翻譯學習者之翻譯錯誤分析：以語料庫為基準之應用研究

Abstract

本研究從翻譯過程及翻譯成品探討學習者之翻譯錯誤，並探討將語料庫方法應用於翻譯錯誤研究之可行性。研究對象為 70 位學生，其中 31 位來自大學部，39 位來自翻譯研究所；所收集的資料為參與者譯自六篇資訊類文本的中文翻譯共 420 篇，以及 70 份訪談謄寫稿與背景問卷。研究者將 420 篇中文翻譯建置為一學習者翻譯語料庫，將其中翻譯錯誤以標記軟體 MMAX2 標記，並以統計方法（T 檢定、Mann-Whitney U 檢定、Kruskal-Wallis H 檢定）檢驗不同組別學生的翻譯錯誤次數在各項類別上是否有差異。訪談資料之分析方法則參考 Strauss and Corbin （2008）的開放編碼、主軸編碼、以及選擇性編碼三步驟，歸納出學生產生翻譯錯誤的原因。 研究結果顯示，語料庫方法可有效應用於翻譯錯誤研究，本研究以MMAX2成功標記中文翻譯語料之錯誤，並可將其搜尋、統計、內容呈現的功能應用於翻譯教學。此外，研究發現大學部學生和研究所學生在某些翻譯錯誤類別上有顯著差異；訪談資料分析結果顯示，學生的翻譯錯誤並不單是因為能力不足所造成，也可能是因為翻譯過程中其他因素相互作用下（如動機、規範、時間、後設認知策略、翻譯任務之目的等）而產生的選擇。本研究之學習者翻譯語料庫建置與標記成果，以及兼顧翻譯過程與成品的錯誤分析之發現，應可作為實徵研究證據，提供翻譯教師和學者參考。

This study aims to investigate the errors of translation learners from both their translation product and their translation process, as well as to test the practicality of a corpus-based method for such analysis. The data collected from 31 undergraduate and 39 graduate student participants were their translations of six informative English texts, which were compiled into a Chinese learner translation corpus and annotated using MMAX2, and the 70 interview transcripts and background questionnaires. Error analyses from the product were carried out through annotating the learner translation corpus and by comparing the differences of error frequency in each text among groups by statistical tests such as the t test, the Mann-Whiney U test and the Kruskal-Wallis H test, while error analyses from the process were conducted by retrospective interviews and the coding procedures proposed by Strauss and Corbin (2008).

The results of the compilation and annotation of the Chinese learner translation corpus, have been successful not only in the error analysis for research but also in its output of the query and statistics function as concrete feedback to individual students and as materials suitable for data-driven and student-based classroom discussions. The flexibility and customization of MMAX2 has also allowed future expansion and modification of this annotated corpus. In addition, results of this study indicate that the undergraduate and graduate students showed a significant difference in some types of errors and their errors were not simply the results of incompetence (lack of abilities) but were also formed under the influence of some interacting factors such as willingness (norm and motivation), time, meta-cognitive strategies, and the Skopos of the translation task. The achievements and findings of this study may offer empirical evidence for increasing our understanding of the translation process of learners through error analysis and present a practical corpus-based method conducive to translation teaching and research.