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.
