{"id":"W2039139328","doi":"10.1016/j.jslw.2014.04.002","title":"Teacher assessment of grammatical ability in second language academic writing: A case study","year":2014,"lang":"en","type":"article","venue":"Journal of Second Language Writing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":29,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University","funders":"","keywords":"Second language writing; Linguistics; Computer science; Academic writing; Writing assessment; Mathematics education; Psychology; Second language; Philosophy","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005807946,0.0001981241,0.0005780934,0.0002760092,0.0001800146,0.0001287419,0.0002324127,0.00009257355,0.004934038],"category_scores_gemma":[0.0006211621,0.0001673642,0.0001611563,0.00006122014,0.000106534,0.0003363084,0.0000681388,0.001995,0.000003678129],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008414654,"about_ca_system_score_gemma":0.00006432403,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004540607,"about_ca_topic_score_gemma":0.002633749,"domain_scores_codex":[0.9970551,0.0007480989,0.001309706,0.0002020677,0.0003536977,0.0003313708],"domain_scores_gemma":[0.9980176,0.000800052,0.0007364054,0.0002156768,0.0001140028,0.0001162715],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00003785164,0.001019319,0.1175474,0.0007404549,0.0001744443,0.004160557,0.8272492,0.00006940684,0.008412013,0.003400156,0.0002161541,0.03697297],"study_design_scores_gemma":[0.001854829,0.0005313294,0.01266729,0.0003996334,0.00007093257,0.00127483,0.9803689,0.001409463,0.0001312058,0.0000862308,0.0009272061,0.0002781396],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9842066,0.0004310209,0.00004579416,0.00007071035,0.0001207428,0.0001609313,0.000007139945,0.00002662256,0.0149305],"genre_scores_gemma":[0.9975423,9.239115e-7,0.0004745282,0.0001373568,0.001118504,0.000004053225,0.000002280176,0.00003058828,0.0006894713],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1531197,"threshold_uncertainty_score":0.9959756,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02830187065077013,"score_gpt":0.3370044620007906,"score_spread":0.3087025913500205,"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."}}