Learners’ Revision Awareness under Synthesized Image and Automated Text Feedback

Abstract

Although many studies have highlighted the benefits of incorporating images into language learning, few have directly compared textual and visual feedback in writing contexts. Recent advances in multimodal generative AI enable the systematic generation of context-sensitive visual feedback, allowing for a more balanced comparison between feedback modalities. This study investigates differences in user perceptions and revision awareness under image-based and textual feedback conditions. A total of 67 Japanese undergraduate students participated in a writing and revision task, and 43 completed a post-task questionnaire. The results indicated no significant differences between the two groups in perceived ease of use, usefulness, enjoyment, attitude, or intention to use the system, suggesting comparable user acceptance across modalities. However, analysis of revision behaviors revealed that learners receiving image-based feedback were significantly more likely to add detailed content to their texts, whereas no significant differences were observed in surface-level revisions such as grammatical or spelling corrections. These preliminary findings suggest that while textual and visual feedback are similarly accepted by learners, image-based feedback encourages meaning-level elaboration.