{"id":"W2072296938","doi":"10.1080/10627190802394222","title":"Validity Issues in Assessing English Language Learners' Language Proficiency","year":2008,"lang":"en","type":"article","venue":"Educational Assessment","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Memorial University of Newfoundland","keywords":"Accountability; Legislation; Language proficiency; Language assessment; Psychology; English language; Pedagogy; Medical education; Mathematics education; Political science; Medicine","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001150633,0.0001587299,0.0001894075,0.0002052734,0.0006371724,0.0001355757,0.0003349012,0.00009339899,0.002625905],"category_scores_gemma":[0.0004617556,0.0001656408,0.00007119505,0.0005360638,0.0002191661,0.0006533614,0.00004390754,0.0002995697,0.00007027048],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005886259,"about_ca_system_score_gemma":0.003765275,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005210772,"about_ca_topic_score_gemma":0.001463454,"domain_scores_codex":[0.9975992,0.0004591345,0.0003167619,0.0003643492,0.0008312202,0.0004293066],"domain_scores_gemma":[0.9986358,0.0006841822,0.0001331898,0.0002065699,0.000192556,0.0001476937],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000002684473,0.000860528,0.7677836,0.00001477494,0.000008443942,0.000008852683,0.1789653,0.00002118979,0.0002182531,0.04534436,0.006036397,0.0007355724],"study_design_scores_gemma":[0.0002125252,0.0000256733,0.7680146,0.00002851808,0.000007542934,0.000002701235,0.2043682,0.0000284185,0.00005974923,0.001216559,0.02577005,0.0002654959],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8618351,0.0002740707,0.00003465049,0.01162068,0.001304911,0.0002741687,0.000006340079,0.00006569202,0.1245843],"genre_scores_gemma":[0.973793,0.000108112,0.003378268,0.0002517315,0.002393031,0.0001205066,0.0001343175,0.00001541935,0.01980559],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1119579,"threshold_uncertainty_score":0.9982858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06769468706590388,"score_gpt":0.4642791324549266,"score_spread":0.3965844453890227,"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."}}