{"id":"W7133205121","doi":"10.1109/ocit66168.2025.11400016","title":"Transforming Medical Machine Translation with Next-Generation Large Language Models","year":2025,"lang":"","type":"article","venue":"","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Machine translation; Leverage (statistics); Health care; Domain (mathematical analysis); Intersection (aeronautics); Translation (biology); Machine translation software usability; Computer-assisted translation","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001584611,0.0003937479,0.0004167019,0.0003456126,0.0005764868,0.000387136,0.000881927,0.0003762858,0.0008340687],"category_scores_gemma":[0.00006558107,0.0003273804,0.0001183945,0.001174429,0.00006610827,0.001530299,0.00008794631,0.001013927,0.00002041413],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001110942,"about_ca_system_score_gemma":0.0009879244,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001499722,"about_ca_topic_score_gemma":0.007183665,"domain_scores_codex":[0.9958627,0.0004865522,0.0007775642,0.0009123051,0.001225394,0.0007354835],"domain_scores_gemma":[0.9985024,0.0001968194,0.0001172327,0.0007117582,0.0001591444,0.0003125997],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00005161409,0.0001596418,0.0005888017,0.0003783294,0.00005473315,0.00005411201,0.01038121,0.01124532,0.0002853977,0.1574498,0.00007845412,0.8192726],"study_design_scores_gemma":[0.001494139,0.0001950665,0.00006649734,0.0003208951,0.00003758722,0.00002776919,0.0002020623,0.9944253,0.0005137182,0.0003686656,0.00201952,0.0003287596],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.009484534,0.005568647,0.9381672,0.02890901,0.0005933213,0.0007724637,0.00001180891,0.0002937383,0.01619927],"genre_scores_gemma":[0.9673582,0.0001706649,0.0272668,0.003161307,0.0002184824,0.00003641857,0.00007016898,0.00002647541,0.001691448],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.98318,"threshold_uncertainty_score":0.9999178,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04114825179416875,"score_gpt":0.3178540664741402,"score_spread":0.2767058146799715,"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."}}