Enhancing accuracy and acceptability of English-Chinese generative AI translation in clinical guidelines: The importance of post-editing

Abstract The accuracy and acceptability of clinical guideline translations from English to various languages are critical to global healthcare, enabling medical practitioners in regions such as China to implement evidence-based recommendations effectively. Despite recent advancements in generative AI translation (GenAIT), its reliability in this specialized domain of clinical translation remains underexplored. This study assesses the necessity of human post-editing (PE) in improving GenAIT outputs of English-to-Chinese (E-C) clinical guideline translations and investigates its impact on readers' cognitive load. Clinical guidelines were translated using ChatGPT 3.5 and DeepSeek, then manually post-edited by professional translators. Errors were collected and categorized, revealing frequent semantic errors (42.6%), followed by syntactic (38.6%) and pragmatic (18.8%) issues. A qualitative assessment using a modified NASA-TLX scale demonstrated that post-editing significantly enhanced translation accuracy and reduced cognitive load, thereby improving comprehension. These findings underscore the value of integrated AI-human workflows, demonstrating that combining GenAIT's computational efficiency with expert PE is crucial for producing reliable, high-quality clinical translations.