Neuro-Symbolic AI Approach to Online Hate Speech Detection

A recent preprint on arXiv (2605.16280) explores a hybrid neuro-symbolic AI framework designed for legal reasoning, focusing on the moderation of online hate speech. This research introduces Rulemapping, a visual logic-tree technique that integrates large language models with deterministic symbolic structures to enhance the assessment of illegality based on statutes. The goal is to mitigate 'scope drift,' a phenomenon where LLMs confuse moral offensiveness with legal illegality. The study frames online content moderation as a stand-in for extensive legal decision-making processes, such as mass administrative cases that require operators to evaluate thousands of cases each day under stringent legal criteria. It assesses if this hybrid approach can balance the transparency of symbolic systems with the adaptability of neural systems.