{"id":"W4395077771","doi":"10.1177/01427237241245500","title":"Evaluating English-language morphological awareness assessments","year":2024,"lang":"en","type":"article","venue":"First Language","topic":"Reading and Literacy Development","field":"Psychology","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Reading comprehension; Reading (process); Comprehension; Set (abstract data type); Psychology; Linguistics; Computer science; Cognitive psychology","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":[0.004020097,0.0006004291,0.0004362965,0.002163598,0.0002373414,0.0007609947,0.0004727454,0.0004724786,0.002433092],"category_scores_gemma":[0.02589823,0.0002169189,0.0004550586,0.0006096195,0.0003697599,0.001312241,0.0008465505,0.0004348532,0.0009166861],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002927729,"about_ca_system_score_gemma":0.0003894531,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00122426,"about_ca_topic_score_gemma":0.003199296,"domain_scores_codex":[0.9975382,0.0006850981,0.0004713881,0.0003327011,0.0008154424,0.0001572134],"domain_scores_gemma":[0.978538,0.009286783,0.003322911,0.0008499924,0.007113812,0.0008885034],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001236587,0.0007834157,0.541521,0.001033693,0.000236362,0.001901318,0.01289654,0.00140132,0.1487454,0.001033788,0.002110539,0.2871001],"study_design_scores_gemma":[0.00003536359,0.001923566,0.8656521,0.000194672,0.0002126913,0.002653445,0.003049393,0.003300693,0.1144066,0.001243337,0.007231865,0.00009637578],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9809489,0.0005273772,0.008314369,0.00007984797,0.00004277885,0.0001587938,0.0004478508,0.0003128725,0.009167258],"genre_scores_gemma":[0.9770542,0.0004632154,0.01840491,0.00008133549,0.00001980445,0.0001453266,0.0005117571,0.00007619834,0.003243316],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004020097,"threshold_uncertainty_score":0.02126056,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05795613321988979,"score_gpt":0.4397926648852263,"score_spread":0.3818365316653365,"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."}}