ChatGPT於軍事文本翻譯應用研究

Abstract

軍隊近年來對翻譯與外語人才資源需求升高，本研究主要探討生成式人工智慧科技ChatGPT在軍事文本翻譯上是否能夠更有效率地處理國際軍事資訊及軍事交流等需求。本文從兩個研究問題出發，分別是譯文準確度和讀者譯文資訊理解程度。研究設計以ChatGPT的GPT-4模型，將國防譯粹中不同主題的8個文章段落翻成中文譯文，並邀集了軍事、外語及翻譯等3位專家針對ChatGPT譯文準確度進行評量，同時以問卷調查軍士官兵對譯文的主觀感受與理解程度，最後再與人工譯文的評分與意見相比較，了解差異，進而探究實際應用的可能。最終，在量性研究後所得分數，ChatGPT譯文的資訊理解與吸收程度在滿分5分的量表中，獲得平均3.92分，對比專業譯者的人工翻譯4.10差0.18，經由分析與個別譯文評分探討，ChatGPT相較於人工譯文並非全都較差，甚至有2段譯文的平均分數高於人工譯者，顯示雖然大部分情況下人工譯文優於ChatGPT譯文，但讀者對於ChatGPT譯文的接受程度也相當高，也同時從研究與語言連貫性分析中，了解ChatGPT在生成軍事譯文上的限制。除了探討ChatGPT在軍事翻譯上的應用外，本研究過程中也對AI工具有其他發現，如操作AI及提詞的使用方式，都將大幅影響產出結果，以及使用線上AI工具的資安隱私問題等，都是值得討論的議題，並且也是具有價值的研究方向，未來如何將人工智慧與人類翻譯者的力量結合，實現更高速度和效率的翻譯作業，才能在安全保密的情況下達到軍事翻譯效率優化。

With an increasing demand for translation and foreign language expertise in the military in recent years, this study primarily investigates the potential of the generative artificial intelligence tool, ChatGPT, in military text translation. The aim is to determine whether it could provide a usable solution to enhance the efficiency of processing international military information and engagement. Our research questions focus on two areas: the accuracy of the translations and the comprehension level of the translated information by the readers.Using the GPT-4 model of ChatGPT, we translated eight different paragraphs from National Defense Digest into Chinese. We then invited experts in military affairs, foreign languages, and translation to evaluate the accuracy of the ChatGPT translations. Additionally, we conducted a questionnaire survey among military noncommissioned officers to understand their subjective impressions and comprehension level of the ChatGPT translations.The outcome of our quantitative research revealed that the comprehension and absorption levels of information in ChatGPT translations averaged at 3.92 out of 5, only 0.18points less than the score of human-translated versions. We found that not all ChatGPT translations underperformed human translations. In fact, two sections scored higher than their human-translated counterparts, indicating a high level of acceptance among readers.While we mainly explored the application of ChatGPT in military translation, we also discovered additional considerations with AI tools. These include the impact of different operational methods on output results and concerns related to cybersecurity and privacy. Our findings open up new directions for future research in the integration of artificial intelligence and human translators to enhance translation speed and efficiency, whilst ensuring security and confidentiality.