{"id":"W3011618373","doi":"10.1075/jslp.19015.tay","title":"Testing the malleability of teachers’ judgments of second language speech","year":2020,"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":"University of Calgary; Concordia University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Psychology; German; Vowel; Intonation (linguistics); Linguistics; Second language; First language; Context (archaeology); Consonant","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00139746,0.0001045705,0.0003013205,0.00008196448,0.00003761772,0.00001258971,0.0003426279,0.0001170808,0.005431641],"category_scores_gemma":[0.0007647199,0.000075023,0.000105838,0.0002384723,0.00008470404,0.00009112788,0.00006139224,0.0005203761,0.0000171597],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004712369,"about_ca_system_score_gemma":0.0001090593,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000965238,"about_ca_topic_score_gemma":0.0001165449,"domain_scores_codex":[0.9984301,0.000273074,0.0006436044,0.0001420251,0.0003137003,0.0001975142],"domain_scores_gemma":[0.9982533,0.0003232296,0.0008535563,0.0002224742,0.0002676608,0.0000797113],"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":[0.000262612,0.0002882934,0.03839678,0.000199715,0.0003138486,0.00006979008,0.2239478,0.00001032039,0.70004,0.00003795432,0.002097518,0.03433534],"study_design_scores_gemma":[0.002012505,0.001179125,0.889504,0.00006557254,0.00007515745,0.00009993882,0.02828353,0.00009385219,0.07767983,0.000240199,0.0006181986,0.0001480691],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9924572,0.001634496,0.00002828643,0.0004636511,0.0001613833,0.0002161615,0.00004215656,0.000006286698,0.004990363],"genre_scores_gemma":[0.9982986,0.00000294383,0.0007196362,0.0002564007,0.000219672,0.000003213244,0.000005577093,0.00001349883,0.0004803994],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8511072,"threshold_uncertainty_score":0.9954776,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05239156274944249,"score_gpt":0.3310371246838275,"score_spread":0.2786455619343849,"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."}}