DIRECTIVE SPEECH ACTS IN RESEARCHER–AI INTERACTION: A REFLEXIVE CASE STUDY OF ACADEMIC JOURNAL SUBMISSION

Abstrak

This study investigates the realisation and interactional functions of directive speech acts in researcher–AI communication through a reflexive case study of an academic journal submission. While directive speech acts have been extensively examined in human–human institutional interactions, limited empirical attention has been paid to how directives are constructed and deployed when the addressee is an artificial intelligence (AI) system operating within digitally mediated academic workflows. As AI systems increasingly assist scholars in writing and navigating editorial platforms, it is necessary to examine how action, authority, and epistemic responsibility are negotiated in human–AI interactions. This study addresses this gap by analysing a naturally occurring dialogue between a researcher and ChatGPT during the real-time submission of a manuscript to an international Q1 journal. The dataset comprised a complete 46-turn interaction sequence, from which 16 researcher-initiated directive moves were identified as primary units of analysis. Adopting a qualitative reflexive case study design, the analysis draws on speech act theory (Searle, 1976), interactional epistemics (Heritage, 2012), and relational work (Locher & Watts, 2008) to examine the linguistic realisation and functional deployment of directives across submission stages. The findings reveal that directives are predominantly realised through interrogative constructions rather than bald imperatives. WH-questions (e.g., “What should I do?”) and yes/no interrogatives pragmatically function as action-eliciting moves, particularly at moments of procedural uncertainty. These interrogatives enable the researcher to seek guidance while mitigating imposition, constructing AI as a procedural authority without an overt command. In addition to interrogatives, mitigated imperatives (e.g. “please explain”) and declarative problem reports (e.g., “I clicked OK, but there is no response”) operate as directive resources, demonstrating that directive force is interactionally achieved rather than strictly encoded in the grammatical mood. Functionally, four directive categories emerged: procedural guidance, explanatory mediation, confirmation for epistemic risk management, and evaluative positioning. Procedural guidance directives are clustered during technically opaque stages, such as PDF generation and final approval. Explanatory directives seek clarification of institutional prompts, positioning AI as an interpretive mediator. Confirmation directives appear before irreversible actions, reflecting heightened epistemic caution. A smaller subset of evaluative directives solicits normative stances regarding open-access decisions, extending human–AI interaction into moral reasoning. Directive density increases in phases characterised by institutional opacity, suggesting a correlation between procedural risks and directive production. Notably, directives are consistently mitigated through relational address terms and politeness markers despite the AI’s nonhuman status. This persistence of mitigation indicates that relational framing may function as a routinised communicative schema rather than solely as a face-protecting adjustment. Overall, this study argues that directives in researcher–AI interactions operate as mechanisms of institutional navigation and epistemic regulation rather than coercive command structures, contributing to empirical insights into the emerging field of human–AI pragmatics.