Study On Translationese In Machine Translation Of Sci-tech Texts

Benefiting from the rapid development of computer technology,machine translation has been widely used in translation practice.Sci-tech(Scientific and technical)texts are considered to be the easiest texts for a machine to converse for concise expressions and specific language formats.However,due to both linguistic and technical factors,unsatisfactory translations are bound to appear in machine translation,which affects the quality of translated texts,and the phenomenon of translationese is one of them.Previous research on translationese has achieved remarkable achievements.However,it’s a pity that the study on translationese of C-E translation is neglected and the study on translationese in machinetranslated texts is few.Besides,the method of qualitative analysis of translation cases is usually used in the related research,quantitative analysis needs to be strengthened.In this thesis,sci-tech texts are used as the corpus to study the translationese in C-E machine translation.In order to present the research process more objectively,four monolingual comparable corpora are established,including source Chinese Corpus of Sci-Tech Texts(CCST),Human-translated English Corpus of Sci-Tech Texts(HECST),Machine-translated English Corpus of Sci-Tech Texts(MECST),and original English Corpus of Sci-Tech Texts(ECST).Based on the theories of comparative linguistics studies of English and Chinese and corpus-based translation studies,quantitative analysis of the linguistic characteristics of English machine translations of sci-tech texts is conducted with the corpus data such as standardized type/token ratio(STTR),high-frequency words,average sentence length and average paragraph length.Then the typical forms of translationese in English machine translation of sci-tech texts are summarized by qualitative analysis.What’s more,the causes of translationese and solutions to translationese are also explored.Major findings can be generalized as follows.Firstly,compared with the original English texts,the lexical richness of the machine-translated texts is reduced.In terms of highfrequency words,more high-frequency words within a certain frequency are found in machine-translated texts than in the original English texts,but the frequency of high-frequency words within a certain ranking in the machine-translated texts is always higher than that in the original English texts.And the average sentence length and the average paragraph length of the machine-translated texts are found to be increased.Secondly,typical forms of translationese in the English machine translation of sci-tech texts can be summarized as inappropriate word selection,word disorder,redundant expressions,lack of logical relations,improper transfer between passive voice and active voice,and insufficient idiomatic structures of the target language.The causes of translationese come from two aspects,namely the inherent differences between English and Chinese and the limitations of the machine translation systems.To avoid translationese in MT,two approaches are proposed in this thesis,namely continuously improving the machine translation system and promoting the post-editing method.The study on translationese in machine translation can provide the linguistic analysis basis for the researchers to improve the machine translation system and present focus on avoiding translationese for posteditors so as to improve the quality of machine translation and promote the long-term development of machine translation.