顺应理论下不同大语言模型的文学类文本翻译效果对比研究——以《雾都孤儿》(Oliver Twist)为例

A Comparative Study of Literary Text Translation Performance among Different Large Language Models under the Adaptive Theory Framework—Taking Oliver Twist as an Example

摘要: 随着大语言模型的发展，其在翻译实践中的使用显著增多。本文选取《雾都孤儿》英文原文为材料，以荣如德经典译本为参照，在顺应理论框架下从时间适应性、语境适应性与结构适应性三个维度对ChatGPT、DeepSeek与文心一言三种大语言模型的中文译文进行对照分析。研究发现，DeepSeek在段落组织、修辞节奏与讽刺氛围的保持上更接近原作；ChatGPT在历史语体与当代可读性之间较为平衡，语境信息交代较为清晰；文心一言叙述连贯、表达清楚，但整体更偏现代化语体，并倾向于强化情绪色彩。本文认为，在强调文体与结构再现的翻译场景中，更适合采用DeepSeek；面向一般读者且强调顺畅阅读时，ChatGPT更具优势；对于说明性较强、需要降低理解门槛的段落，文心一言的译文更为稳妥。

Abstract: With the advancement of large language models, their application in translation practice has significantly increased. This paper selects the English original text of Oliver Twist as source material and uses Rong Rudé’s classic translation as a reference. Within the framework of the Conforming Theory, it conducts a comparative analysis of the Chinese translations produced by three large language models—ChatGPT, DeepSeek, and Wenxin Yiyan—across three dimensions: temporal adaptation, contextual adaptation, and structural adaptation. The findings reveal that DeepSeek maintains greater fidelity to the original in paragraph organization, rhetorical rhythm, and the preservation of satirical atmosphere. ChatGPT strikes a relatively balanced approach between historical register and contemporary readability, with clearer contextual information delivery. Wenxin Yiyan offers coherent narration and clear expression, though its overall style leans more toward modernity and tends to amplify emotional tones. This paper argues that DeepSeek is more suitable for translation scenarios emphasizing stylistic and structural fidelity; ChatGPT holds advantages when targeting general readers and prioritizing smooth readability; and Wenxin Yiyan’s translations prove more reliable for highly explanatory passages requiring simplified comprehension thresholds.