{"id":"W4400889759","doi":"10.1111/desc.13551","title":"Limited evidence of test‐retest reliability in infant‐directed speech preference in a large preregistered infant experiment","year":2024,"lang":"en","type":"article","venue":"Developmental Science","topic":"Language Development and Disorders","field":"Psychology","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba; University of British Columbia","funders":"Leibniz-Gemeinschaft; Norges Forskningsråd; Social Sciences and Humanities Research Council of Canada; National Natural Science Foundation of China; National Science Foundation","keywords":"Psychology; Reliability (semiconductor); Preference; Test (biology); Developmental psychology; Active listening; Context (archaeology); Speech perception; Audiology; Perception; Statistics; Communication; Medicine","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":[0.001861308,0.000265202,0.0003182275,0.0006776941,0.0000766627,0.00009662112,0.0007563749,0.0001142006,0.001184998],"category_scores_gemma":[0.001260988,0.0002337451,0.00004255019,0.003888532,0.0004394444,0.0006641973,0.0003950463,0.0002747095,0.0001305407],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005327415,"about_ca_system_score_gemma":0.0008891367,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001243221,"about_ca_topic_score_gemma":0.001447775,"domain_scores_codex":[0.9966766,0.00009726152,0.0007867279,0.001011348,0.0006618764,0.0007661803],"domain_scores_gemma":[0.9988237,0.0004438784,0.0001053785,0.0004042306,0.00009152431,0.0001312896],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001397971,0.0005688295,0.9029529,0.0001385956,0.00001092156,0.0001880745,0.04659978,0.000002409427,0.03717243,0.0003897918,0.000294946,0.0115415],"study_design_scores_gemma":[0.0007166966,0.0001045845,0.9748422,0.001031419,0.000003542825,0.00002488784,0.003232638,0.0004095883,0.01846011,0.0001093861,0.0006595401,0.0004053995],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9847082,0.001293507,0.0000232748,0.0001357089,0.0004702449,0.0007343753,0.00001269658,0.0001513505,0.01247068],"genre_scores_gemma":[0.994741,0.00004073885,0.004268812,0.0001055241,0.000009343012,0.0001795773,0.00001064285,0.00001499229,0.0006293877],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07188927,"threshold_uncertainty_score":0.9997281,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04609901152125097,"score_gpt":0.3373769925537226,"score_spread":0.2912779810324717,"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."}}