Training in Pre-MT Editing With the Controlled Language: A Pilot Study

[[abstract]]本研究旨在探究翻譯訓練中應用前機器翻譯編輯之效應。為進行研究，在前機器翻譯編輯的過程中，我們運用控制語言來修改、編輯原文，再將編輯後之原文送入機器翻譯系統。一般而言，控制語言可限制原文中的詞彙及文法，因此降低原文的複雜程度，並闡明原文中較模糊的部份。本研究運用十萬字的雙語平行語料庫，語料庫取自史宗玲教授的國科會計劃，語料內容則為使用手冊及產品說明書。此平行語料庫包含原文、機器翻譯產出之翻譯、經過前編輯的原文及產出之翻譯。此語料庫的機器翻譯成果分別由歐泰科技公司研發的譯經（TransWhiz）及Google Translate（Google翻譯）產出，而此兩種機器翻譯系統分別使用規則法及統計法，故翻譯結果不同。 依據現有研究成果，在翻譯訓練中採用前機器翻譯編輯對學生之效益如下：首先，學生必須分析原文句構，並在機器翻譯系統產出合宜翻譯前反覆修改原文，故學生可藉由前機器翻譯編輯辨別中、英文的語言差異，並提升其改寫能力。其次，學生可透過此練習了解機器翻譯在詞彙及句法方面的限制，且熟悉機器翻譯系統的操作，進而提升機器翻譯之應用效率。最後，學生可瞭解一系列用於前機器翻譯編輯的規則，以便於日後翻譯時依循之。此外，作者依據自身從事前機器翻譯編輯的經驗，針對前編輯的規則進行粗略評估。雖然本研究之語料庫數量有限，對編輯規則的評估亦太過主觀，但本研究對前機器翻譯編輯之訓練提出實用建議，將可作為未來研究控制語言及機器翻譯之參考。[[abstract]]This study aims to investigate the effectiveness of adopting pre-machine translation (pre-MT) editing in translation training In the process of pre-MT editing the controlled language is used to revise and edit the source text (ST) for MT application Generally the controlled language helps restrict the vocabulary and grammar of the ST thus reducing the complexity as well as clarifying the vague expression in the ST This study develops a 100 000-word parallel corpus of user manuals and product instructions retrieved from Prof Shih Chung-ling’s NSC project The corpus consists of STs MT outputs pre-edited STs and MT outputs of pre-edited STs The MT outputs are created by TransWhiz and Google MT that respectively adopt rule-based and statistics-based approaches so different performance are identified The findings show that adopting pre-MT editing benefits student much in translation training First pre-MT editing practice enables trainee translators to identify the linguistic differences between English and Chinese and to improve their paraphrase skill because they have to analyze the ST structure and repeatedly revise the ST until the satisfying MT output is produced Next pre-MT editing helps trainee translators to improve working efficiency through MT application and makes them understand some MT constraints in the lexical and syntactic aspects In addition this study introduces a set of useful guidelines for translation practitioners to follow in pre-MT editing Finally a rough assessment on the pre-editing principles is made based on the author’s empirical experience Despite the small sample size (100 378 words in total) and the subjective assessment this study provides some significant insights about benefits of pre-MT editing in translation training The results of this study can also serve as a reference for the future research on the controlled language and MT applicatio