Improving Web-Based Oov Translation Mining For Query Translation

Query translation is the most widely used approach for cross-language information retrieval (CLIR). The major challenge of query translation is translating Out-Of-Vocabulary (OOV) terms. This paper proposes three methods to improve OOV translation mining for query translation. Firstly, Co-occurrence information is utilized to extract topic words and to expand the source language query with the translations of topic words for collecting relevant bilingual snippets. Secondly, an improved frequency change measurement method which combines context dependency is utilized to extract valid OOV translation candidates from noisy, small-sized bilingual snippets. Thirdly, for choosing the proper translation, a combination model considering frequency-distance, surface patterns matching and phonetic features is proposed to pick out the appropriate translation(s). Experimental results show that this OOV translation mining approach for query translation has substantial CUR performance improvement.