{"id":"W4411724140","doi":"10.2196/69485","title":"Performance of Large Language Models in the Non-English Context: Qualitative Study of Models Trained on Different Languages in Chinese Medical Examinations","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Context (archaeology); Natural language processing; Language model; Artificial intelligence; English language; Word error rate; Benchmark (surveying); Linguistics; Psychology; Mathematics education","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":[0.02832357,0.0007141845,0.000537261,0.001232191,0.001476039,0.003163529,0.002158685,0.001131006,0.001351195],"category_scores_gemma":[0.1063101,0.0004058606,0.0007317042,0.0008935304,0.002820429,0.005200924,0.003058234,0.001537753,0.0005447966],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004123603,"about_ca_system_score_gemma":0.003926151,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02021644,"about_ca_topic_score_gemma":0.01521313,"domain_scores_codex":[0.9846255,0.01051306,0.0005529074,0.001449093,0.002113502,0.0007459157],"domain_scores_gemma":[0.9358944,0.05084076,0.002214725,0.003089117,0.006337652,0.00162329],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.000381046,0.0004188066,0.4471572,0.0008498334,0.0002040636,0.001320817,0.3848746,0.01671299,0.00671407,0.003305544,0.004396364,0.1336646],"study_design_scores_gemma":[0.00006072643,0.001311316,0.2601255,0.00146758,0.0006742299,0.001893501,0.360593,0.3119268,0.01954919,0.006413663,0.03550047,0.0004840822],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9887263,0.0003676419,0.005449712,0.001210781,0.0000210121,0.0001186221,0.0001306178,0.0001536096,0.003821811],"genre_scores_gemma":[0.9959379,0.0001495147,0.00282618,0.000142236,0.000006393833,0.00005531278,0.0001708405,0.00004357856,0.0006681234],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02832357,"threshold_uncertainty_score":0.1497912,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06632131130418917,"score_gpt":0.4693720188904845,"score_spread":0.4030507075862953,"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."}}