Download Full PDF

季瑞骅

石昱欣

张天

赵文书

南京大学，中国
 

摘要

生成式人工智能正成为国际中文教育资源开发的重要支撑，但现有研究多关注生成效率与跨模态联动，对生成结果是否符合专业命题要求及其对学习者参与和表达动力的影响缺乏检验。本文围绕看图说话试题图像的自动生成，构建场景、人物、动作、表情、道具五要素生成模板，设计两种教师参与度不同的智能体，以 27 道试题为基准生成图片，从教师评价与学生体验两层面评估。结果显示，不同方案教育应用效果差异显著，较优方案接近原图水平，较弱方案在教师评价及学生情感接受、话题联想和表达意愿方面均不足。研究表明，生成式人工智能支持国际中文教育资源开发切实可行，但成效不仅取决于模型能力，还与试题需求描述的组织方式及教师参与程度密切相关。

 

关键词

生成式人工智能，看图说话，图像生成，学习者体验

 

Generative AI–Supported Image Generation for Picture-Description Speaking Test Items

 

Ruihua Ji

Yuxin Shi

Tian Zhang

Wenshu Zhao

Nanjing University, China

 

Abstract

Generative artificial intelligence (GenAI) is becoming an important support for resource development in international Chinese language education. However, existing research has focused mainly on generation efficiency and cross-modal content integration, with little examination of whether the generated outputs truly meet professional item-writing requirements or how they affect learners’ participation and motivation for oral expression. Against this background, this study investigates the automatic generation of images for picture-description speaking test items in international Chinese language education. A generation template comprising five elements, i.e., scene, person, action, expression, and prop, was constructed, and two generation agents involving different degrees of teacher participation were designed accordingly. Using 27 picturedescription speaking items as benchmarks, the two agents generated corresponding images, and an evaluation framework was established from two perspectives: professional teacher evaluation and learner experience. The results show significant differences among the generation schemes in educational effectiveness: the better-performing scheme was generally comparable to the original images, whereas the weaker scheme underperformed in teacher evaluations as well as in students’ affective acceptance, topic association, and willingness to speak. The findings indicate that GenAIsupported resource development for international Chinese language education is feasible in practice, but its effective application depends not only on model capability but also on how item requirement descriptions are organized and on the degree of teachers’ professional involvement.

 

Keywords

Generative artificial intelligence, picture description, image generation, learner experience