Abstract Participatory AI governance assumes that deliberation among diverse stakeholders can shape regulatory design. In practice, deliberative input affects governance only when it can be articulated as an institutional rule. This article examines the linguistic conditions under which normative claims acquire this form. Drawing on Crawford and Ostrom’s ADICO framework, Institutional Grammar is often treated as a neutral analytic tool, but it effectively defines the syntactic structure through which obligations become institutionally legible. Within this framework, a rule attributes responsibility to a defined actor (Attribute), encodes a deontic operator (must, may, must not), specifies an action (Aim), and defines the conditions under which the action applies (Condition), potentially accompanied by sanctions (Or else). In public deliberation, normative claims about justice, harm, or responsibility are typically expressed through descriptions of situations and shared experiences rather than through actor-directed prescriptions, making them difficult to translate into rule-centered institutional forms. To examine the implications of this gap, the study analyzes AI-mediated deliberations on AI harm prevention conducted in four typologically distinct languages (English, Basque, Czech, and Hebrew) and compares them with institutionally structured formulations derived from Article 1 of the EU AI Act. While all four languages produce ADICO-compatible rules under explicit rule-formulation constraints, unconstrained deliberation organizes obligation differently. Normative claims commonly appear through impersonal constructions, relational framings, and process-oriented accounts rather than actor-bound prescriptions. These findings reveal a structural gap between deliberative reasoning and the grammatical requirements of institutional rule-making. The article therefore proposes Institutional Grammar Pluralism, arguing that contemporary regulatory systems privilege a narrow syntactic format for recognizing obligation and thereby filter out significant dimensions of participatory reasoning. The challenge for participatory AI governance is thus not only to expand participation but also to reconsider the linguistic conditions under which normative claims can become regulatory commitments.
Denisa Kera (Mon,) studied this question.