{"id":"W4312069611","doi":"10.31234/osf.io/uwche","title":"Limited evidence of test-retest reliability in infant-directed speech preference in a large pre-registered infant experiment","year":2022,"lang":"en","type":"preprint","venue":"","topic":"Language Development and Disorders","field":"Psychology","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba; University of British Columbia","funders":"Leibniz-Gemeinschaft; Social Sciences and Humanities Research Council of Canada; Norges Forskningsråd; National Science Foundation","keywords":"Reliability (semiconductor); Psychology; Preference; Active listening; Test (biology); Context (archaeology); Audiology; Developmental psychology; Speech perception; Sample size determination; Perception; Medicine; Statistics; Communication","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001670399,0.0005776506,0.001019551,0.0007009088,0.00004194982,0.00004343835,0.001139959,0.0005348081,0.01238549],"category_scores_gemma":[0.002805231,0.0005523073,0.0001711064,0.0009871345,0.0001211133,0.0001437501,0.002075414,0.00132755,0.0000311543],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004154197,"about_ca_system_score_gemma":0.0005098216,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01801567,"about_ca_topic_score_gemma":0.008999072,"domain_scores_codex":[0.9946923,0.0006822667,0.001683999,0.001556576,0.0006258853,0.0007590072],"domain_scores_gemma":[0.9958609,0.001133109,0.0006271357,0.002106368,0.0001561398,0.0001163634],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009835792,0.002237939,0.9562024,0.0005698071,0.00006452333,0.0001270059,0.03624819,0.00007645163,0.0008218487,0.0001595499,0.001350153,0.001158472],"study_design_scores_gemma":[0.00250252,0.0002950361,0.9888671,0.001129559,0.00002605089,0.000004966392,0.003480617,0.000457497,0.001245106,0.0004063757,0.0007152343,0.000869929],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9725203,0.002880412,0.00003093102,0.0002503615,0.0007137286,0.002742198,0.00009360483,0.0002628195,0.02050566],"genre_scores_gemma":[0.9922988,0.0001825299,0.002418522,0.0001957374,0.0000211576,0.001634031,0.0002114248,0.00004762174,0.002990206],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03276758,"threshold_uncertainty_score":0.9996929,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06775932560258904,"score_gpt":0.3576466124999326,"score_spread":0.2898872868973436,"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."}}