When Algorithms Translate: Rethinking Lefevere's Manipulation Theory in the Age of AI - A Comparative Analysis of Todd Foley's and ChatGPT's English Translations of Yu Hua's Sleeper Bus

Abstract

André Lefevere's manipulation theory posits that translation is a form of rewriting governed by the triadic forces of ideology, poetics, and patronage. The advent of large language models, however, is fundamentally reshaping this theoretical landscape. When AI acquires the capacity for literary translation, the traditional triadic relationship of "human—culture—text" evolves into a four-way interaction of "human—algorithm—culture—text," with the agent of manipulation extending from the human translator to non-human algorithmic systems. This paper takes Yu Hua's short story Sleeper Bus — published in The Paris Review in spring 2026 — as a case study, juxtaposing the professional human translation by Todd Foley (the English translator of Yu Hua's Wencheng) with the machine translation produced by ChatGPT. Drawing on Lefevere's classical framework, this study proposes the concept of "digital manipulation" to examine how AI translation diverges from human translation along the three axes of ideology, poetics, and patronage. The findings reveal that Foley's translation exhibits a highly self-conscious cultural mediation, maintaining a delicate balance between domestication and foreignization. ChatGPT's translation, by contrast, displays a tendency toward "smooth universalism," systematically effacing the cultural specificity of contemporary China. The "patron" behind AI translation is no longer an institutional force but rather the corpus bias embedded in training data and the probabilistic logic of the algorithm itself. This discovery suggests that translation studies must incorporate algorithms into the genealogy of manipulative agents, thereby responding to the fundamental transformation of translational practice in the AI era.