Most of the existing methods for bilingual word embedding only considershallow context or simple co-occurrence information. In this paper, we proposea latent bilingual sense unit (Bilingual Sense Clique, BSC), which is derivedfrom a maximum complete sub-graph of pointwise mutual information based graphover bilingual corpus. In this way, we treat source and target words equallyand a separated bilingual projection processing that have to be used in mostexisting works is not necessary any more. Several dimension reduction methodsare evaluated to summarize the BSC-word relationship. The proposed method isevaluated on bilingual lexicon translation tasks and empirical results showthat bilingual sense embedding methods outperform existing bilingual wordembedding methods.
展开▼
