Iterative reordering and word alignment for statistical MT

In this study I investigate reordering for statistical machine translation. In particular I focus on improving both reordering rules and word alignment, by iterating these two steps, since they both depend on each other. I also investigate how three reordering approaches used at decoding time, monotone decoding, a simple distortion penalty, and a lexicalized reorering model, interplay with the reordering rules. No consistent improvements were seen by performing the iterative reordering and alginment, the overall performance varied quite a lot between different metrics though. But the overall best performance were found by performing two iterations of reordering, and using a distortion penalty.