{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06188726,0.0006645472,0.0005915463,0.0005975334,0.001506006,0.00080962,0.0005048208,0.0006912081,0.0008659784],"category_scores_gemma":[0.01086393,0.0007281937,0.0003251091,0.002104161,0.001036803,0.001724448,0.00008173718,0.0009704262,0.00009924043],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003235803,"about_ca_system_score_gemma":0.004140561,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01726838,"about_ca_topic_score_gemma":0.03954547,"domain_scores_codex":[0.9451559,0.04969443,0.001111755,0.001207103,0.001787731,0.001043111],"domain_scores_gemma":[0.9928559,0.004962023,0.0005337362,0.0006250392,0.0007248838,0.0002984022],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00002613379,0.0006071826,0.03920039,0.0003724995,0.0001721814,0.000006663743,0.6559798,0.0324258,0.01097934,0.0200803,0.004141919,0.2360078],"study_design_scores_gemma":[0.0008170041,0.0002398721,0.6555179,0.001634302,0.001403069,0.00009663992,0.1174957,0.1271357,0.004324467,0.04813877,0.04208126,0.001115196],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.649007,0.008037942,0.3103821,0.02459695,0.002910825,0.001080123,0.00005399395,0.0004464278,0.003484592],"genre_scores_gemma":[0.8432263,0.002247575,0.1389354,0.0001764676,0.00155279,0.0004135733,0.00072841,0.0001279978,0.0125915],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6163176,"threshold_uncertainty_score":0.9997939,"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."}}