Adjacent reordering phrase-based translation models

We study a number of statistical phrase-based translation models that are constrained by adjacent reordering. Under this constraint, we derive polynomial decoding algorithms respectively for basic and Koehn´s phrase-based models, with an implication that many statistical phrase-based models may also be decodable in polynomial time. Using NIST as a metric, we show that the presented decoding algorithms can achieve a relative improvement of about 2.7% over Pharaoh under the same experimental conditions.