形式衔接与语义偏移：《活着》人机翻译对比

Formal Cohesion and Cognitive Disjunction: A Comparative Analysis of Human and AI Translations of To Live

摘要: 文学翻译最棘手的问题始终在于深层文化连贯的重构，人工智能在此领域的表现究竟如何，尚待系统检验。以衔接与连贯理论为框架，结合“意图–认知–语境”动态模型，本文对比分析了《活着》白睿文译本与AI译本在文化认知映射、情感潜文本传递和审美风格统摄三个维度的差异。分析表明，AI在显性语法衔接层面已逼近人工译者水平，但在深层连贯重构上却呈现“形式连贯–语义偏移”的系统性偏差。这一偏差的根源在于AI以概率匹配为基础的运作机制无法完成意图感知与认知语境构建。基于上述发现，本文将“深层连贯”细化为可操作的三个子维度，进而提出人机协作的“三明治模型”，以期为AI时代的文学翻译提供理论参照与方法论框架。

Abstract: Reconstructing profound cultural coherence has long stood as the most intractable bottleneck in literary translation, and the practical performance of artificial intelligence within this field still awaits systematic empirical examination. Anchored in cohesion and coherence theory and supplemented by the dynamic “intention-cognition-context” analytical model, this paper compares and contrasts Michael Berry’s authorized English rendition and AI-produced translations of Yu Hua’s To Live across three core dimensions: cultural cognitive mapping, conveyance of implicit affective subtext, and holistic regulation of aesthetic stylistic unity. Analytical results demonstrate that artificial intelligence has nearly caught up with human translators in terms of explicit grammatical cohesion, yet it falls prey to a systematic flaw characterized by “formal coherence alongside cognitive disjunction” when reconstructing deep-seated textual coherence. Rooted in AI’ s probability-matching operational logic, such structural deficiency stems from the algorithm’s inherent inability to accomplish intentional perception and the construction of context-bound cognitive frameworks. Drawing on the above analytical outcomes, this study disassembles deep-level coherence into three operable sub-dimensions and proposes a human-machine collaborative “Sandwich Model”, so as to provide theoretical references and a feasible methodological framework for literary translation research in the artificial intelligence era.