{"id":"W2230811337","doi":"10.5430/elr.v4n4p58","title":"Comparison of Inter-rater Reliability of Human and Computer Prosodic Annotation Using Brazil’s Prosody Model","year":2015,"lang":"en","type":"article","venue":"English Linguistics Research","topic":"Phonetics and Phonology Research","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Kappa; Prosody; Reliability (semiconductor); Correlation; Syllable; Population; Inter-rater reliability; Tone (literature); Annotation; Pearson product-moment correlation coefficient; Statistics; Cohen's kappa; Computer science; Speech recognition; Mathematics; Natural language processing; Artificial intelligence; Linguistics; Rating scale; Medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04843904,0.0007615587,0.0007051942,0.001934071,0.0007390611,0.001551376,0.0008352972,0.0005974541,0.0009084932],"category_scores_gemma":[0.143198,0.0004999881,0.001033527,0.001018604,0.001629717,0.001347779,0.00247048,0.0006292978,0.0006222887],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006846886,"about_ca_system_score_gemma":0.0006914796,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003012959,"about_ca_topic_score_gemma":0.00514788,"domain_scores_codex":[0.9476781,0.03162955,0.004035278,0.007410824,0.008385543,0.0008606964],"domain_scores_gemma":[0.8510253,0.09551711,0.008933377,0.01497709,0.02869629,0.0008507561],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003910075,0.0004702497,0.5885324,0.001104235,0.003182111,0.0004703369,0.06974463,0.01071076,0.04210446,0.005242711,0.004028209,0.2704999],"study_design_scores_gemma":[0.0003104363,0.002175445,0.8241158,0.0005785152,0.0009873457,0.00177531,0.01580194,0.1035181,0.03003739,0.007520702,0.01260512,0.0005739639],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8871545,0.0006451418,0.1014348,0.000141047,0.0002007724,0.0007735562,0.0004754929,0.0003747994,0.008800007],"genre_scores_gemma":[0.9819065,0.0001203648,0.01635796,0.00004892348,0.00002507664,0.0004723671,0.0003301702,0.0001072056,0.0006314177],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04843904,"threshold_uncertainty_score":0.2561732,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2703584596292362,"score_gpt":0.5320530570359621,"score_spread":0.2616945974067259,"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."}}