A Post-Editing Study On C-E Machine Translation Of Net Literature

In recent years,a large quantity of Chinese net literature has been translated overseas,which makes a great significance for Chinese culture to spread abroad.Based on its hypertext characteristics and commercial mechanism,the translation activity of net literature has formed a unique translation mode,the“Machine Translation + Post-Editing”（MTPE）translation mode.While chronically the combination of computer technology and translation activities has been mainly practiced to applied text,and the theoretical guidance and the evaluation system of literary text is not yet mature.Thus,based on the general Translation Quality Assessment（TQA）model of House and the Translation Error Ratio（TER）translation quality evaluation method based on edit-distance,this study made a general description to the Chinese-English machine translation quality of net literature,summarized the main error classifications,and made a formal description of its post-editing operations,to explore one mode that can be used in the post-editing practice of net literature machine translation.Through the comparison and analysis of the triplet data of the source text（src）,the machine translation（mt）,and the post-edited translation（pe）,the results show that,in terms of readability,the machine translation system with the support of neural network technique can basically meet the needs of readers.In terms of language expression,at the lexical level,the main error types in MT are terms,conjunctions,quantifiers,tenses,genders,net words,names,places,idioms,and repetitive expressions.At the syntactic level,there are sentences without subjects,modifiers,tenses,sentence patterns,word orders,mistranslations,and punctuation errors.The textual level is mainly manifested in anaphora,substitution,ellipsis,cohesion,oral function and misidentification of information focus.In terms of style,different themes and types of net literature have different characteristics of text and narrative rules.The evaluation results of edit-distance TER index are different,and the average is 48.42%.In terms of post-editing operation index,the operation parameters of TER index based on edit-distance mainly focus on replace operations and addition operations.The average index of editing words is 22.65 in one sentence,and there is a slight difference in post-editing operation indexes between different types of text.Thus,based on the results,at the lexical level in post-editing work,this research is represented by the extraction,recovery and addition of terms to form a glossary,which is ready for editing and translation work latter.Paying close attention to the error types of non-term words in different texts,the editor can process the text words in advance.At the syntactic level,the net literature corpus can be customized through syntactic supplement,voice substitution,subject conversion and other forms to improve the quality of translation.At the discourse level,according to different text styles and edit-distance,the rule-based Chinese analysis and English generation rule system learning should be strengthened,and a suitable text discourse model of net literature should be established in big data.Similarly,the post-edited translation is aligned with the original text to form a memory of translation data,which makes the MT system more intelligent.The research on the post-editing mode of C-E machine translation of net literature provides a practical exploration of machine translation in literary translation,and also provides some guides for the exploration of the quality assessment of net literature translation and the development of literary translation system.