Fast and Accurate Bilingual Lexicon Induction via Matching Optimization

Zewen Chi, Heyan Huang, Shenjian Zhao, Heng-Da Xu, Xianling Mao

Abstract: Most recent state-of-the-art approaches are proposed to utilize the pre-trained word embeddings for bilingual lexicon induction. However, the word embeddings introduce noises for both frequent and rare words. Especially in the case of rare words, embeddings of which are always not well learned due to their low occurrence in the training data. In order to alleviate the above problem, we propose BLIMO, a simple yet effective approach for automatic lexicon induction. It does not introduce word embeddings but converts the lexicon induction problem into a maximum weighted matching problem, which could be efficiently solved by the matching optimization with greedy search. Empirical experiments further demonstrate that our proposed method outperforms state-of-the-arts baselines greatly on two standard benchmarks.