{"id":"W2889522036","doi":"10.5539/ies.v11n9p12","title":"Interrater Scoring of Public Speaking Performances in English Language Teacher Education Program","year":2018,"lang":"en","type":"article","venue":"International Education Studies","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Grading (engineering); Psychology; Inter-rater reliability; Mathematics education; Curriculum; Rating scale; Qualitative research; Qualitative property; Empowerment; Pedagogy; Teacher education; 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":[0.02093384,0.000471583,0.0006228063,0.002306049,0.001056489,0.0007834503,0.0007445411,0.0003836149,0.001298902],"category_scores_gemma":[0.03499464,0.0002599053,0.0005817857,0.0009795635,0.001083277,0.0005155978,0.002267616,0.0004836014,0.0007063716],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000478773,"about_ca_system_score_gemma":0.0009420495,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001055889,"about_ca_topic_score_gemma":0.002529584,"domain_scores_codex":[0.9765413,0.01071057,0.003463607,0.001882654,0.006753326,0.0006484881],"domain_scores_gemma":[0.9616939,0.01195921,0.002900279,0.00418554,0.01839451,0.0008664828],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001479306,0.0005495353,0.625302,0.001102461,0.0003948571,0.0007339625,0.07690802,0.001046164,0.03695838,0.001043616,0.002331216,0.2521505],"study_design_scores_gemma":[0.00005079648,0.001437759,0.906138,0.0002860134,0.0002422288,0.001420038,0.04308059,0.006374667,0.03238069,0.001370554,0.00707315,0.000145505],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9812029,0.0002396525,0.01187831,0.00006031666,0.00008022247,0.0004611701,0.0002401789,0.0001209219,0.005716195],"genre_scores_gemma":[0.9871467,0.0001721186,0.01049661,0.00001478188,0.00001849292,0.0003414558,0.0002363538,0.00003726082,0.001536214],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02093384,"threshold_uncertainty_score":0.11071,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08813912167534482,"score_gpt":0.4762929714795188,"score_spread":0.3881538498041739,"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."}}