part of speech tag (henceforth we call “POS tag”) to each word. Then we extract an uninterrupted sequence of nouns as a candidate term. As easily known they are single nouns or compound nouns. On the other hand, to process English corpora, Brill’s tagger only assigns a POS tag to each word in English sentences. Then we extract, as a candidate term, the following four types of linguistic patterns written in regular expressions defined in Perl language and their combinations using assigned POS tags. (1) [noun] (2) adjective (noun | adjective) noun (3) noun “of” noun (4) foreign word