Abstract
Machine Translation is one of the most difficult problems in the field of natural language processing. In the past, MT has been applied to professional communication in the process of translating technical and corporate document in a specific domain. Recently, because of the rapid development of Internet and the need to access information across the language, people began to look into the role that MT can play in Cross Language Information Retrieval. The prevalent approach to CLIR is based on translation of query phrases. We propose a noval approach based on Statistical Phrase Translation Model (SPTM), aimed at achieving a tighter estimation of phrase translation probability.Experiments were conducted using bilingual phrases in the BDC Electronic Chinese-English Dictionary. The training of alignment model is done by the EM-algorithm. For evaluation, we adapted the methodology used by Och et al. (2000) to assess the performance of the experiment. We obtained the recall rate of 92.0%, the precesion rate of 91.3% and the error rate of 8.4%.The effect of Chinese segmentation and initial model of EM algorithm was also studied. We found that Chinese segmentation can improve the traning result slightly. A better initial model was found to improve the performance of the EM algorithm significantly.