From Linguistic Rights to Explainable Learning Analytics: A Cross-jurisdictional Policy Framework for Generative AI Feedback in Professional Interpreter Education

This methodological article proposes a cross-jurisdictional framework that links linguistic rights and procedural justice to explainable learning analytics for GenAI feedback in interpreter education. We define Procedural-Justice Consistency (PJC) indicators, speech-act preservation, presupposition fidelity, and tone/force consistency, and operationalise them through a minimal validation model. Using a bilingual dataset from criminal court interpreting (≈ 3,250 minutes of recordings; ≈112,500 transcribed words), we reanalyse lawyer questioning and witness testimony stylistics with interpretable decision trees and SHAP summaries, and triangulate findings with participant reflections on fairness and trust. Results show that light, discourse-oriented features can learn holistic “manner-of-speech” risks, while micro-errors remain better suited to human review. We provide force-aware prompts, indicator-linked explanations, and lean audit logging that map directly to institutional governance. The framework supports cross-jurisdictional alignment by keeping a stable PJC core and tunable thresholds for different regulatory profiles.