{"id":"W4366998711","doi":"10.5539/ells.v13n2p46","title":"An Empirical Study on the Validity of the AES Systems Juku and iWrite for Continuation Writing Task Assessment","year":2023,"lang":"en","type":"article","venue":"English Language and Literature Studies","topic":"Technology-Enhanced Education Studies","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Task (project management); Continuation; Scoring system; Narrative; Grammar; Psychology; Computer science; Mathematics education; Linguistics; 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":[0.0409973,0.00039015,0.000502222,0.002262332,0.0008145898,0.001789882,0.0007887137,0.0005427134,0.000978531],"category_scores_gemma":[0.1493911,0.0003795575,0.0005672884,0.001713447,0.001575442,0.002169263,0.001758909,0.0008490955,0.0003986675],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009425064,"about_ca_system_score_gemma":0.001511713,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001413657,"about_ca_topic_score_gemma":0.002222656,"domain_scores_codex":[0.9590757,0.02078717,0.005168469,0.002608654,0.01147283,0.0008870638],"domain_scores_gemma":[0.7991347,0.1255018,0.02008408,0.01047706,0.04264269,0.002159735],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004814316,0.0005237711,0.8482197,0.0004202342,0.0001266173,0.0001358243,0.02570327,0.0004395135,0.005129576,0.0011442,0.0005803489,0.1170955],"study_design_scores_gemma":[0.00007797502,0.002304301,0.9475232,0.0004567553,0.0001447012,0.0005187393,0.02731458,0.007852711,0.008493907,0.0006125804,0.004607002,0.00009352185],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9938029,0.000133911,0.002731755,0.00006479515,0.00001872654,0.0001320784,0.00004333056,0.00001792156,0.003054625],"genre_scores_gemma":[0.9940148,0.0001230888,0.005041467,0.00002966276,0.00001013016,0.0002204689,0.0000984779,0.00001617354,0.0004457459],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0409973,"threshold_uncertainty_score":0.216817,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0533578777450666,"score_gpt":0.4335936760121358,"score_spread":0.3802357982670692,"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."}}