Propasafe-Hybrid: A Text-Based Hybrid Propaganda Detection Tool (ILA 2026 Presentation)

This presentation introduces Propasafe-Hybrid, a hybrid system for sentence-level propaganda detection that combines offline transformer-based classification with selective large language model (LLM) explainability. The system operates through a two-stage pipeline: a local BERT-based classifier evaluates every sentence and filters out non-propagandistic content, while only high-confidence candidates are passed to an LLM for rhetorical technique labeling and explanation. This design enables cost-aware, privacy-conscious, and scalable analysis by reducing unnecessary LLM usage. Propasafe-Hybrid identifies propagandistic techniques such as loaded language, obfuscation, and appeal to fear, and generates concise natural-language rationales that make these techniques visible to readers. By bridging computational detection and linguistic interpretation, the system supports both large-scale rhetorical analysis and media literacy. This presentation also discusses methodological design, system architecture, limitations, and future directions, including multilingual extension and user evaluation. Presented at the 69th Annual Conference of the International Linguistic Association (ILA 2026), John Jay College of Criminal Justice, CUNY.