A unigram orientation model for statistical machine translation

In this paper, we present a unigram segmentation model for statistical machine translation where the segmentation units are blocks: pairs of phrases without internal structure. The segmentation model uses a novel orientation component to handle swapping of neighbor blocks. During training, we collect block unigram counts with orientation: we count how often a block occurs to the left or to the right of some predecessor block. The orientation model is shown to improve translation performance over two models: 1) no block re-ordering is used, and 2) the block swapping is controlled only by a language model. We show experimental results on a standard Arabic-English translation task.