Research On Statistical Method Of Machine Translation Evaluation

Recently, research on machine translation evaluation (MTE) has become hot topic on natural language processing research, as a significant guideline to the development of translation systems. The design and implementation of reliable translation evaluation and development of the translation system, is a very time-consuming and labor-intensive-job, with high complexity. As a result, some researchers make in-depth study on translation quality evaluation.This dissertation aims at three main aspects:First, for improving the performance of MTE methods, a translation evaluation method base-on language modeling is proposed according to the property of the problem. After normalizing the scores, the automatic score and the human score get a correlation coefficient of 0.79.Second, base on the BLEU, application of the case-based method in the quality evaluation of English-Chinese translation is studied. The impact of the Chinese word model and Chinese character model is studied on the research of Chinese language at the special problems. The result of automatic evaluation with human-scored evaluation are compared to verify the effectiveness of automatic evaluation.Third, base on the existing machine translation evaluation methods, some methods are applied to English-Chinese machine translation quality evaluation. After normalizing the system scores from various evaluation methods, a comparison is made to verify the validation of automatic MTE methods, so that we can make a further research in the future.