{"id":"W4401042030","doi":"10.18653/v1/2024.semeval-1.239","title":"CLaC at SemEval-2024 Task 2: Faithful Clinical Trial Inference","year":2024,"lang":"en","type":"article","venue":"","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"SemEval; Inference; Computer science; Task (project management); Natural language processing; Artificial intelligence","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":[],"consensus_categories":[],"category_scores_codex":[0.0004846085,0.0001492747,0.0001854831,0.00003273322,0.00005313672,0.0000526381,0.0002109402,0.0003985356,0.0005271909],"category_scores_gemma":[0.0006015153,0.0001058977,0.0001937968,0.0001017751,0.0002485772,0.000001761666,0.0002835579,0.0001996585,0.0004039276],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001325818,"about_ca_system_score_gemma":0.0001393193,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000010262,"about_ca_topic_score_gemma":0.00004699384,"domain_scores_codex":[0.9986215,0.00009300478,0.0003590016,0.0005119504,0.0001563615,0.0002582363],"domain_scores_gemma":[0.9993525,0.0001360067,0.00002989731,0.0003193335,0.00003095825,0.0001313319],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.005078793,0.00020272,0.002106204,0.00008103299,0.0003153695,0.00006850591,0.00007142807,0.000001922713,0.03191198,0.001028055,0.6586918,0.3004422],"study_design_scores_gemma":[0.003807256,0.001233277,0.0004032308,0.00002496284,0.00003125074,0.00001266806,0.00003899088,0.0002139643,0.004122974,0.0002524925,0.9896522,0.0002067513],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.953046,0.004038756,0.006440511,0.002138665,0.005118311,0.0002926487,0.000038646,0.0002316328,0.02865486],"genre_scores_gemma":[0.9437041,0.0005213017,0.001403205,0.0006922799,0.001436734,0.00002454364,0.00007342155,0.00001922916,0.05212517],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3309604,"threshold_uncertainty_score":0.5772371,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06425536072042741,"score_gpt":0.4095967690482611,"score_spread":0.3453414083278337,"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."}}