{"id":"W4285116834","doi":"10.18653/v1/2022.ltedi-1.6","title":"Detoxifying Language Models with a Toxic Corpus","year":2022,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; University of Toronto","funders":"Vector Institute; Canadian Institute for Advanced Research","keywords":"Debiasing; Computer science; Resource (disambiguation); Natural language processing; Process (computing); Artificial intelligence; Programming language; Psychology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.002931538,0.0009329448,0.0005966008,0.0009336976,0.0006469767,0.001366168,0.000959,0.0009468501,0.002565198],"category_scores_gemma":[0.01751075,0.0005617777,0.0006863886,0.0006699561,0.0008905584,0.002589663,0.002346243,0.002017666,0.001052554],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005934343,"about_ca_system_score_gemma":0.001145707,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001951588,"about_ca_topic_score_gemma":0.003453722,"domain_scores_codex":[0.9988644,0.0005623288,0.00007190255,0.0002183477,0.0002202838,0.00006276418],"domain_scores_gemma":[0.9931542,0.004496485,0.0003803646,0.00112552,0.0007490335,0.00009442316],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0007169407,0.0003055838,0.004117765,0.0004204219,0.0001575299,0.0007768481,0.0009009735,0.5851732,0.0698877,0.03480473,0.003117666,0.2996207],"study_design_scores_gemma":[0.00003061643,0.00009738217,0.0002639073,0.00002049162,0.00003088491,0.00007869893,0.00009608667,0.9625748,0.0192642,0.01526462,0.002257468,0.00002094494],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1900945,0.0005123971,0.8019194,0.0007285209,0.0000975016,0.0001241359,0.0002613956,0.003225164,0.003036994],"genre_scores_gemma":[0.6860712,0.0004794531,0.3062491,0.0004796713,0.00009598678,0.0002882696,0.001235636,0.0005272054,0.004573471],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.002931538,"threshold_uncertainty_score":0.01550364,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0142464298903263,"score_gpt":0.2480982024963722,"score_spread":0.2338517726060459,"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."}}