{"id":"W4401042513","doi":"10.18653/v1/2024.naacl-demo.7","title":"jp-evalb: Robust Alignment-based PARSEVAL Measures","year":2024,"lang":"en","type":"article","venue":"","topic":"Fuzzy Logic and Control Systems","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of British Columbia","keywords":"Parseval's theorem; Computer science; Mathematics; Fourier transform; Fourier analysis","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006773568,0.002637938,0.002504591,0.005651014,0.001894194,0.004506422,0.005206347,0.002828784,0.009739972],"category_scores_gemma":[0.02665647,0.001431603,0.001690381,0.004187365,0.001502842,0.007301233,0.005863826,0.004199637,0.009789902],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001049755,"about_ca_system_score_gemma":0.002529228,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002373051,"about_ca_topic_score_gemma":0.003154978,"domain_scores_codex":[0.9910926,0.002562016,0.0007895776,0.001739938,0.003325714,0.0004900736],"domain_scores_gemma":[0.9910222,0.003441819,0.0006205178,0.002413175,0.00223152,0.0002707923],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001572965,0.0003466668,0.002370487,0.000776812,0.0005127813,0.0002252534,0.0004224764,0.02860285,0.01836921,0.05827343,0.1417174,0.7468097],"study_design_scores_gemma":[0.0003951754,0.0004762758,0.003404493,0.0001958395,0.0001927574,0.0006406083,0.0002785918,0.7131569,0.05118775,0.1452083,0.0845594,0.0003040539],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008549549,0.001078691,0.9221937,0.0002729999,0.000431784,0.0001625345,0.003008992,0.06109229,0.003209352],"genre_scores_gemma":[0.09252908,0.0004520467,0.8699921,0.0003477977,0.0003033821,0.0006711429,0.01816037,0.01411198,0.003432115],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.009739972,"threshold_uncertainty_score":0.03582245,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03207759730768253,"score_gpt":0.2273532539965208,"score_spread":0.1952756566888383,"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."}}