{"id":"W2239662196","doi":"10.5539/elt.v9n2p102","title":"How Do Raters Judge Spoken Vocabulary?","year":2016,"lang":"en","type":"article","venue":"English Language Teaching","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Psychology; Vocabulary; Consistency (knowledge bases); Variation (astronomy); Linguistics; Rasch model; Rating scale; Natural language processing; Cognitive psychology; Artificial intelligence; Developmental psychology; Computer science","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":[0.0374997,0.0005619207,0.0007688188,0.001777273,0.000513392,0.002208573,0.0006322378,0.0009443423,0.001057953],"category_scores_gemma":[0.186664,0.0003272029,0.0006169627,0.0009353265,0.001002819,0.002088906,0.0009367802,0.0007220027,0.001666007],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004127542,"about_ca_system_score_gemma":0.000340629,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001679566,"about_ca_topic_score_gemma":0.002438713,"domain_scores_codex":[0.9651423,0.02182763,0.003862713,0.00260049,0.005899199,0.0006677971],"domain_scores_gemma":[0.8811514,0.06713166,0.01490013,0.006370541,0.02828646,0.002159811],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00213631,0.0002623007,0.6106347,0.0007182505,0.001031108,0.0005556619,0.04297036,0.001704309,0.02849867,0.002017301,0.007347709,0.3021232],"study_design_scores_gemma":[0.0003256206,0.002670479,0.8469407,0.0008183672,0.0006119848,0.002707838,0.05460494,0.03074882,0.02992362,0.00830465,0.02155329,0.0007896842],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9578246,0.001645194,0.02573323,0.0008358393,0.00036378,0.0003002879,0.0003907598,0.0002454447,0.01266089],"genre_scores_gemma":[0.985638,0.0005341708,0.011314,0.0002360281,0.0001156297,0.0002587031,0.0003270413,0.00008553999,0.001490943],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0374997,"threshold_uncertainty_score":0.1983197,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008731158519110651,"score_gpt":0.2708562456947734,"score_spread":0.2621250871756627,"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."}}