Stacking for Statistical Machine Translation⇤

We propose the use of stacking, an ensem-ble learning technique, to the statistical machine translation (SMT) models. A diverse ensem-ble of weak learners is created using the same SMT engine (a hierarchical phrase-based sys-tem) by manipulating the training data and a strong model is created by combining the weak models on-the-fly. Experimental results on two language pairs and three different sizes of train-ing data show significant improvements of up to 4 BLEU points over a conventionally trained SMT model.