{"id":"W3200855460","doi":"10.1111/flan.12571","title":"Effects of benchmarking and peer‐assessment on French learners' self‐assessments of accentedness, comprehensibility, and fluency","year":2021,"lang":"en","type":"article","venue":"Foreign Language Annals","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec à Montréal; Concordia University","funders":"","keywords":"Fluency; Pronunciation; Benchmarking; Psychology; Peer assessment; Self-assessment; Medical education; Mathematics education; Pedagogy; Linguistics; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004341958,0.0006736619,0.00068362,0.0004461684,0.0004727363,0.000562199,0.0004836102,0.0003454695,0.001406168],"category_scores_gemma":[0.0216041,0.000166265,0.0004413711,0.0001676903,0.000399316,0.0004619375,0.0008759809,0.0004373651,0.0002157708],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002982597,"about_ca_system_score_gemma":0.0005582588,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001014526,"about_ca_topic_score_gemma":0.001722516,"domain_scores_codex":[0.9944958,0.003886083,0.000216672,0.00033798,0.000843149,0.0002202458],"domain_scores_gemma":[0.9862066,0.008583248,0.001332646,0.0007882228,0.001423424,0.001665829],"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":[0.01096486,0.03063497,0.1310785,0.0007513518,0.0005799508,0.0005827834,0.01976237,0.002592078,0.0728767,0.0002579186,0.001192504,0.7287261],"study_design_scores_gemma":[0.001094191,0.09629948,0.816176,0.0002808449,0.0006273501,0.0006367526,0.01118226,0.005962277,0.06049796,0.0005446963,0.006454932,0.0002432871],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9989597,0.0000583097,0.0002545122,0.00001491248,0.000008109822,0.00003961266,0.000006154713,0.00001780844,0.0006408877],"genre_scores_gemma":[0.9980878,0.00007085226,0.00120716,0.00001022979,0.0000100028,0.0001148415,0.00002329892,0.000003802319,0.0004718557],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004341958,"threshold_uncertainty_score":0.02296275,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02665246318528015,"score_gpt":0.3227988995325941,"score_spread":0.2961464363473139,"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."}}