The Rhetorical Features of Human-Driven and AI-Generated Texts from a Critical Discourse Analysis Perspective

ABSTRACT. This study explores the relation between the rhetorical features of human-authored academic prose and those of machine-generated academic prose through a method of qualitative text analysis. Utilizing van Dijk's (1995) socio-cognitive model and analyzing academic writing at three levels of text structure; synthesizing, semantics, and schematic through inside analysis and comparing/contrasting these levels of text to find similarities and differences. Examples of similarities/differences include originality and creativity, tone and voice, consistency and accuracy, objectivity and subjectivity, error patterns, discursivity, intertextuality, adaptability and the complexity of text generation. Results of the analysis reveal human-authored academic writing has a higher level of rhetorical complexity, contextual sensitivity, and deeper degrees of discursivity due to the writers' applications of critical thinking, varied stylistic choice, and intertextually sourced meanings constructed from multiple contexts including personal and sociocultural perspectives. On the other hand, machine-generated academic texts are grammatically correct and neutral in tone, are structurally correct, and as such have little contextual awareness. In a number of cases, machine-generated academic texts lack intertextuality; thus, the content produced lacks adaptive reasoning and is typically predictable and formulaic. In contrast, although AIs are capable of producing academic-text-related outputs, they cannot produce the depth of nuanced meaning making as found in the academic text created by human authors. Thus, the current study suggests that AIs may be used to assist and augment the human writing process with technical tasks such as brainstorming and paraphrasing.