Research on Deep Learning HMM Word Alignment

Abstract

Word alignment is an essential step for machine translation. According to the over-fitting of traditional word alignment model and the weakness for the context description, the deep learning hidden HMM word alignment combined a multi-layer neural network with an undirected probabilistic graph model, use the similarity of the word and context information word alignment to be a more precise modeling. Experimental results show that, compared with the reference system, this model can significantly improve the effect of word alignment, which is applicable.