{"id":"W3201341200","doi":"10.18653/v1/2021.emnlp-main.42","title":"Mitigating Language-Dependent Ethnic Bias in BERT","year":2021,"lang":"en","type":"preprint","venue":"Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing","topic":"Topic Modeling","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"Ministry of Science and ICT, South Korea; National Research Foundation of Korea; National Research Foundation","keywords":"Categorical variable; Computer science; Ethnic group; German; Turkish; Natural language processing; Metric (unit); Language model; Linguistics; Arabic; Gender bias; Word (group theory); Artificial intelligence; Psychology; Machine learning; Sociology; Social psychology","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.00341118,0.0006098373,0.001083116,0.0004929094,0.0001040924,0.0006559828,0.003456197,0.0006001926,0.0000280216],"category_scores_gemma":[0.004886336,0.0004694732,0.000294429,0.001628519,0.0001406528,0.0005375389,0.004271333,0.004159542,0.000001511088],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000362532,"about_ca_system_score_gemma":0.0006179531,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002725924,"about_ca_topic_score_gemma":0.00011052,"domain_scores_codex":[0.9949394,0.0004199713,0.001263432,0.00159534,0.001028883,0.0007529196],"domain_scores_gemma":[0.9971314,0.0005690432,0.0009790106,0.0006969838,0.0005005389,0.000123007],"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":[0.00002406915,0.0001706505,0.00330039,0.001405034,0.00002994481,0.00005567894,0.04818734,0.0003724118,0.0475841,0.0009772104,0.000007799454,0.8978854],"study_design_scores_gemma":[0.0008848872,0.0000528455,0.004773921,0.01255929,0.00005037079,0.00005099981,0.01708641,0.8057991,0.1461475,0.01127354,0.000006764019,0.001314376],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.921169,0.02498014,0.04105638,0.005124675,0.001551102,0.001009563,0.000005599928,0.0001901768,0.004913353],"genre_scores_gemma":[0.6134404,0.00004622275,0.3857994,0.0003651028,0.0001038441,0.00005339186,0.000002216561,0.0000274414,0.0001619281],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.896571,"threshold_uncertainty_score":0.9997757,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1389738437343006,"score_gpt":0.44294323317028,"score_spread":0.3039693894359795,"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."}}