Machine Translation Error Analysis And Post-Editing Strategies

In recent years,artificial intelligence and machine translation(MT)technologies have progressed rapidly.Still,MT cannot be expected to match human translations in terms of quality,though it excels in translation speed.In order to optimize the balance between translation quality and efficiency,post-editing(PE)has been widely adopted in the language industry.In this report,the author applied PE in her translation,analyzed the error types of MT,and put forward suggestions for dealing with the most frequent MT errors.The thesis is based on the translation of the first chapter of the 2018 Economic Report of the President,a typical kind of economic text characterized by numerous long sentences,abundant technical terms,complex sentence structures(parallel structures,subordinate structures and insertion elements).The author first selected an MT engine(i.e.,Google Translate)based on translation quality out of the eight engines provided by JEEMAA,an online computer-aided translation platform,and pre-translated the source text.The errors in the translation of one part of the source text were categorized,counted,and corrected.Cases were analyzed.It was found that in this MT exercise,errors at the lexical level incude mistranslation,undertranslation,overtranslation and omission,Errors at the syntactic level are omission or mistranslation of clauses and inappropriate order of words.Errors at the textual level mainly refer to lack of coherence,inconsistent format and lack of briefness.It was also found that errors at the lexical level were the commonest compared with errors at syntactic and textual levels.The four most frequent types of errors that the MT engine made were mistranslation,undertranslation,inconsistent format,and inappropriate word order.For coping with such errors,it is suggested that the post-editor analyze the errors,confirm the meaning,convert part of speech and restructure long sentences when needed.