{"id":"W4405168385","doi":"10.5753/stil.2024.31170","title":"LLM-SEMREL: Towards a Better Coreference Resolution for Portuguese","year":2024,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Blutip (Canada)","funders":"","keywords":"Coreference; Portuguese; Computer science; Resolution (logic); Natural language processing; Artificial intelligence; Linguistics; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004289424,0.001629432,0.00160198,0.005697046,0.001611305,0.004878347,0.002254729,0.001631885,0.007610143],"category_scores_gemma":[0.01011505,0.001349886,0.002401468,0.003668139,0.0008429234,0.007483839,0.003723641,0.001664277,0.005397541],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001404635,"about_ca_system_score_gemma":0.003144856,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004877496,"about_ca_topic_score_gemma":0.007696984,"domain_scores_codex":[0.9958295,0.001450269,0.0006039196,0.0009682111,0.0009642814,0.0001837731],"domain_scores_gemma":[0.9970341,0.001065277,0.0003169987,0.0009869552,0.0004695709,0.0001270558],"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":[0.001456924,0.0005736162,0.007105936,0.005809103,0.0004856718,0.00295049,0.004762737,0.0332795,0.02183411,0.1578001,0.1126049,0.6513368],"study_design_scores_gemma":[0.0002222517,0.0002248209,0.003808928,0.001170275,0.0003895193,0.001924919,0.001765886,0.1399376,0.02010776,0.07180935,0.7583661,0.0002727712],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02813471,0.004103838,0.8691781,0.002309356,0.0006142138,0.000632714,0.0417312,0.02959243,0.02370338],"genre_scores_gemma":[0.1839743,0.003854154,0.7153931,0.001197729,0.0002211937,0.0007806717,0.08269485,0.005037933,0.006846075],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007610143,"threshold_uncertainty_score":0.02545846,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02760809426622047,"score_gpt":0.3072224662490973,"score_spread":0.2796143719828768,"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."}}