Discourse-Aware Policy Analysis with Argumentation: A Hybrid LLM-Symbolic Framework for Disaster Governance

Short summary

Researchers present Apaf, a hybrid LLM-symbolic pipeline that applies critical discourse analysis to disaster-risk-reduction policy documents from the USA, UK, Canada, and Australia. The system classifies arguments into deliberative or managerial frames, then uses deterministic rules over LLM-extracted features to identify four frame-mediated relation subtypes (agency reduction, agenda shift, instrumental support, normative support). The resulting argument graphs are shown to be accurate, interpretable, and stable across jurisdictions, offering domain experts an inspectable alternative to end-to-end LLM summarization.

• •Apaf is a hybrid LLM-symbolic pipeline that operationalizes critical discourse analysis as a quantitative bipolar argumentation framework
• •It classifies policy arguments into deliberative vs. managerial frames and detects four frame-mediated relation subtypes via deterministic rules
• •Evaluated on 100 sub-documents of disaster-risk-reduction policy from four countries, producing accurate and cross-jurisdictionally stable argument graphs

Generated with AI, which can make mistakes.