{"id":"W1889588694","doi":"10.24908/pceea.v0i0.4785","title":"EVALUATING THE USABILITY OF AN AUTOMATED COURSE-SPECIFIC VOCABULARY TO MEASURE LANGUAGE PROFICIENCY IN ORAL PRESENTATIONS","year":2013,"lang":"en","type":"article","venue":"Proceedings of the Canadian Engineering Education Association (CEEA)","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science; Usability; Vocabulary; Natural language processing; Field (mathematics); Measure (data warehouse); Artificial intelligence; Course (navigation); Term (time); Word (group theory); Linguistics; Human–computer interaction; Engineering; Data mining; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.001096679,0.00008967407,0.0001074049,0.0001994723,0.0001340238,0.00008774216,0.0007753132,0.00008215737,0.0000135289],"category_scores_gemma":[0.0008838866,0.00007390991,0.00003418387,0.001072326,0.00002364479,0.0004289683,0.00004460248,0.000177363,0.000005403106],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007147872,"about_ca_system_score_gemma":0.0009734555,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006172314,"about_ca_topic_score_gemma":0.002495255,"domain_scores_codex":[0.9989049,0.000027012,0.0002869448,0.0001954501,0.0003732114,0.0002124345],"domain_scores_gemma":[0.9986733,0.00008104138,0.0002345319,0.0002109824,0.0007055468,0.00009460031],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000004536212,0.001154565,0.6811763,0.0002384835,0.0001162906,1.207624e-7,0.03441349,0.0140371,0.05931467,0.1545433,0.03116056,0.02384059],"study_design_scores_gemma":[0.00007911136,0.00003574875,0.9367639,0.00004740672,0.000007592595,0.000001194589,0.001577004,0.05779384,0.002789502,0.0006484304,0.0001524813,0.0001037397],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9822851,0.00004586919,0.0001308893,0.01591446,0.0004420709,0.0007092144,0.000004112535,0.00009198391,0.0003762709],"genre_scores_gemma":[0.9894363,6.243051e-7,0.009997703,0.0001212111,0.00003081867,0.0002119453,0.000003298585,0.000007469795,0.0001906851],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2555876,"threshold_uncertainty_score":0.9330733,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0220139720686023,"score_gpt":0.3259191465731555,"score_spread":0.3039051745045532,"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."}}