{"id":"W3175560839","doi":"10.18653/v1/2021.acl-long.325","title":"How is BERT surprised? Layerwise detection of linguistic anomalies","year":2021,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; University of Toronto","funders":"Social Sciences and Humanities Research Council of Canada; Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Computer science; Volume (thermodynamics); Linguistics; Joint (building); Computational linguistics; Natural language processing; Artificial intelligence; Philosophy; Engineering; Physics; Structural engineering","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00006963856,0.00006138818,0.00009960594,0.00003999071,0.00003543357,0.00009217297,0.0002205127,0.00003429592,0.00002094041],"category_scores_gemma":[0.0001570252,0.00005842567,0.00004582752,0.0001870621,0.00001236073,0.0001234541,0.0001386994,0.00004924632,0.00000604308],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000142645,"about_ca_system_score_gemma":0.0000505005,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004630096,"about_ca_topic_score_gemma":0.0000446831,"domain_scores_codex":[0.9993641,0.00002356797,0.0001166629,0.0002284518,0.0001517873,0.0001154195],"domain_scores_gemma":[0.9992633,0.00005315088,0.00004109436,0.0004272917,0.0001799213,0.0000352337],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00001400543,0.0003136054,0.01099342,0.0003368261,0.0001481532,0.0002415258,0.02100691,0.0005565076,0.2796156,0.1595537,0.0007612502,0.5264585],"study_design_scores_gemma":[0.0002250383,0.0000386471,0.001066615,0.00002207207,0.000008590257,0.00002482188,0.0001466402,0.234623,0.7495478,0.00819753,0.005914049,0.0001852291],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0842524,0.0002547337,0.9105391,0.0007303538,0.0003753929,0.00003226589,5.541402e-7,0.00007880475,0.00373641],"genre_scores_gemma":[0.9447157,0.000009552692,0.05165927,0.0001752587,0.00006126009,0.000001846848,2.553123e-7,0.000003453956,0.003373413],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8604633,"threshold_uncertainty_score":0.238253,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01879292536861095,"score_gpt":0.2284517515391103,"score_spread":0.2096588261704994,"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."}}