{"id":"W3200891190","doi":"","title":"Inspecting the Factuality of Hallucinated Entities in Abstractive Summarization.","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Hallucinating; Automatic summarization; Computer science; Natural language processing; Artificial intelligence; Information retrieval","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":[],"consensus_categories":[],"category_scores_codex":[0.0003477189,0.0001590198,0.0002470964,0.0001654712,0.00008489141,0.00009754935,0.001172991,0.0001575073,0.00002136354],"category_scores_gemma":[0.0001095462,0.0001604257,0.000107055,0.0004926445,0.0000746207,0.000386239,0.001553574,0.0005469146,0.000002578835],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001877642,"about_ca_system_score_gemma":0.00024508,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002176621,"about_ca_topic_score_gemma":0.0007532628,"domain_scores_codex":[0.9986393,0.0001976076,0.000249414,0.0006333538,0.00009672099,0.0001835396],"domain_scores_gemma":[0.9983748,0.0002009929,0.0002953982,0.0008882394,0.0002059111,0.00003462691],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000008591863,0.0001130602,0.02212457,0.0001221647,0.00008884644,0.0001514539,0.006672441,0.7607765,0.0001019778,0.2089219,0.000006245017,0.0009123262],"study_design_scores_gemma":[0.0003794988,0.00001373176,0.0538019,0.0002294703,0.00003253877,0.000002396949,0.003771675,0.8994634,0.001700123,0.04021598,0.00002145394,0.0003678515],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5949146,0.00003752427,0.4028111,0.00006144633,0.0002234628,0.0001136989,0.000003313991,0.00004522433,0.00178968],"genre_scores_gemma":[0.998632,0.00006160723,0.0009858154,0.00002538773,0.00001861904,4.782057e-7,0.000007950787,0.000006169204,0.0002619411],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4037174,"threshold_uncertainty_score":0.6541972,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09452848709654141,"score_gpt":0.1948844294690246,"score_spread":0.1003559423724832,"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."}}