关联理论视角下大语言模型英译提示词设计探究——以《中国的能源转型》为例

A Study on Prompt Design for LLM-Assisted English Translation from the Perspective of Relevance Theory—A Case Study of China’s Energy Transition

摘要: 以ChatGPT为代表的大语言模型(LLMs)助手打破了传统翻译模式，对过往翻译活动及翻译评价也带来冲击。本研究以“最佳关联”为翻译评价标准，以《中国的能源转型》白皮书中特色文本为分析对象，围绕关联理论中的认知努力、认知语境、语境效果等核心概念，结合PICCO结构化提示词框架，设计英译提示词，并与朴素提示词(naive prompts)下生成的译文和官方译文作比较。分析发现：在关联理论指导下设计的结构化提示词能够在一定程度上引导模型助手显化译文主体，重组逻辑关系，处理隐喻表达等，降低目标读者的认知努力并增强译文的语境效果。研究结果初步表明“最佳关联”评价标准在翻译提示词设计中的适用性，也旨在为未来人机协同翻译实践提供参考。

Abstract: Large language model (LLM) assistants represented by ChatGPT have broken the traditional mode of translation and posed challenges to previous translation activities and translation evaluation. Taking optimal relevance as the criterion for translation evaluation and selected characteristic expressions from the white paper China’s Energy Transition as the object of analysis, this study designs English translation prompts based on the core concepts of Relevance Theory, including cognitive effort, cognitive context and contextual effects, and the PICCO structured prompting framework. It then compares the translations generated under naive prompts with those generated under PICCO prompts and the official translation. The analysis finds that structured prompts designed under the guidance of Relevance Theory can, to a certain extent, guide LLM assistants to make implicit subjects explicit, reorganize logical relations and handle metaphorical expressions, thereby reducing the target reader’s cognitive effort and enhancing the contextual effects of the translation. The findings preliminarily indicate the applicability of optimal relevance as an evaluative criterion in translation prompt design and provide a reference for future human-AI collaborative translation practice.