基于人工智能指令语生成高三英语阅读理解试卷——以ChatGPT 5.6为例

Prompt-Driven Generation of Senior Tree English Reading Comprehension Tests—A Case Study of ChatGPT 5.6

摘要: 随着生成式人工智能的快速发展，大语言模型逐渐被应用于试卷开发，为教师开展自动化命题提供了新的可能。本研究旨在设计AI指令语，运用ChatGPT生成一份高三英语阅读理解试卷并探讨可行性及其存在的问题。研究依据《普通高中英语课程标准(2017年版2025年修订)》以及《中国英语能力等级量表》的相关阅读技能，确定本试卷拟考查的阅读能力构念，包括提取关键细节信息、概括主旨要义、推断词义、推断上下文隐含的语义逻辑关系等。在此基础上，研究者编制命题细目表，对语篇难度、体裁、题材、题型、阅读技能和认知层级作出规定，并将上述要求转化为AI指令语，运用ChatGPT (GPT-5.6)生成一套高三英语阅读理解试卷。随后，从iTEST英语文本指难针分析试题难度以及从构念效度视角和布鲁姆认知目标层次考察阅读技能可行性与一致性。结果显示，ChatGPT虽能快速生成结构完整、题型多样的试卷，但部分题目存在目标构念与实际考查不一致和题干表述晦涩等问题，因此，AI可以提高命题效率，但最终生成的试题仍需教师进行专业审核与修订。

Abstract: As generative artificial intelligence has developed rapidly, large language models have been applied to generate examination papers, which offers a promising prospect for teachers to implement automated item‑generation. This study aimed to use ChatGPT to generate English reading comprehension test for Chinese Grade 12 students by designing AI prompts and to explore its feasibility and existing problems. Based on the relevant reading skills outlined in the Curriculum Standards for General Senior High School English (2017 Edition, Revised in 2025) and the Chinese English Proficiency Scale, this study identified the reading ability constructs to be assessed in this test, including extracting key details, summarizing main ideas, inferring word meanings, and deducing implicit semantic and logical relationships in context. Based on this, the researcher developed a test-item specification table by specifying requirements for text difficulty, genre, topic, question types, reading skills, and cognitive levels, and then converted these requirements into AI prompts. ChatGPT (GPT-5.6) was then used to generate a set of senior high school English reading comprehension test papers. Subsequently, the difficulty of the test items was analyzed by using the iTEST English Text Difficulty Index, and the feasibility and consistency of the reading skills were examined from the perspectives of construct validity and Bloom’s taxonomy. The results showed that while ChatGPT can quickly generate test papers with complete structures and diverse question types, some problems exhibited inconsistencies between the intended constructs and what was actually assessed, as well as obscure question stems. Therefore, although AI can improve test-development efficiency, the final test items still require professional review and revision by teachers.