This study takes TM/Win and Trados to explore the effects of the alignment function of Translation Memory (TM) systems on different types of text. In the first phase, this study performs some TM test on three types of texts (18 texts, total word count is 10,942 words). Then, all tested texts can be further divided into five types of writing to examine the connection between text types and TM tools. In this study, there is a new scoring table which is created for evaluating the reliability of aligned pairs. All results of evaluations indicate that the alignment pairs generated by TM/Win and Trados are different, and the alignment performance of both TM tools varies with the types of text. This study also creates some pre-editing rules based on the variation of sentence structures and segmentation marks. The result of outside test with these rules indicates: the correctness rate of tested text with Trados has risen to 87% (initial test: 32%), and the correctness rate with TM/Win has even reached 98% (initial test: 49%).
