Detecting Emergency Lawmaking with Machine Learning: A Multi-Source Model of Crisis-Responsive Legislative Initiatives

The paper introduces a machine-learning framework for identifying legislative initiatives that respond to crisis or emergency situations, even when the crisis is not explicitly named in the bill title or metadata. The model treats “crisis responsiveness” as a multi-dimensional signal expressed across the legislative text and its institutional context. On the content side, it captures (i) outlier provisions that depart from a jurisdiction’s baseline policy repertoire, (ii) temporariness and sunset structures (including renewal clauses, delegated extension mechanisms, and retroactive validation), and (iii) doctrinal signatures consistent with emergency-governance patterns, such as accelerated administrative discretion, derogation from ordinary safeguards, and the reconfiguration of oversight, procurement, or enforcement powers. These features are complemented by justification-statement indicators (problem framing, urgency language, proportionality claims, references to exceptional necessity), debate-level cues (compression of deliberation, cross-party rhetorical alignment or contestation over rights and proportionality), and procedural markers (fast-track pathways, shortened committee stages, bundling of heterogeneous measures, and unusual sequencing of readings). The architecture combines supervised classification with weak supervision from crisis-timeline anchors and anomaly detection to improve recall for novel emergencies. We validate the approach on a manually coded corpus spanning multiple crisis types (public health, security, economic shock, and natural disasters), reporting performance against expert labels and conducting ablation tests to quantify the contribution of each evidence channel. Beyond classification accuracy, we provide interpretable outputs -- provision-level rationales and “emergency pattern” scores -- designed for legal and political-science applications, including monitoring rights-restrictive drift, mapping opposition strategies, and comparing emergency lawmaking styles across jurisdictions and time.