{"id":"W2965600044","doi":"10.1111/modl.12579","title":"Weighing Up Exercises on Phrasal Verbs: Retrieval Versus Trial‐and‐Error Practices","year":2019,"lang":"en","type":"article","venue":"Modern Language Journal","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":68,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"Victoria University of Wellington; University of Victoria","keywords":"Implementation; Meaning (existential); Computer science; Test (biology); Significant difference; Psychology; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.005526667,0.001049861,0.00127103,0.000757302,0.000422986,0.000959155,0.001641648,0.000995202,0.003313052],"category_scores_gemma":[0.03736,0.0004110537,0.0003395374,0.0004376762,0.0009102039,0.001415747,0.001324192,0.0007985857,0.0007796877],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000286037,"about_ca_system_score_gemma":0.0004317725,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002200167,"about_ca_topic_score_gemma":0.0003015559,"domain_scores_codex":[0.9951736,0.002220661,0.0007195681,0.0007726828,0.0008926984,0.0002208361],"domain_scores_gemma":[0.964353,0.02614521,0.003959896,0.002957804,0.00146684,0.001117208],"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":[0.02750725,0.05217065,0.03991683,0.002229315,0.0003120143,0.0003729645,0.007472324,0.004407755,0.1942065,0.001472467,0.001083192,0.6688488],"study_design_scores_gemma":[0.009736572,0.2556313,0.2380236,0.001582681,0.001414916,0.001666947,0.007086618,0.04350399,0.4195035,0.00858632,0.01273336,0.0005302259],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9957793,0.00009937953,0.002952884,0.00003444108,0.00001868902,0.0002364554,0.00001448109,0.00006231343,0.0008020891],"genre_scores_gemma":[0.9804901,0.0001959425,0.01642771,0.00005631565,0.00003821709,0.0007305747,0.00005935391,0.00003530858,0.001966334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005526667,"threshold_uncertainty_score":0.02922815,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05336397463982574,"score_gpt":0.3910841418882373,"score_spread":0.3377201672484116,"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."}}