{"id":"W3018695736","doi":"10.18653/v1/2020.acl-srw.16","title":"Considering Likelihood in NLP Classification Explanations with Occlusion and Language Modeling","year":2020,"lang":"en","type":"preprint","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Banting and Best Diabetes Centre, University of Toronto; Bundesministerium für Bildung und Forschung","keywords":"Computer science; Artificial intelligence; Natural language processing; Language model; Context (archaeology); Occlusion","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0001753066,0.000162441,0.0001906624,0.0001661823,0.00005964623,0.0001942599,0.0003779096,0.000117493,0.000004107309],"category_scores_gemma":[0.00003323046,0.0001473898,0.00002051493,0.0001382274,0.00001172436,0.0002147301,0.001062139,0.0003922055,0.000003895854],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000606368,"about_ca_system_score_gemma":0.0001444342,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004463072,"about_ca_topic_score_gemma":0.0004166512,"domain_scores_codex":[0.9986593,0.00004106766,0.0002682464,0.0006474932,0.0002127442,0.0001711083],"domain_scores_gemma":[0.9992546,0.00004367014,0.00008009935,0.0004933575,0.00004725917,0.00008101839],"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":[0.00002704813,0.0001479006,0.006586459,0.0006544849,0.00008068168,0.000357597,0.05787612,0.6866227,0.007064507,0.1594194,0.0001118106,0.0810513],"study_design_scores_gemma":[0.0001749562,0.000009117467,0.0004043382,0.0001258273,0.000004505921,0.00001138579,0.0005632623,0.9921682,0.00009567169,0.006257962,0.000004782059,0.0001800145],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1380771,0.0001574765,0.8566727,0.003179243,0.00008034689,0.0002246645,0.000001562978,0.0001935672,0.001413355],"genre_scores_gemma":[0.7826071,0.0000327589,0.2170766,0.0002022654,0.00003318724,0.00001848417,0.00000933096,0.000008813032,0.00001151977],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6445301,"threshold_uncertainty_score":0.6010382,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0539578300993899,"score_gpt":0.2734041729442856,"score_spread":0.2194463428448957,"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."}}