{"id":"W4399386809","doi":"10.1075/jslp.23033.joh","title":"Assessing pronunciation using dictation tools","year":2024,"lang":"en","type":"article","venue":"Journal of Second Language Pronunciation","topic":"Phonetics and Phonology Research","field":"Psychology","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec à Montréal; Concordia University","funders":"Social Sciences and Humanities Research Council","keywords":"Dictation; Pronunciation; Computer science; Rubric; Reliability (semiconductor); Natural language processing; Speech recognition; Test (biology); Artificial intelligence; Psychology; Linguistics; Mathematics education","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00132918,0.0001066839,0.0001772887,0.0003631973,0.00007689944,0.0003658115,0.0001234521,0.0001481924,0.001961797],"category_scores_gemma":[0.0002040318,0.00009199876,0.00009306733,0.0003028859,0.00002501652,0.001147571,0.00001867542,0.0004319487,0.00005889941],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002733161,"about_ca_system_score_gemma":0.0002587281,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002283044,"about_ca_topic_score_gemma":0.00002640885,"domain_scores_codex":[0.998601,0.0001797283,0.0004938223,0.0001609324,0.0003405765,0.0002239021],"domain_scores_gemma":[0.9990473,0.0001843373,0.0003301231,0.0001202135,0.000254635,0.00006338692],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.000188939,0.0003072178,0.003927278,0.000367146,0.0005519505,0.0005481475,0.09224027,0.0002052189,0.5301393,0.00161541,0.002966631,0.3669425],"study_design_scores_gemma":[0.004518917,0.001733475,0.8627349,0.001485064,0.0006046022,0.002453019,0.03492078,0.02773557,0.0437158,0.00725638,0.01162651,0.001214979],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9858698,0.005026444,0.004618862,0.0001969467,0.001491676,0.0001866007,0.00001048353,0.00003041412,0.002568803],"genre_scores_gemma":[0.997011,0.00001254727,0.001384123,0.00007025917,0.0007910729,0.000006265786,0.00002273336,0.00002312247,0.0006788515],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8588076,"threshold_uncertainty_score":0.9989505,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07403334702608028,"score_gpt":0.4177569239165626,"score_spread":0.3437235768904823,"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."}}