In this paper, we propose a novel dependency-based bracketing transduc- tion grammar for statistical machine translation, which converts a source sen- tence into a target dependency tree. Dif- ferent from conventional bracketing trans- duction grammar models, we encode tar- get dependency information into our lex- ical rules directly, and then we employ two different maximum entropy models to determine the reordering and combi- nation of partial dependency structures, when we merge two neighboring blocks. By incorporating dependency language model further, large-scale experiments on Chinese-English task show that our sys- tem achieves significant improvements over the baseline system on various test sets even with fewer phrases.

Click Here to View the Paper In XML Format