{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004128755,0.0001638101,0.0003423266,0.00009346927,0.0001714637,0.00008808885,0.00008629732,0.00005121009,0.0003458153],"category_scores_gemma":[0.0000989383,0.0001432787,0.00007000132,0.00004825547,0.0001237601,0.0001345746,0.0001086943,0.000239991,0.000001417177],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001212341,"about_ca_system_score_gemma":0.00004433061,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003459777,"about_ca_topic_score_gemma":0.00006023813,"domain_scores_codex":[0.9986771,0.0002245262,0.0002832602,0.000285415,0.0003225789,0.0002071731],"domain_scores_gemma":[0.9990442,0.000320947,0.0001758452,0.0002078638,0.0001870906,0.00006403036],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0001125436,0.002658588,0.151011,0.007956746,0.00123265,0.0004606934,0.3676867,0.000108994,0.01891949,0.3503135,0.004467481,0.09507168],"study_design_scores_gemma":[0.01048354,0.007330792,0.6022525,0.007714937,0.00117753,0.00009653534,0.2338427,0.007263461,0.05029793,0.01890583,0.0575092,0.003125091],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9629314,0.001658234,0.00003581882,0.00007061523,0.0001007016,0.0001522658,0.00002149211,0.00004035951,0.03498911],"genre_scores_gemma":[0.9977693,0.00008019611,0.0005689255,0.0001715562,0.0001286097,0.000006038197,0.00004517837,0.00001699253,0.0012132],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4512416,"threshold_uncertainty_score":0.5842734,"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."}}