<aside> ℹ️ This session features at PAIRS 2026 Online on 17th February 2026 13:30 UTC. Registered participants will receive zoom links to join the session via e-mail.
</aside>
[https://drive.google.com/file/d/1nkn6SU9XrgaOvHegEGcX45Vltu3633RQ/view](https://drive.google.com/file/d/1nkn6SU9XrgaOvHegEGcX45Vltu3633RQ/view)
This paper examines why participatory AI governance so often fails to translate into institutional rules, even when deliberation produces rich normative consensus. We investigate this problem by analyzing how normativity is linguistically articulated before rule formation, and how these forms align (or fail to align) with Institutional Grammar (ADICO), the dominant framework for institutional rule representation. Using controlled multilingual deliberations conducted in English, Basque, Czech, and Hebrew, we asked AI agents to deliberate on the question “How to prevent AI harm?” without prompting rule-like formulations. Final-round proposals (“how things should be”) were then systematically compared to EU AI Act Article 1, treated as a canonical example of ADICO-structured institutional normativity. Across all four languages, including English, the language in which ADICO was developed, deliberative outputs consistently resist decomposition into attributed, deontically qualified actions with sanctions. Instead, normativity is articulated through states, relationships, processes, and collective orientations, with agency distributed, implicit, or backgrounded. We show that this resistance is not noise or cultural idiosyncrasy, but a stable grammatical pattern shaped by linguistic affordances and discourse structure. Institutional Grammar successfully captures normativity after institutional translation, but systematically misrecognizes normativity as it emerges in deliberation. On this basis, we argue that participatory failure is not primarily a problem of inclusion or deliberative quality, but of institutional legibility: institutions recognize only one grammar of normativity. The paper concludes by proposing Speculative Institutional Grammar(s) and Pluralism as a design principle for participatory AI governance, and motivates a workshop-based format to explore how alternative normative grammars can be preserved, translated, or negotiated rather than prematurely compressed into rule form.