A head-annotated synchronous context-free grammar for hierarchical phrase-based translation

Hierarchical phrase-based translation is a state-of- the-art model. However, it does not consider the linguistic knowledge deeply as pure syntax-based model. In this paper, we propose a novel head-annotated synchronous context-free grammar for hierarchical phrase- based translation model. Our model contains several nonterminals in the synchronous rules. It can integrate some linguistic knowledge from head-modifier relationships of the words in source language. The experimental results obtain significant improvements in BLEU score over the NIST-05 task.