An Exploration on Improving Statistical Machine Translation Performance by Using Post-Editing Information

Manually Post-editing (PE) is a traditional and effective way of improving machine translation outputs. However, it is costly and time consuming. In this paper, manually PE knowledge including word and phrase revision information is used for updating statistical machine translation (SMT) model, the updated system can avoidsimilar mistakes and achieve better translation performance. A number of SMT model compatible features are extracted from PE process, and then an updating process is implemented to combine such PE knowledge into the original SMT model. Experiments on Chinese to English translation are carried out. Results show that our approach could improve the performance of baseline SMT system. Additionally, the updated SMT model has the capability of generating user expected outputs through PE information combination process.