Investigating English-Chinese Word Level Alignment by Using Semantic Similarities and Linguistic Knowledge

To tackle the precision problems that statistical models of alignment face in aligning translation units across language families, this paper proposeg a new approach to Chinese and English word level alignment. Using an operating definition of alignment and calculating the semantic similarities and applying linguistic knowledge, this research proposes an algorism with optimal length segmentation and dynamic extraction strategies. Considerations of relations in close proximity and a relative distortion method together with a few linguistic rules are taken to deal with word match irregularities and phrase identification issues. Empirical test results show this method can reach an 87.81% of word alignment with a precision rate as high as 90.1% though the training corpus is relatively small.