{"id":"W2325923846","doi":"10.5539/elt.v9n5p98","title":"Comparison of Native and Non-native English Language Teachers’ Evaluation of EFL Learners’ Speaking Skills: Conflicting or Identical Rating Behaviour?","year":2016,"lang":"en","type":"article","venue":"English Language Teaching","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Pronunciation; Fluency; Psychology; First language; Language proficiency; Mathematics education; Vocabulary; Pearson product-moment correlation coefficient; Test (biology); Rating scale; Significant difference; Linguistics; Developmental psychology; Statistics; Mathematics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006327671,0.0003085495,0.000626428,0.0002698724,0.0005405921,0.000131036,0.0002442826,0.000131047,0.000661188],"category_scores_gemma":[0.01592745,0.0002225968,0.0001255019,0.00006664312,0.000318184,0.0006587442,0.0001529877,0.001133818,0.000003094175],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001487652,"about_ca_system_score_gemma":0.00006954822,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001205384,"about_ca_topic_score_gemma":0.0007533842,"domain_scores_codex":[0.9959583,0.001556283,0.00083755,0.0004391778,0.0008008035,0.0004079353],"domain_scores_gemma":[0.9964271,0.00198018,0.0008252199,0.000281536,0.0003935315,0.00009244023],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000243845,0.0001308329,0.02913148,0.00006012117,0.0001018756,0.000006642566,0.925996,0.00003986742,0.002935245,0.001732768,0.00003818942,0.03980258],"study_design_scores_gemma":[0.001717712,0.0001921808,0.001454088,0.001105576,0.0002420815,0.000001490768,0.9899716,0.001281002,0.003316659,0.00001725027,0.0003614884,0.0003388846],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9658999,0.0004353285,0.0004346865,0.00001973865,0.0004793251,0.0003648285,0.00002810626,0.0001745247,0.03216361],"genre_scores_gemma":[0.9968219,0.000002143884,0.0006513355,0.00002239585,0.001204987,0.00001679561,0.00002661445,0.00006000285,0.001193801],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06397557,"threshold_uncertainty_score":0.9923618,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04161681684629527,"score_gpt":0.364497786540308,"score_spread":0.3228809696940127,"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."}}