厦门大学学报（自然科学版）2025，Vol.64Issue(6)：958-969,后插1-后插7,19.DOI:10.6043/j.issn.0438-0479.202412010
基于大语言模型的智能译后编辑系统构建与应用
Construction and application of intelligent post-editing editing system based on large language model
摘要
Abstract
[Objective]Although neural machine translation(NMT)has become the mainstream method for current machine translation(MT),the need to post-edit outputs of NMT models in critical scenarios to correct errors and improve quality remains.[Methods]Herein we propose an automatic translation post-editing system called Smart Suggest AutoEdit(SSAE),based on the large language model GPT-4o mini.The SSAE system ensures consistency and accuracy in terminology usage during the translation process by introducing term constraints and a self-feedback mechanism for multidimensional translation suggestions,thereby improving the overall quality of the critical scenarios translation.[Results]Experimental results show that the system performs meaningful and reliable edits on translations,hence helping to enhance their overall quality and eliminating major errors in the translations.When handling texts in specialized fields,the system performs very satisfactorily,thus reducing terminology errors and consistency issues significantly.[Conclusion]Notably,our proposed automatic translation post-editing system has achieved state-of-the-art performance in the WMT-23 terminology translation tasks for Chinese-English,German-English,and English-Czech.关键词
神经机器翻译/大语言模型/自动译后编辑Key words
neural machine translation/large language models/automatic post-editing分类
信息技术与安全科学引用本文复制引用
赵泽龙,朱俊国..基于大语言模型的智能译后编辑系统构建与应用[J].厦门大学学报（自然科学版）,2025,64(6):958-969,后插1-后插7,19.基金项目
国家自然科学基金地区基金(62166022) （62166022）
云南省"兴滇英才支持计划"(KKXX202403023) （KKXX202403023）
