本文以美國人工智慧研究實驗室OpenAI 所開發的大語言模型ChatGPT為研究物件，探討新一代人工智慧技術在翻譯實踐中的應用潛力及其局限性ChatGPT 憑藉其大規模語言訓練與生成能力，能夠快速提供流暢的譯文，顯著減輕用戶在翻譯過程中所投入的時間。然而，“人工翻譯—尤其是文學翻譯—是否將被完全取代”，仍是一個有待深入探討的問題。本研究依據Mona Baker（1992）所提出的詞語層面翻譯策略分類（本文稱為“貝克分類”），以錢鐘書小說《圍城》的西班牙語譯本（Taciana Fisac,1992）為參照，選取其中25 個具有代表性的語段，從文化詞項、隱喻表達、句式結構及語用適配等維度，系統對比人工譯本與ChatGPT 自動生成譯本之間的差異。分析表明，ChatGPT 在百科知識的資訊轉換和基本語義傳達方面表現優異，能夠高效完成程式化語言的翻譯任務；然而，在處理具有深厚文化內涵的詞彙、文學性修辭以及語體風格統一等方面，仍存在明顯不足。本研究認為，儘管ChatGPT 可作為輔助工具提升翻譯效率，其在複雜文學翻譯中仍無法取代人工翻譯。這也再次印證Nord（1997）提出的“功能＋忠誠”原則，機器無法像人類譯者那樣靈活選擇翻譯策略來保證譯文符合各方需求的目的。人工智慧與人工翻譯在實踐中可形成互補關係，未來應在人機協同模式下進一步探索其在漢西翻譯實踐應用中的合理路徑。
This study examines the application potential and limitations of ChatGPT, a large language model developed by OpenAI, in the context of translation practice. Leveraging its extensive language training and generative capabilities, ChatGPT can rapidly produce fluent translations, significantly reducing the time invested by users. However, the question of whether artificial intelligence can fully replace human translation—particularly in literary translation—remains open for further investigation. Based on the classification of translation strategies at the word level proposed by Mona Baker (1992), referred to in this paper as the “Baker Taxonomy,” this research systematically compares 25 representative segments from the Spanish translation of Fortress Besieged by Qian Zhongshu (translated by Taciana Fisac, 1992) with corresponding versions generated by ChatGPT. The analysis focuses on dimensions such as culture-specific terms, metaphorical expressions, syntactic structures, and pragmatic adaptation. The findings indicate that while ChatGPT excels in conveying encyclopedic knowledge and basic semantic meaning, and performs efficiently in handling formulaic language tasks, it shows notable deficiencies in processing culturally nuanced vocabulary, literary rhetoric, and maintaining stylistic consistency. The study concludes that although ChatGPT can serve as an auxiliary tool to enhance translation efficiency, it cannot replace human translators in complex literary translation contexts. This further corroborates Nord’s (1997) “Function plus Loyalty” principle, underscoring that machines lack the human translator’s ability to flexibly choose strategies that align with varied communicative purposes and stakeholders’ expectations. Ultimately, artificial intelligence and human translation can form a complementary relationship. Future efforts should explore appropriate pathways for human-AI collaboration in Chinese-Spanish translation practice.