{"id":"W4210824192","doi":"10.1558/jmtp.20395","title":"Language assessment frameworks as language policy","year":2021,"lang":"en","type":"article","venue":"Journal of Multilingual Theories and Practices","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; York University","funders":"","keywords":"Language policy; Language assessment; Language proficiency; Curriculum; Pedagogy; Language industry; Perception; Language education; Multilingualism; Psychology; Mathematics education; Comprehension approach; Linguistics; Sociology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002041792,0.000122812,0.0002432716,0.0001286943,0.0004060782,0.0003563819,0.0001959022,0.0002020468,0.00111217],"category_scores_gemma":[0.01749977,0.000103617,0.00009930549,0.0003006077,0.000220986,0.000516449,0.0000533322,0.000773486,0.00001291477],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006700037,"about_ca_system_score_gemma":0.002369494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003470627,"about_ca_topic_score_gemma":0.0006464437,"domain_scores_codex":[0.9981016,0.0006709149,0.0003824515,0.0001444886,0.0004420035,0.0002585344],"domain_scores_gemma":[0.996249,0.001820405,0.001041888,0.0001561422,0.0004704689,0.0002620544],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00014801,0.0003177194,0.002270454,0.00004790575,0.0001432236,0.0003078155,0.5446175,0.000007472735,0.001823022,0.2040666,0.0003273652,0.245923],"study_design_scores_gemma":[0.0006081288,0.0001149144,0.001659556,0.00009552336,0.00007657127,0.0001840445,0.6012416,0.00001989162,0.002739498,0.008831181,0.3842127,0.0002163499],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8813667,0.004245859,0.000045201,0.01459125,0.0008736147,0.00007159267,0.000006364417,0.00002099643,0.09877839],"genre_scores_gemma":[0.9812041,0.002616343,0.004110683,0.002070977,0.003431927,0.000001255314,0.00000252073,0.00001282668,0.006549403],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3838853,"threshold_uncertainty_score":0.9998009,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0416195823282649,"score_gpt":0.5476509293673879,"score_spread":0.506031347039123,"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."}}