{"id":"W2035007197","doi":"10.3138/cmlr.64.4.555","title":"Towards Defining a Valid Assessment Criterion of Pronunciation Proficiency in Non-Native English-Speaking Graduate Students","year":2008,"lang":"en","type":"article","venue":"Canadian Modern Language Review/ La Revue canadienne des langues vivantes","topic":"Phonetics and Phonology Research","field":"Psychology","cited_by":89,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"McGill University","keywords":"Pronunciation; Intelligibility (philosophy); Psychology; Language proficiency; Graduate students; Mathematics education; Linguistics; Computer science; Pedagogy","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001350511,0.0002707161,0.0006343063,0.0005149441,0.0001713701,0.00001969429,0.0005413987,0.0001598218,0.0001697601],"category_scores_gemma":[0.00076064,0.0002674116,0.0001142185,0.0005971123,0.0002327009,0.0001153467,0.00007741262,0.000517412,0.00001532243],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008071511,"about_ca_system_score_gemma":0.0004473292,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2504711,"about_ca_topic_score_gemma":0.3302398,"domain_scores_codex":[0.9972399,0.0005154625,0.0005949397,0.0005014493,0.0002092174,0.0009390235],"domain_scores_gemma":[0.9983274,0.0001792064,0.0002367318,0.0004761479,0.0003857151,0.0003947861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00003240987,0.0001978922,0.1457397,0.002568096,0.0001908792,0.00351501,0.8052963,0.0000227001,0.004125203,0.000625563,0.0004136524,0.03727262],"study_design_scores_gemma":[0.0006418648,0.0001814357,0.9878378,0.003110183,0.00005086724,0.0001560309,0.006208686,0.0005414248,0.00009076526,0.0001706971,0.0006231087,0.0003871163],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9721628,0.017558,0.0001100568,0.0001160183,0.0002580388,0.0009324967,0.0002093936,0.00002495267,0.008628242],"genre_scores_gemma":[0.9956913,0.0029679,0.0004037513,0.0002099165,0.00008439671,0.0001971133,0.0001430044,0.00004697263,0.0002556356],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8420981,"threshold_uncertainty_score":0.9999778,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05463594671732009,"score_gpt":0.3628819215498003,"score_spread":0.3082459748324802,"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."}}