Extending phrased-based SMT with dependency based reordering model

Phrase based SMT (SMT) is state-of-the-art model, especially, on large scale training data. However, the capacity of its reordering is based on simple flat features which cannot handle complicated reordering cases. To address this issue, this paper proposes a dependency based reordering model. By doing this, we exploit the way how to utilize the structured linguistic analysis information in source language. With simple parameter estimation and small scale of training mode, the performance of the phrase based SMT is improved.