Bayesian Subtree Alignment Model based on Dependency Trees

Word sequential alignment models work well for similar language pairs, but they are quite inadequate for distant language pairs. It is difficult to align words or phrases of distant languages with high accuracy without structural information of the sentences. In this paper, we propose a Bayesian subtree alignment model that incorporates dependency relations between subtrees in dependency tree structures on both sides. The dependency relation model is a kind of tree-based reordering model, and can handle non-local reorderings, which sequential word-based models often cannot handle properly. The model is also capable of handling multilevel structures, making it possible to find many-to-many correspondences automatically without any heuristic rules. The size of the structures is controlled by nonparametric Bayesian priors. Experimental alignment results show that our model achieves 3.5 points better alignment error rate for English-Japanese than the word sequential alignment model, thereby verifying that the use of dependency information is effective for structurally different language pairs.