A CLIR-oriented OOV translation mining method from bilingual webpages

摘要

Translating unknown terms is a major bottleneck for cross-language IR. An effective solution to relevant webpage detection, translation extraction with correct boundaries, and candidate translation ranking is proposed. Topic word translations are used to expand the source query and collect bilingual search engine snippets. Then an improved Frequency Change Measurement method is used to extract valid candidates from noisy, small bilingual corpora. To choose the translation, frequency-distance, surface patterns and phonetic features are used to pick out the correct translation. Experimental results show an impressive performance for unknown term translation mining.