{"id":"W3019416653","doi":"10.18653/v1/2021.acl-long.416","title":"StereoSet: Measuring stereotypical bias in pretrained language models","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":102,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Canadian Institute for Advanced Research","funders":"","keywords":"Stereotype (UML); Race (biology); Set (abstract data type); Scale (ratio); Language model; Computer science; Measure (data warehouse); Natural language processing; Artificial intelligence; Psychology; Cognitive psychology; Social psychology; Data mining; Geography","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003901105,0.0008149431,0.0007968158,0.001462239,0.0005171322,0.001188396,0.001193027,0.001216418,0.003500977],"category_scores_gemma":[0.02043377,0.0004472496,0.0005572517,0.001374069,0.0004011138,0.002855204,0.001738355,0.001690464,0.001454289],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006505046,"about_ca_system_score_gemma":0.0006612309,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003879194,"about_ca_topic_score_gemma":0.006937301,"domain_scores_codex":[0.9978973,0.0008545154,0.0001185745,0.0006324283,0.0003614269,0.0001357195],"domain_scores_gemma":[0.9905591,0.006257132,0.0005054501,0.00153713,0.0008223476,0.0003189739],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.007627557,0.000942898,0.1374095,0.000971775,0.002024877,0.0002968036,0.001735034,0.07888819,0.04030051,0.008536872,0.05903788,0.6622281],"study_design_scores_gemma":[0.0004531708,0.0008055322,0.0803431,0.0001227534,0.000412114,0.0005406787,0.0007799158,0.8472123,0.03010189,0.02996169,0.009092116,0.0001747498],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7834313,0.002683966,0.1884729,0.0005544223,0.0005632018,0.0002151468,0.009603158,0.007632134,0.006843806],"genre_scores_gemma":[0.9386572,0.0003383627,0.04138286,0.0002095349,0.0001513136,0.000186872,0.01650101,0.0007979414,0.001774949],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003901105,"threshold_uncertainty_score":0.02063125,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1300569881173056,"score_gpt":0.2873309265116926,"score_spread":0.157273938394387,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}