Annotator Positionality as Signal: Psychometric Weighting for Anti-Autistic Ableism Detection

In three linesarXiv study on anti-autistic ableism detection in LLMs. Researchers propose a psychometrically-weighted evaluation framework anchored in annotator positionality. Models produce harmful outputs, mislabel community-reclaimed language, and rely on surface-level keyword matching rather than context (speaker identity, in-group solidarity vs. out-group harm).

Summary generated by Claude — human-verified