{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004272232,0.0008389969,0.0006941632,0.001094268,0.0008362728,0.001104427,0.0007463333,0.0006599454,0.001094152],"category_scores_gemma":[0.01623391,0.0003104157,0.0005035033,0.001146101,0.000544026,0.002454872,0.001749516,0.001099458,0.0005672492],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000648275,"about_ca_system_score_gemma":0.0009455655,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007853147,"about_ca_topic_score_gemma":0.017672,"domain_scores_codex":[0.9982111,0.0009891189,0.00009777729,0.0002643817,0.0002802784,0.0001573593],"domain_scores_gemma":[0.9901862,0.006344861,0.0008828929,0.00113197,0.001232534,0.0002215069],"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":[0.00105989,0.0003117787,0.1883346,0.0003801563,0.0004619137,0.0006083252,0.003674763,0.3355414,0.01955374,0.02452218,0.007216659,0.4183345],"study_design_scores_gemma":[0.00002800053,0.00009348388,0.02014456,0.00003303359,0.0000610635,0.0002474686,0.001220544,0.9478813,0.007470477,0.01879464,0.003973505,0.00005197468],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7023168,0.0005701412,0.2899924,0.0005063235,0.00006307322,0.00007768157,0.0004201418,0.001561986,0.004491512],"genre_scores_gemma":[0.9588619,0.00009810258,0.0386928,0.00008421125,0.00003480532,0.00003776191,0.0007932257,0.0002068275,0.001190367],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007853147,"threshold_uncertainty_score":0.02259403,"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."}}