An improved maximal entropy based bracketing transduction grammar translation model with ensemble learning
- 软件学院－已发表论文 
With an important characteristic that using discriminative lexicalized reordering model to capture the phrase movement during translation, the maximum entropy based bracketing translation grammar (MEBTG) has become one of the research hotspots of statistical machine translation in recent years. However, the research of this model is far from mature. Specifically, MEBTG system tends to suffer from the problems related to over-fitting of the reordering examples. To solve this problem, we propose to apply ensemble learning framework to improve discriminative reordering model of MEBTG system. In the specific implementation, we first respectively try bagging and cross-validation to construct multiple basic classifiers, and then investigate two integration methods to combine the results of these classifiers. The experimental results on large-scale data set demonstrate the effectiveness of our method. ? 2014 Binary Information Press.