知识翻译学将翻译界定为地方性知识的世界性再生产。以此审视牡丹文化意象的机器翻译,可发现一个深层困境:机器输出的不是“知识重构”,而是 “知识扁平化”,即多层叠加的地方性知识被简化为单维度的普世化概念。通过对比分析古典诗词、植物志、旅游宣传三类文本的机器翻译与人工译本,系统诊断其 “知识识别误判”“背景知识剥离”“语用知识失效” 三重问题,揭示生成式AI统计模仿的本质。构建基于知识运作流程的译后编辑干预策略模型,从知识单元的精确化、背景知识的补偿化、语用知识的适配化三个层面,推动PE范式从“纠正—修补”向 “评估—优化” 演进。以“知识扁平化”概念丰富了知识翻译学对“知识重构”的理解,为AI时代中国文化知识的精准外译提供理论框架与操作路径,揭示人机协同中不可让渡的“人的判断”。

Transknowletology (a theory defining translation as the global reproduction of local knowledge) reveals a deep-seated dilemma when applied to machine translation of peony cultural images: what machines produce is not “knowledge reconstruction” but “knowledge flattening”—the multilayered local knowledge is reduced to one-dimensional universal concepts. Through a comparative analysis of machine translations (including traditional NMT and generative AI) and human translations of three text types—classical poetry, botanical records, and tourism promotions—this study systematically diagnoses three interconnected failures in handling peony-related cultural knowledge: misidentification of knowledge, loss of contextual knowledge, and failure of pragmatic knowledge. It further reveals the nature of generative AI: statistical imitation of knowledge rather than rational cultural judgment. In response, this study constructs a post-editing intervention model based on the knowledge operation process, advancing the PE paradigm from “correction-patching” to “evaluation-optimization” through three strategies: precision of knowledge units, compensation of contextual knowledge, and adaptation of pragmatic knowledge. This research not only offers a theoretical framework and operational path for the accurate translation of Chinese cultural knowledge in the AI era, but also enriches the understanding of “knowledge reconstruction” in Transknowletology with the concept of “knowledge flattening,” revealing the non-negotiable “human judgment” in human-machine collaboration.