{"id":"W4290958425","doi":"10.1080/15434303.2022.2073886","title":"Developing a Scenario-Based English Language Assessment in an Asian University","year":2022,"lang":"en","type":"article","venue":"Language Assessment Quarterly","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Language assessment; Linguistics; Language proficiency; Psychology; Mathematics education; Sociology; Computer science","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008401234,0.0002346399,0.0002734024,0.0003467374,0.0008733291,0.0003067803,0.0004307335,0.00004170007,0.002972894],"category_scores_gemma":[0.000009299368,0.0002585398,0.00009371968,0.0001467065,0.00006157136,0.0004276185,0.00006055425,0.0009889649,0.00001006102],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007123305,"about_ca_system_score_gemma":0.0003705552,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002901461,"about_ca_topic_score_gemma":0.009161063,"domain_scores_codex":[0.9975241,0.0008623251,0.0002697662,0.0004425501,0.0004500706,0.0004512009],"domain_scores_gemma":[0.9992201,0.00008864541,0.0001543348,0.0003898806,0.00004854992,0.00009853939],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004240245,0.000570468,0.01077061,0.00006672284,0.00004902412,0.0009867682,0.8673573,0.0003600421,0.0003492221,0.08566462,0.0002832735,0.03349951],"study_design_scores_gemma":[0.001443496,0.0008800743,0.009062035,0.00005364723,0.00002917051,0.000003068094,0.9456292,0.003114318,0.000007014879,0.00007223843,0.03919638,0.0005093745],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9560176,0.00004801408,0.001063788,0.0006381067,0.0005079249,0.0002812499,0.00005653458,0.0003691141,0.04101768],"genre_scores_gemma":[0.9937437,3.553579e-7,0.002901983,0.0004564562,0.0003092686,0.00002449106,0.0003074609,0.00003902467,0.002217246],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08559238,"threshold_uncertainty_score":0.9999867,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01712562869737202,"score_gpt":0.2795077938743099,"score_spread":0.2623821651769379,"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."}}