Improving Bilingual Lexicon Extraction Performance from Comparable Corpora via Optimizing Translation Candidate Lists

摘要

In this paper, we propose a novel method to optimize translation candidate lists derived from window-based approach for the task of bilingual lexicon extraction. The optimizing process consists of two cross-comparisons between 1th translation candidate of each target word, and between set of all the 1th candidates and that of each word's 2th to Nth ones. Experiment results demonstrate that the proposed method leads to a significant improvement on accuracy over window-based approach in bilingual lexicon extraction from both English-Chinese and Chinese-English comparable corpora.