{"id":"W1991701363","doi":"10.1111/1540-4781.00137","title":"Decision Making while Rating ESL/EFL Writing Tasks: A Descriptive Framework","year":2002,"lang":"en","type":"article","venue":"Modern Language Journal","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":236,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Christian Studies","funders":"","keywords":"Test of English as a Foreign Language; Active listening; Psychology; Think aloud protocol; Protocol analysis; Test (biology); Descriptive statistics; Mathematics education; Writing assessment; Reading (process); Exploratory research; Linguistics; Computer science; Language assessment","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008169149,0.0002190944,0.0002541502,0.0001795232,0.001704774,0.001256424,0.0002576452,0.00009022548,0.00755089],"category_scores_gemma":[0.0005683097,0.000186331,0.0001633989,0.00004866901,0.00006987704,0.0005002432,0.00007180097,0.001623922,0.0001610592],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001072851,"about_ca_system_score_gemma":0.00001475625,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003698337,"about_ca_topic_score_gemma":0.00006816648,"domain_scores_codex":[0.9981907,0.0002061164,0.0004459251,0.0002430239,0.00043065,0.0004835953],"domain_scores_gemma":[0.9989086,0.0003885333,0.0002883719,0.0002011575,0.0001015614,0.0001118033],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001888146,0.00006655608,0.0003746259,0.00001745513,0.00005171221,0.0003771864,0.5709116,0.0007187782,0.0002977766,0.003249785,0.002463863,0.4214517],"study_design_scores_gemma":[0.002641585,0.0006494786,0.0005823235,0.01109833,0.0002834469,0.002416484,0.5481986,0.3145657,0.0001786919,0.03706276,0.07981063,0.002511885],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8286656,0.007188904,0.1125559,0.0002873986,0.001028444,0.0001177783,0.000009489173,0.0002454316,0.049901],"genre_scores_gemma":[0.9788742,0.00002282545,0.01464797,0.0005702988,0.003341511,0.000003943288,0.00000183701,0.00005612956,0.002481301],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4189398,"threshold_uncertainty_score":0.9997804,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04833117962473061,"score_gpt":0.2782858427122726,"score_spread":0.229954663087542,"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."}}