{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001954458,0.0006847063,0.0009283808,0.0008434701,0.000496686,0.001574808,0.001210367,0.0009466661,0.001471608],"category_scores_gemma":[0.01178575,0.0003388404,0.000356372,0.0008128161,0.000586122,0.002838267,0.001231155,0.001704202,0.001146141],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004901467,"about_ca_system_score_gemma":0.0005945562,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002691412,"about_ca_topic_score_gemma":0.003370237,"domain_scores_codex":[0.9990994,0.0002278197,0.00004562786,0.0002928383,0.0001699347,0.0001644577],"domain_scores_gemma":[0.9969587,0.001196221,0.0003729781,0.0004951716,0.0007489586,0.0002279585],"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":[0.003892255,0.0002296092,0.11928,0.0003135963,0.0004403613,0.001166012,0.0009216905,0.02665633,0.0848094,0.01159966,0.0602296,0.6904616],"study_design_scores_gemma":[0.00005082615,0.0002262714,0.04245364,0.00007005658,0.0002564959,0.0006203429,0.0007832038,0.8649251,0.02725971,0.05336215,0.009909981,0.00008216089],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6712379,0.003717066,0.301698,0.007925926,0.00123208,0.00004572017,0.002119623,0.004867021,0.007156631],"genre_scores_gemma":[0.9734292,0.0003789286,0.02210959,0.0004537509,0.0002677896,0.00001759917,0.001324106,0.000316196,0.001702741],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.002691412,"threshold_uncertainty_score":0.01033622,"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."}}