以電腦技術自動採用翻譯資源

Abstract

電腦輔助翻譯工具現今廣泛運用於翻譯實務，其中翻譯記憶系統可儲存既有之原文與譯文為平行語料庫形式，再以相似度對應擷取相似內容，由此利用先前翻譯之重複內容，遂廣運用於重複性高之技術性文件。然翻譯記憶資料來源有限，須仰賴既有平行文本，為增加資料來源，本文承前人研究，討論以可比語料庫建立翻譯記憶資料之可行性，並探討其他資料來源。陳碧珠(2011)、李佳陵(2017)等先前研究指出，可比語料庫於翻譯工具中之使用似乎利用率不高，與所建立之資料庫大小不符，因此本文亦從翻譯記憶之相似度對應方法著手，探討如何提高翻譯記憶系統擷取相似內容之效率，以此提出改善翻譯工具之建議。 關鍵詞：電腦輔助翻譯工具，翻譯記憶，可比語料庫，相似度比對，機器翻譯

Computer assisted tools (CAT) are widely-used in today’s translation work, and translation memory (TM) systems help translators deal with repetitive expressions or contents in similar contexts for maintaining consistency or saving time and efforts. The data in translation memory systems are stored as parallel corpora, but when there is no previously aligned language pairs for professional or technical texts of a specific field, the TM is empty and translators need to start from scratch. Chen (Chen, 2011) suggests that comparable corpora of naturally produced texts in the working languages may help. With the help of machine translation, comparable corpora can be turned into parallel corpora for use in TM systems. However, current TM systems are limited in matching similar contents. The study suggests that term weighting techniques used in information retrieval(Gao, 2002) be adopted in order to find out truly important and relevant contents while matching. In the proposed method, content words that carry important information are given more weight while functional words or expressions that are less specific to a text are deemed as less important. The study tests the use of comparable corpora and the adoption of term weighting by comparing texts of a prospectus and its translation. Keywords: computer assisted translation tool (CAT), translation memory (TM), comparable corpora, similarity measure, machine translation