Analysis Of Software-based Translation Errors And Post-editing Strategies For Energy Policy Texts

In recent years,neural machine translation has been greatly improved.More and more neural machine translation systems claim to be able to simulate the learning mode of the human brain and output excellent translation,which has led many scholars to research on the quality of machine translations.This thesis aims to explore the types of errors made by neural machine translation system in translating energy policy texts,and to propose post-editing strategies.Based on former research on individual neural machine translation systems,the author selects Deep L neural machine translation system to start the study.Moreover,the author classifies the errors in the machine translation into three levels.Finally,under the guidance of the Skopos theory,the author searches for post-editing strategies for the current machine translation errors of this type of text.It is found that: although the current neural machine translation has made great improvements in technology,errors such as punctuation errors,omission,lexical errors,grammatical errors and incoherence still occur in translating energy policy texts.Punctuation errors and omission errors are found to occur 138 times,with a frequency of 46.3%.The author finds that the lexical errors and grammatical errors occur 91 times,with a frequency of 30.5%.Statistically,the problems of illogicality between sentences occurs 69 times,with a frequency of 23.2%.Accordingly,the author put forward corresponding solutions to the aforementioned errors: revising punctuation and supplementing translations;correctly translating vocabulary and grammar and adjusting sentence structure.This report provides a reference for the types of errors of neural machine translation system in translating energy policy texts and provides three post-editing strategies.