PATAS：個人化適應性翻譯輔助系統之設計與實現

PATAS: A Personalized Adaptive Translation Assistant System

現今市面上的計算機輔助翻譯系統，大多把焦點放在如何提高翻譯效率，以減少重複翻譯的功能特點上，對於如何提升翻譯品質並沒有太多的著墨。因此，本論文提出PATAS系統（Personalized Adaptive Translation Assistant System），希望讓計算機輔助翻譯系統不僅能夠透過翻譯記憶來提高翻譯效率，還能夠應用文件採擷的技術來提昇翻譯品質。整體系統的主要功能如下： （1）利用段落自動分群技術，提示翻譯者可以一併修改內容相似的段落，以解決不同時間點所採用的翻譯角度不一致問題，或不同譯者合作翻譯同一著作的不一致性問題。 （2）系統也利用翻譯者的修正行為，找出內文的權威段落（Authority Paragraph）與集散段落（Hub Paragraph），用以協助維持譯文的一致性。 （3）系統透過共現詞（Co-occurrence Word）的採擷以及事件追蹤（Event Tracking）機制，找出個人翻譯習慣用語，以解決翻譯風格不一致的問題。 （4）提出運用原文與譯文文件摘要之間的段落交集，來量化翻譯品質，以做為評估翻譯品質好壞的參考指標。 本系統與翻譯界廣受歡迎的SDL Trados計算機輔助翻譯系統進行比較，經評估後均獲得相當正面的評價，顯示利用文件採擷技術應用於計算機輔助翻譯系統上，不僅可以提高翻譯效率，對於譯文品質也有實質上的提升。

In this paper, we will propose a general framework for improving the efficiency, and the quality of a translation process based on the application of text mining. We have designed and implemented a prototype system, named PATAS (Personalized Adaptive Translation Assistant System), to be the testbed of our approach. The main features of our system include: (1) Use paragraph clustering to deal with inconsistent translation viewpoints. (2) Track the behavior modification log to find the authority and hub paragraphs for keeping meaningful consistency for related sentences or paragraphs. (3) Find co-occurrence words and provide event tracking mechanism to adjust the idioms and translation styles of various translators. (4) Finally, document summarization of original text and translated text are compared by cross-intersection to measure the quality of translation. We have compared our system features with the most popular computer assisted translation system, SDL Trados, to verify the effectiveness. The conducted survey shows that our approach does not only improve the efficiency of translation, but also promote the quality of translation.