Statistical machine translation model based on a synchronous tree-substitution grammar

Abstract

A translation model based on synchronous tree-substitution-grammar is presented in this paper. It can elegantly model the global reordering and discontinuous phrases. Furthermore, it can learn non-isomorphic tree-to-tree mappings. Experimental results on two different data sets show that the proposed model significantly outperforms the phrase-based model and the model based on synchronous context-free grammar. © by Institute of Software, the Chinese Academy of Sciences. All rights reserved.