{"id":"W3177509237","doi":"10.1017/s0272263121000437","title":"WHAT THE RESEARCH SHOWS ABOUT WRITTEN RECEPTIVE VOCABULARY TESTING","year":2021,"lang":"en","type":"article","venue":"Studies in Second Language Acquisition","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University","funders":"","keywords":"Content (measure theory); Vocabulary; Psychology; Computer science; Linguistics; Mathematics","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002560922,0.0002521621,0.0003671892,0.0002684148,0.0005731232,0.0002155211,0.0003334822,0.0001796712,0.08583632],"category_scores_gemma":[0.001251229,0.0002039375,0.0001082985,0.001233061,0.0004475835,0.0004648754,0.0003270768,0.001022824,0.0006802809],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001938374,"about_ca_system_score_gemma":0.00007212392,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008975899,"about_ca_topic_score_gemma":0.00049082,"domain_scores_codex":[0.9957979,0.00180608,0.0004895753,0.000719892,0.0004272372,0.0007593559],"domain_scores_gemma":[0.9955704,0.002890696,0.000131674,0.0007867821,0.0005449736,0.00007549706],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000214079,0.0003745737,0.001449186,0.0002358536,0.0006941579,0.009056025,0.7747592,0.00003505567,0.01918869,0.003850487,0.05389431,0.1362484],"study_design_scores_gemma":[0.001009796,0.0001310446,0.01911068,0.0004943236,0.00002738802,0.0003886726,0.9687337,0.00002077178,0.001752557,0.0006722123,0.007330949,0.0003279002],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7081795,0.2431276,0.00001588233,0.0009477822,0.001952321,0.0003374122,0.00001757659,0.000102017,0.04531989],"genre_scores_gemma":[0.9622643,0.0004165325,0.0003324942,0.01810819,0.001398051,0.0002801332,0.0001216062,0.00005846227,0.01702023],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2540848,"threshold_uncertainty_score":0.9149994,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1284684008275283,"score_gpt":0.4416823069257763,"score_spread":0.313213906098248,"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."}}