AIWizards at MULTIPRIDE: A Hierarchical Approach to Slur Reclamation Detection
The episode discusses a paper titled "AIWizards at MULTIPRIDE: A Hierarchical Approach to Slur Reclamation Detection." The hosts explain that researchers use a two-stage process: first, using a LLM to assign fuzzy labels about community membership from user profiles, and second, using a BERT-like model to detect if slurs are reclaimed. They conclude that integrating user identity information via a learned gating mechanism significantly improves accuracy in distinguishing harmful use from reclaimed language.