| Neural machine translation(NMT)is an end-to-end coding-decoding structure,which obtains mapping relations between natural languages through deep learning neural networks.As the latest development stage of machine translation,neural machine translation(NMT),compared with traditional phrase-based machine translation(PMT),has significantly improved the fidelity and fluency of its output results,and NMT has made many significant technological breakthroughs.However,there are still many problems in neural network machine translation,and the quality of translation still needs to be improved.This thesis selects 5958 medical terms as material in the form of document uploaded to Google and Baidu online translation platform,and uses the translation result as the research object.This thesis refers to the existing translation quality assessment models and classifies the types of errors into four categories.The results show that:(1)two types of neural machine translation errors occur frequently,namely,mistranslation,redundancy,omission,and word order;(2)the output of Baidu and Google Translate is different because of various training corpus and the major errors of the two Translates are caused by the lack of context and cultural awareness;(3)suggestions for improving machine translation quality from human interference are: literal translation,free translation,annotation and parallel text could be applied into handling mistranslation;addition and omission could be used to tackle omission and redundancy respectively;for words order,re-arrangement of words order could be adopted.In view of the above findings,the author believes that the mode of “machine translation + primary proofreading + advanced proofreading(with medical background)” should be adopted for translation in medical field,so as to effectively guarantee the quality of translation. |
PDF Full Text Request