{"id":"W4400898279","doi":"10.1111/jedm.12406","title":"Using Automated Procedures to Score Educational Essays Written in Three Languages","year":2024,"lang":"en","type":"article","venue":"Journal of Educational Measurement","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Mathematics education; Natural language processing; Computer science; Psychology; Linguistics; Artificial intelligence; Philosophy","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":[0.001102068,0.0001387808,0.0001619733,0.0006667032,0.00005993382,0.0002923589,0.0007003304,0.00004925052,0.00005033983],"category_scores_gemma":[0.0008987308,0.0001131882,0.00006712338,0.0007852433,0.00001965469,0.0006381894,0.00007587326,0.0002432007,0.0000129128],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007342731,"about_ca_system_score_gemma":0.003160986,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000397938,"about_ca_topic_score_gemma":0.00008423333,"domain_scores_codex":[0.9980659,0.00005126423,0.0004649051,0.0002267284,0.00099195,0.0001992339],"domain_scores_gemma":[0.9986936,0.000107439,0.0001549158,0.0001724255,0.0007522161,0.0001193593],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00008792432,0.001947171,0.03756291,0.001448024,0.0003726188,0.0001348939,0.007767938,0.002945095,0.2279967,0.5953418,0.09334922,0.03104571],"study_design_scores_gemma":[0.0008710664,0.0004514336,0.1640259,0.01567071,0.0001263697,0.002465985,0.0003995183,0.0379039,0.0263518,0.7483385,0.001868346,0.001526378],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.3117565,0.1270766,0.2627513,0.2867894,0.006964274,0.001536591,0.00001426534,0.000789537,0.002321594],"genre_scores_gemma":[0.7424302,0.000005522385,0.2567846,0.0002916866,0.0004289958,0.00001361513,9.531548e-7,0.00001024614,0.00003419501],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4306737,"threshold_uncertainty_score":0.5607453,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0603400789110201,"score_gpt":0.3634531269538653,"score_spread":0.3031130480428452,"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."}}