{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02128162,0.0005454235,0.0006455831,0.002621906,0.0003305221,0.001507981,0.0009503207,0.000467511,0.001088268],"category_scores_gemma":[0.1027879,0.0002917103,0.000333018,0.001662673,0.0004451779,0.0009377371,0.001491909,0.0003732115,0.001019137],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004912257,"about_ca_system_score_gemma":0.0005887844,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002478464,"about_ca_topic_score_gemma":0.003757104,"domain_scores_codex":[0.9744989,0.01299545,0.002237253,0.003230101,0.006473709,0.0005646432],"domain_scores_gemma":[0.9108998,0.04337325,0.01257838,0.009787687,0.0223869,0.0009740327],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000782742,0.0003415301,0.6564419,0.0001938939,0.0002079065,0.00005282017,0.001972976,0.002935129,0.008339801,0.0004100496,0.001863033,0.3264583],"study_design_scores_gemma":[0.00007562163,0.001078667,0.9453582,0.00008697459,0.00006107474,0.0001683933,0.0007175581,0.03431036,0.01363444,0.0006055379,0.003832048,0.00007112524],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9695404,0.0002420071,0.0234801,0.0001026277,0.0000599851,0.0003507065,0.0008214836,0.0006748813,0.004727806],"genre_scores_gemma":[0.9818934,0.00007940573,0.01562844,0.00005361543,0.00002098089,0.0002859289,0.0006243987,0.00005102903,0.001362797],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02128162,"threshold_uncertainty_score":0.1125493,"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."}}