On Key Technologies For Statistic-Based Fuzzy Translation Memory

Because the size of the translation memory library and the repetition rate of the translation task are both limited,the traditional translation memory system can only play better in the specific translation task.However,while in the face of large-scale task or the task whose field is not limited,the ability of the traditional system will always be limited because of the incomplete translation memory library.In this thesis,we propose a framework of combining statistical machine translation and translation memory,by using the translation memory information in the statistical machine translation.Different with the current statistical machine translation system,the system in this thesis is based on the text which need to be translated first,and search from the translation memory library for the related segments which match the task as a reference.Then we input the translation memory reference and the text to be translated into the translation system.The translation memory reference segments will guide the decoding process.And after the re-rank of the translation hypothesis during the decoding process,the translation hypothesis which has some same words with the translation memory will get a high score,so that it can have a higher probability to be in the final translation result.The system in this thesis uses the IWSLT 2010 evaluation data as data set,and it does better on the translation performance than the traditional translation memory system and statistical machine translation system.The final translation result of the system in this thesis is better than the result of the traditional machine translation system on the language fluency,and at the same time,the system in this thesis can use the translation memory without the manual help.In a word,the system in this thesis has a good application prospect.