{"id":"W4244295316","doi":"10.31219/osf.io/gcetv","title":"Validity of Automated Text Evaluation Tools for Written-Expression Curriculum-Based Measurement: A Comparison Study","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Institute of Education Sciences; University of Houston; U.S. Department of Education","keywords":"Metric (unit); Computer science; Measure (data warehouse); Expression (computer science); Natural language processing; Curriculum-based measurement; Artificial intelligence; Machine learning; Curriculum; Sample (material); Data mining; Medical physics; Psychology; Medicine; Engineering; Curriculum development","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":[],"consensus_categories":[],"category_scores_codex":[0.01065892,0.000221522,0.0005382936,0.0001434855,0.0005264499,0.0004792625,0.0003496863,0.0003101963,0.0001980456],"category_scores_gemma":[0.006525791,0.0002195753,0.0002149785,0.0002641337,0.0000784954,0.0001582322,0.0001062597,0.0002740113,0.000002929763],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005118012,"about_ca_system_score_gemma":0.002340212,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0043357,"about_ca_topic_score_gemma":0.003244216,"domain_scores_codex":[0.9944251,0.001847625,0.0007432225,0.000587284,0.00209435,0.0003024623],"domain_scores_gemma":[0.9946082,0.0005184144,0.000672721,0.0004265975,0.003673283,0.0001007933],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003205533,0.05905994,0.4418157,0.006945289,0.0009001475,0.000004269299,0.2206648,0.02900117,0.02474919,0.001726888,0.08370803,0.1311041],"study_design_scores_gemma":[0.00633166,0.0008552557,0.08647984,0.006719238,0.001948387,3.476597e-7,0.5099379,0.3542676,0.02757745,0.0009603136,0.002983179,0.001938843],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9850442,0.0001918781,0.003748114,0.0006282181,0.001098205,0.004190414,0.00002252854,0.0005217472,0.004554728],"genre_scores_gemma":[0.994664,0.000006402051,0.003832691,0.00002252513,0.0003424524,0.0007779058,0.0002883255,0.00001982058,0.00004588781],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3553358,"threshold_uncertainty_score":0.8954022,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2613056123284385,"score_gpt":0.4539135859038707,"score_spread":0.1926079735754322,"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."}}