{"id":"W4245310439","doi":"10.31219/osf.io/yrvq5","title":"Accuracy of Automated Written Expression Curriculum-based Measurement Scoring","year":2020,"lang":"en","type":"preprint","venue":"","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; Institute of Education Sciences; U.S. Department of Education","keywords":"Curriculum; Word (group theory); Narrative; Computer science; Expression (computer science); Writing assessment; Natural language processing; Artificial intelligence; Curriculum-based measurement; Mathematics education; Psychology; Linguistics; Pedagogy; Curriculum development","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001474363,0.0001792647,0.0003096868,0.0001080648,0.000302754,0.0001389801,0.0004421337,0.0002408791,0.0001599192],"category_scores_gemma":[0.002559594,0.0001753417,0.0001427435,0.0002171329,0.00009683832,0.00007894166,0.0001801368,0.0003564561,0.0000206404],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002437397,"about_ca_system_score_gemma":0.001409134,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006208532,"about_ca_topic_score_gemma":0.0002846213,"domain_scores_codex":[0.9974464,0.0003349335,0.0004408108,0.0004213738,0.001079402,0.0002770979],"domain_scores_gemma":[0.9983317,0.0001548461,0.0004580299,0.0002940356,0.0005997709,0.0001615975],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0002005017,0.005476388,0.09751156,0.01374985,0.0005748273,0.00004030695,0.1299714,0.008279174,0.252614,0.03456553,0.3613754,0.09564096],"study_design_scores_gemma":[0.006034614,0.0005222439,0.04589061,0.06398861,0.001097241,0.000002345605,0.1689844,0.1046262,0.4962186,0.01931041,0.08571704,0.007607664],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7417297,0.00197291,0.007068005,0.03397244,0.005439782,0.003255466,0.00005588411,0.008235791,0.19827],"genre_scores_gemma":[0.9936662,0.00004859557,0.005401345,0.0001133111,0.0005898968,0.00005719188,0.00003079413,0.00001896161,0.00007367403],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2756584,"threshold_uncertainty_score":0.9385484,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09362668226288563,"score_gpt":0.371740380094759,"score_spread":0.2781136978318734,"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."}}