{"id":"W4366728273","doi":"10.21432/cjlt28241","title":"Automated Scoring of Speaking and Writing: Starting to Hit its Stride","year":2023,"lang":"en","type":"article","venue":"Canadian Journal of Learning and Technology","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; Queen's University","funders":"","keywords":"Context (archaeology); Computer science; Section (typography); Notation; Process (computing); Data science; Linguistics; Programming language","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"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":[0.02760308,0.001025613,0.001729808,0.01008534,0.001390231,0.007659514,0.002869798,0.001495497,0.003998785],"category_scores_gemma":[0.1136908,0.0005195185,0.0006991252,0.007916173,0.003774503,0.007822584,0.003492916,0.001715127,0.00237469],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003168901,"about_ca_system_score_gemma":0.009426083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01321992,"about_ca_topic_score_gemma":0.0261796,"domain_scores_codex":[0.9597065,0.01973972,0.003083673,0.002170183,0.0146912,0.0006086275],"domain_scores_gemma":[0.7817769,0.1313469,0.01468449,0.01149306,0.05903718,0.00166146],"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":[0.00005502312,0.00003873797,0.00494524,0.003933137,0.00004689888,0.00003637431,0.002053512,0.0003615712,0.0004601106,0.005921303,0.005078916,0.9770692],"study_design_scores_gemma":[0.0001302126,0.001075249,0.1275848,0.04497427,0.0004476075,0.002229206,0.02292407,0.01098488,0.01264651,0.09760503,0.6786442,0.0007539868],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.1245994,0.4662618,0.2287905,0.04189258,0.004690608,0.0009220329,0.001646175,0.002720129,0.1284767],"genre_scores_gemma":[0.5341803,0.2279015,0.2102573,0.006228691,0.003482946,0.0009456343,0.001981459,0.0006091806,0.01441304],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02760308,"threshold_uncertainty_score":0.1459808,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01922601542231894,"score_gpt":0.3105721623918642,"score_spread":0.2913461469695453,"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."}}