{"id":"W3200130628","doi":"10.18653/v1/2021.emnlp-main.122","title":"Conditional probing: measuring usable information beyond a baseline","year":2021,"lang":"en","type":"article","venue":"Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing","topic":"Topic Modeling","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; National Science Foundation","keywords":"Baseline (sea); USable; Representation (politics); Computer science; Word (group theory); Property (philosophy); Identity (music); Natural language processing; Artificial intelligence; Speech recognition; Mathematics","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":[0.001636078,0.0002062415,0.0003194269,0.0001802948,0.0001520078,0.0003591051,0.0009267881,0.0001171263,0.00003433302],"category_scores_gemma":[0.002299751,0.0001537158,0.00009881689,0.001155187,0.00007211012,0.001596427,0.0004073569,0.0006685505,0.000003507406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000972817,"about_ca_system_score_gemma":0.0003183481,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007011138,"about_ca_topic_score_gemma":0.000002072315,"domain_scores_codex":[0.9979332,0.0001127577,0.0005825192,0.0004243411,0.0006028978,0.0003443049],"domain_scores_gemma":[0.9983127,0.0002114967,0.0003517098,0.0002215985,0.0008316123,0.00007087248],"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":[0.0000288514,0.0001170812,0.001080561,0.0004767421,0.00001849079,0.000006424037,0.006868131,0.0002631702,0.08288294,0.03435062,0.00007671498,0.8738303],"study_design_scores_gemma":[0.0005747359,0.00002921559,0.0006044231,0.0008219858,0.00001582847,0.00005556452,0.001390572,0.7257151,0.2306124,0.03947164,0.0003370058,0.0003715196],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07204081,0.004151265,0.8936958,0.01162127,0.0007981763,0.000618497,0.000008526075,0.000190671,0.01687493],"genre_scores_gemma":[0.562755,0.00001000793,0.4364134,0.0006235649,0.00005015413,0.0000178486,0.000002746876,0.000006057494,0.0001211819],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8734587,"threshold_uncertainty_score":0.6268346,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05992423051901477,"score_gpt":0.3690637550504681,"score_spread":0.3091395245314533,"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."}}