{"id":"W4414211502","doi":"10.7202/1119632ar","title":"La grille critériée : toujours une bonne méthode d’évaluation ? Comparaison de différentes méthodes d’évaluation des éléments paraverbaux dans les productions orales d’élèves de 11-12 ans","year":2024,"lang":"fr","type":"article","venue":"Mesure et évaluation en éducation","topic":"French Language Learning Methods","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"General interest; Notation; World wide","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.02790929,0.001782158,0.00146196,0.003953041,0.001412508,0.006188824,0.001598512,0.001116104,0.01110911],"category_scores_gemma":[0.08151022,0.001050565,0.001586287,0.003137664,0.002146363,0.003086125,0.003046986,0.001701007,0.003836808],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001863349,"about_ca_system_score_gemma":0.003117127,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0064721,"about_ca_topic_score_gemma":0.01333045,"domain_scores_codex":[0.9684706,0.0164461,0.003309646,0.003476941,0.007668812,0.000627882],"domain_scores_gemma":[0.9212595,0.04772342,0.003794335,0.005756707,0.02060042,0.0008656589],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002151721,0.0001928363,0.03932137,0.005123911,0.0008043516,0.0002373768,0.03097891,0.002149488,0.08069404,0.01768638,0.004019467,0.8166402],"study_design_scores_gemma":[0.0007241618,0.0042646,0.4061578,0.006219184,0.002012481,0.001905898,0.04379962,0.03853328,0.1919054,0.03370339,0.2690841,0.001690083],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2121368,0.006133189,0.7232073,0.001192571,0.001035264,0.002883494,0.001244712,0.002116012,0.05005065],"genre_scores_gemma":[0.3639337,0.002177472,0.6072994,0.0002956551,0.0001205278,0.003592015,0.0009647679,0.002006803,0.01960969],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02790929,"threshold_uncertainty_score":0.1476002,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.155698039103682,"score_gpt":0.4375736953203973,"score_spread":0.2818756562167153,"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."}}