{"id":"W6931679768","doi":"10.5683/sp3/toxzsq","title":"A Theoretical Framework for Evaluating Narrative Surprise in Large Language Models (WNU 2025)","year":2025,"lang":"en","type":"dataset","venue":"Borealis","topic":"Urban Green Space and Health","field":"Environmental Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Surprise; Narrative; Language understanding; Natural language; On Language; Narrative network","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":[0.0116972,0.001717139,0.0009326001,0.002880456,0.001510403,0.003299771,0.003430476,0.002583595,0.01331258],"category_scores_gemma":[0.05996295,0.000558915,0.001682103,0.003288948,0.001083549,0.003273538,0.004593547,0.003303207,0.01067703],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002984342,"about_ca_system_score_gemma":0.002955339,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01891766,"about_ca_topic_score_gemma":0.03125506,"domain_scores_codex":[0.9904128,0.005518968,0.0008291286,0.001286892,0.001550083,0.0004021141],"domain_scores_gemma":[0.9850519,0.009312435,0.0006168486,0.002724711,0.001838269,0.0004557835],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006128929,0.000383551,0.008202164,0.001329664,0.0002847603,0.0001802873,0.0004614531,0.01794848,0.0007978109,0.02673396,0.8766482,0.06641676],"study_design_scores_gemma":[0.00105319,0.0004571431,0.0156342,0.001023404,0.0002821982,0.0006639193,0.001591617,0.2936785,0.003814708,0.1637612,0.517792,0.0002479881],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.02062566,0.001924525,0.08536933,0.00668713,0.0006445893,0.0009124115,0.8564472,0.01231334,0.01507589],"genre_scores_gemma":[0.05303427,0.0002307176,0.0700114,0.001122811,0.0001116016,0.002061538,0.8709059,0.000474796,0.002047007],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.01891766,"threshold_uncertainty_score":0.06186146,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02592288425587333,"score_gpt":0.3543021331889294,"score_spread":0.3283792489330561,"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."}}