Experiments for various alignment models in Chinese-to-English SMT

In this paper we analyze and compare three basic word alignment models to acquire the useful knowledge and parameters for the statistical machine translation. A series of experiments are made and a very detailed analysis based on the experiments is given. According to the different features of the three models, the merits and demerits of the models are described and analyzed respectively. Finally, the word alignment probabilities and the phrases extracted using HMM Viterbi-alignment are used to develop the Chinese-to-English statistical machine translation system. The experimental results show that the translation quality will be increased with the preprocessing of the training data, joining of the dictionary and enlarging the translation grain appropriately.