{"id":"W4407179463","doi":"10.1017/s0272263125000075","title":"The impact of collocational proficiency features on expert ratings of L2 English learners’ writing","year":2025,"lang":"en","type":"article","venue":"Studies in Second Language Acquisition","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Linguistics; Psychology; Language proficiency; Language assessment; Mathematics education; Philosophy","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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":[0.0007751847,0.0001755827,0.0003148339,0.0002585581,0.0002035429,0.0000206347,0.0002153987,0.0001005798,0.004851248],"category_scores_gemma":[0.0007572438,0.0001263983,0.0001538197,0.000518842,0.0002840863,0.00008282594,0.00006653131,0.000264696,0.000006675938],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001203718,"about_ca_system_score_gemma":0.00006445756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001687007,"about_ca_topic_score_gemma":0.00004353142,"domain_scores_codex":[0.9984295,0.0003006516,0.000513491,0.000288175,0.0001863896,0.0002817926],"domain_scores_gemma":[0.9978642,0.001237561,0.0002798038,0.0003433649,0.0002517069,0.00002330445],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.001121396,0.0007044277,0.005796156,0.0003158098,0.001148403,0.00008584275,0.8493575,0.0004891964,0.02324073,0.04352437,0.03961173,0.03460443],"study_design_scores_gemma":[0.001898131,0.0005040485,0.1243461,0.0005170039,0.00002268479,0.00001080912,0.8651624,0.00003735823,0.006723386,0.0003032345,0.0002326525,0.0002421621],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9031239,0.04210052,0.0000150321,0.00009191207,0.0004898823,0.0002986521,0.00002262194,0.00003527771,0.05382217],"genre_scores_gemma":[0.9961123,0.00004745394,0.00005342333,0.001127468,0.000180983,0.00008667919,0.00003211884,0.00001290502,0.002346661],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.11855,"threshold_uncertainty_score":0.9960585,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01855927005041019,"score_gpt":0.402168913415247,"score_spread":0.3836096433648368,"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."}}