Large-scale Expected BLEU Training of Phrase-based Reordering Models

Recent work by Cherry (2013) has shown that directly optimizing phrase-based reordering models towards BLEU can lead to significant gains. Their approach is limited to small training sets of a few thousand sentences and a similar number of sparse features. We show how the expected BLEU objective allows us to train a simple linear discriminative reordering model with millions of sparse features on hundreds of thousands of sentences resulting in significant improvements. A comparison to likelihood training demonstrates that expected BLEU is vastly more effective. Our best results improve a hierarchical lexicalized reordering baseline by up to 2.0 BLEU in a single-reference setting on a French-English WMT 2012 setup.