生成式人工智能在法律文本翻译中的表现研究——以《民法典》总则第六章民事法律行为英译为例

Research on the Performance of Generative Artificial Intelligence in Legal Text Translation—A Case Study of English Translation of Civil Juristic Acts in Chapter VI of the General Part of the Civil Code

摘要: 生成式人工智能的快速发展为法律翻译带来了新的机遇与挑战。本研究选取《民法典》总则第六章民事法律行为作为研究语料，从术语、语义、语法及其他四个维度，对比分析ChatGPT-4与DeepSeek-V3两款生成式人工智能在英译中的表现。研究发现：两款生成式人工智能在语法层面表现良好，但在术语与语义层面存在较严重的问题，核心短板集中在需要专业法律知识支撑的领域。在错译和漏译上，DeepSeek-V3整体表现略优于ChatGPT-4。上述问题反映出目前生成式人工智能在法律翻译中仍存在局限性，难以精确把握法律内涵。本研究揭示了生成式人工智能在法律翻译中的能力边界，为法律翻译实践与人工智能应用提供参考。

Abstract: The rapid development of generative artificial intelligence presents both new opportunities and challenges for legal translation. This study selects Chapter VI Civil Juristic Acts in the General Part of the Civil Code as the research corpus, and comparatively analyzes the English translation performance of two generative AI models, namely ChatGPT-4 and DeepSeek-V3, from four perspectives including terminology, semantics, grammar and other aspects. The research results reveal that both models show sound performance in grammatical expression, but have prominent defects in terminology translation and semantic interpretation, with their main deficiencies lying in fields that demand specialized legal knowledge. In terms of mistranslation and omission, DeepSeek-V3 outperforms ChatGPT-4 slightly on the whole. These phenomena demonstrate that current generative artificial intelligence still has obvious limitations in legal translation and cannot accurately grasp the inherent legal implications. This study defines the capability boundaries of generative AI applied to legal translation, and offers valuable references for legal translation practice and the practical application of artificial intelligence.