From “Correct Translation” to “Great Translation”: Research on Human–Machine Collaborative Translation and Translators’ Digital Intelligence Literacy: A Comparative Analysis of Translations Produced by ChatGPT, DeepSeek, and Human Translators

Abstract

In the context of generative artificial intelligence (AIGC) reshaping the translation industry, human–machine collaboration has become an irreversible new paradigm for translating social-science and humanities academic texts. Taking Cultural History of Thirty Keywords as the corpus,, this study systematically compares three English translation versions produced by ChatGPT-5.4-nano, DeepSeek-V3.2, and human–machine collaboration. Based on a five-dimensional evaluation framework—cultural imagery, conceptual interpretation, logical expression, semantic-pragmatic correspondence, and norm-abiding output—the findings show that large language models perform exceptionally well in grammatical transformation and the transmission of surface-level information. However, when confronted with social-science texts that embed historical depth and value judgments, they remain clearly insufficient in terms of the depth of cultural interpretation, contextual fit, and ethical prudence. Translator intervention is therefore still necessary to accomplish “in-depth post-editing.” Building on these results, the paper proposes that translators enhance digital intelligence literacy in three areas: the ability for technological co-orchestration, the ability for critical review and optimization, and the ability for ethically grounded subject decision-making. This shift reframes translators from “tool users” to “top-level designers” and “quality reviewers,” thereby establishing a sustainable complementary mechanism between algorithms and the humanities.