Engagement Functions of AI-Generated Text from the Perspective of Appraisal Theory

Abstract

Within the framework of Appraisal Theory, this corpus-based study investigates the use and functions of engagement strategies in human-written and AI-generated academic discourse. A total of 190 academic texts (95 human-written from the BAWE corpus and 95 generated by DeepSeek-V3) were analyzed. A mixed-methods approach was adopted, involving both quantitative and qualitative analyses of Dialogic Expansion and Dialogic Contraction strategies. The results show that human writing contains significantly more engagement resources than AI-generated texts overall. This difference holds for both Dialogic Expansion and Dialogic Contraction. AI-generated texts also exhibit frequent vague Attribution and localized overuse of Proclaim resources. These patterns are associated with weaker stance construction, less effective dialogic negotiation, and weaker reader-oriented persuasion. Based on the observed differences, this study offers implications for academic writing instruction and human-AI collaboration.