摘要：

目的 系统评估大语言模型（large language model，LLM）在辅助医学期刊英文摘要编校中的实际效能，并凝练可落地的编辑工作流程。 方法 选取11种代表性医学期刊的100篇中英文摘要，构建基于“术语-语法-规范-表达-逻辑”五维框架的英文摘要评估体系，对比3种LLM对医学期刊英文摘要问题的自动检测能力，并实施双盲人工复核。 结果 3种LLM累计发现1197个问题，人工审核后共采纳1113个（93%）修改意见。LLM能有效检出英文摘要中存在的多样化问题。这些问题以术语准确性为主（408个，占36.7%），其次是表达地道性、语法正确性和学术规范性。不同LLM在评估维度上呈现一定的互补优势，协同使用可全面提升英文摘要编校质量。 结论 LLM是赋能英文摘要编校的有效辅助工具，可提升工作效率与文本质量，实践策略包括优化输入指令、多模型互补集成、结构化评估框架、明确人机边界以及坚守伦理底线等。

关键词: 大语言模型, 医学期刊, 英文摘要, 编校质量, 人机协作

Abstract:

Purposes To systematically evaluate the effectiveness of large language models （LLMs） in editing English abstracts for medical journals and to distill actionable editorial workflow guidance. Methods We selected 100 paired Chinese-English abstracts from 11 representative medical journals and established a five-dimensional evaluation framework （terminology， grammar， conventions， expression， and logic）. We compared the automatic error-detection performance of three LLMs， followed by double-blind human adjudication. Findings The three LLMs collectively identified 1197 issues， of which 1113 （93%） were confirmed after double-blind review. Detected problems were dominated by terminological accuracy （408， 36.7%）， followed by idiomatic expression， grammatical correctness， and adherence to academic conventions. Different LLMs exhibited complementary strengths across evaluation dimensions； integrated use can enhance overall abstract quality. Conclusions LLMs are effective assistive tools for English-abstract editing in medical journals， improving editorial efficiency and textual quality. Recommended practices include refined prompt design， multi-model integration， structured assessment frameworks， clear human–machine role boundaries， and strict adherence to ethical standards.

Key words: Large language model, Medical journal, English abstract, Copyediting and proofreading quality, Human-machine collaboration