前編輯與機器翻譯：海洋英文閱讀教材之研製用途

Pre-Editing and Machine Translation: Developing Marine English Reading Materials

臺灣本地出版社發行的海洋英文教材匱乏，導致很難推動海洋與英文整合之教育。基於此故，作者提議應用科技輔助方法自我研製海洋英文閱讀文本。該方法乃是使用控制性語言前編改寫海洋議題的中文文本，然後使用線上神經機器翻譯系統，譯成英文文本。本研究透過文本分析發現，如果與非控制性文本之英語機器文本相比較，控制性前編文本的英文機器譯文含有較多單字，可符合大學入學考試中心所頒布的一至六級高中詞彙表。此外，也較少使用複雜句與連寫句，所以具有較高的可讀性及可理解性。正因為控制性文本之英文機器譯文的詞彙與句構複雜程度較低，學生給予其可理解性之評分亦較非控制性文本之英語機器文本高。另一方面，針對學生閱讀機器譯文感到困難的阻礙要素，閱讀非控制性英語機器譯文時，高達86.6%（71/82）的學生選擇困難的詞彙，而閱讀控制性英語機器譯文時，僅有54.9%（45/82）贊同此要素。於句構方面，閱讀非控制性英語機器譯文時，48.8%（40/82）的學生選擇英文文法錯誤，但另一群組中，僅有28%（23/82）的學生贊同此原因。由此可知，前編輯後的海洋英文機器譯本，可提升學生的可理解度，故本論文強調科技輔助教材研製的效益，也呼籲高中教師可使用控制性中文與機器翻譯為自己的學生研製可理解的海洋英文閱讀教材。

This paper proposes the application of an online neural machine translation (MT) system and controlled language (CL) to develop marine English reading materials for senior high school students. Scant marine English reading materials have been issued by local publishers in Taiwan, making it challenging to integrate marine ecology and English education. The author of this paper proposes a technology-enhanced approach to self-develop marine English reading materials by adapting Chinese source texts with CL and then translating them into English by using the MT system of Google Translate. Textual analysis revealed that English MT output of CL texts, when compared with uncontrolled texts, used more words from levels one to six of the word list issued by Taiwan's College Entrance Examination Center (CEEC). The MT output of CL texts contained few complex, run-on sentences, so readability and comprehensibility were relatively high. Due to lower lexical and syntactic complexity, English MTs of CL texts received higher scores than those of uncontrolled texts when assessed by student respondents in a questionnaire-based survey. In response to a question regarding the linguistic factors that affect comprehension, 86.6% (71/82) of students reading uncontrolled texts complained about difficult words, whereas only 54.9% (45/82) of students reading CL texts encountered this problem. Regarding the syntactic factor, 48.8% (40/82) of students reading uncontrolled texts noticed English grammatical errors, compared with only 28% (23/82) of students reading CL texts. These findings highlight the effectiveness of boosting the comprehensibility of the MTs of marine English reading texts through pre-editing. This paper concludes by describing the technology-enhanced approach and calling for the use of CL and MT to produce comprehensible marine English reading texts for senior high school students.