{"id":"W4388495886","doi":"10.5430/wjel.v14n1p121","title":"An Investigation of Dynamic Assessment on EFL Learners’ Speaking Performance","year":2023,"lang":"en","type":"article","venue":"World Journal of English Language","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Fluency; Psychology; English as a foreign language; Mathematics education; Strengths and weaknesses; Qualitative research; Foreign language; Social psychology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00080878,0.0001075396,0.0001946678,0.0003934288,0.00004647516,0.00002027946,0.0002589188,0.00004804849,0.0009534442],"category_scores_gemma":[0.00006846301,0.00008458902,0.00007975885,0.0005855123,0.000052143,0.0001724406,0.00001454178,0.0003824166,0.00004430863],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005807489,"about_ca_system_score_gemma":0.00004353976,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001183013,"about_ca_topic_score_gemma":0.00001295849,"domain_scores_codex":[0.9987439,0.0001770265,0.000414943,0.0001501301,0.0003266509,0.0001873505],"domain_scores_gemma":[0.9989116,0.0001731042,0.0004042341,0.0002098741,0.0001960046,0.0001052221],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000473065,0.001429319,0.6164042,0.0001393926,0.0004383351,0.0002542909,0.1751304,0.001551474,0.01463141,0.01098053,0.01137335,0.1671942],"study_design_scores_gemma":[0.0006329382,0.0006121435,0.977209,0.0001435867,0.00002259327,0.000006996057,0.01927673,0.0001075023,0.000150766,0.0002508343,0.001466567,0.0001203226],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.96694,0.00007552707,0.00001178059,0.0002176218,0.002047204,0.0000679568,0.00000581153,0.00003137435,0.03060272],"genre_scores_gemma":[0.9968225,0.00001710928,0.000664938,0.0002611893,0.0005508814,0.000004343015,0.0000266365,0.00001462375,0.001637769],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3608049,"threshold_uncertainty_score":0.9999598,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02978198847730963,"score_gpt":0.3798158850834386,"score_spread":0.3500338966061289,"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."}}