{"id":"W2116334677","doi":"10.5539/elt.v3n1p113","title":"Direct Spoken English Testing Is still a Real Challenge to be Worth Bothering About","year":2010,"lang":"en","type":"article","venue":"English Language Teaching","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Section (typography); Presentation (obstetrics); Reliability (semiconductor); Psychology; Spoken language; Linguistics; Natural language processing; Computer science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04251283,0.000673538,0.00141966,0.001254118,0.001028296,0.003349538,0.002798262,0.001286558,0.004442465],"category_scores_gemma":[0.1369322,0.0003365315,0.0004810803,0.0005927164,0.003791409,0.004481529,0.002857644,0.002572376,0.003527084],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009214442,"about_ca_system_score_gemma":0.002125339,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001560706,"about_ca_topic_score_gemma":0.004373115,"domain_scores_codex":[0.9567166,0.02168343,0.003381409,0.002519155,0.01509824,0.0006011531],"domain_scores_gemma":[0.7346219,0.1876933,0.01111645,0.01985429,0.04225856,0.004455615],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00033619,0.0006711852,0.06940297,0.001885297,0.000157947,0.001060167,0.01161803,0.0006774035,0.01085634,0.006223868,0.01791242,0.8791983],"study_design_scores_gemma":[0.000256751,0.01214584,0.2910313,0.007525907,0.000288574,0.03089895,0.08313371,0.005425326,0.047177,0.09478506,0.426422,0.000909602],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"commentary","genre_scores_codex":[0.4798597,0.03077842,0.3263402,0.07961382,0.009919287,0.001985227,0.001134811,0.002995891,0.06737264],"genre_scores_gemma":[0.7245861,0.0106307,0.2248047,0.009432564,0.004529697,0.00139783,0.0006704539,0.000997193,0.02295078],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.04251283,"threshold_uncertainty_score":0.2248319,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02473490046603004,"score_gpt":0.2573063900499266,"score_spread":0.2325714895838965,"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."}}