A Study on the Realization of Interpersonal Meaning in Generative AI-Generated Essays Based on Persona Prompts

Abstract

This study utilized persona prompts with varying writer and reader occupations and ages to generate argumentative texts using generative AI, examining how tenor differences were realized in interpersonal lexicogrammar from a systemic functional perspective. A total of 70 texts were generated across six combinations of three writer and two reader personas, plus one unspecified condition. While the variation of types of modality across conditions was not substantial, its realization and orientation showed clearer differences. The unspecified condition had the highest density of modalities but the lowest rate of metaphorical realization. Objectified metaphorical modality was more common among policy-maker readers, whereas subjectified metaphorical modality was more frequent among middle school readers. These findings do not represent the linguistic characteristics of actual social groups. Rather, they suggest that AI-generated texts reflect social representations of expertise, status, age, and participant relations through lexicogrammatical choices. For prompt literacy education, the findings highlight the need to specify desired linguistic features and examine how AI fills in unspecified elements. For grammar education, they suggest moving beyond formal correctness to analyzing lexicogrammatical choices in relation to context and interpersonal meanings.