Data Selection and Smoothing in an Open-Source Systemfor the 2008 NIST Machine Translation Evaluation

摘要

This paper gives a detailed description of a statistical machinetranslation system developed for the 2008 NIST open MT eval-uation. The system is based on the open source toolkit Moseswith extensions for language model rescoring in a second pass.Significant improvements were obtained with data selectionmethods for the language and translation model. An improve-ment of more than 1 point BLEU on the test set was achieved bya continuous space language model which performs the proba-bility estimation with a neural network. The described systemhas achieved a very good ranking in the 2008 NIST open MTevaluation.