{"id":"W4414016026","doi":"10.1162/opmi.a.21","title":"Issues of Generalization From Unreliable or Unrepresentative Stimuli: Broad Lessons From Lexical Ambiguity","year":2025,"lang":"en","type":"article","venue":"Open Mind","topic":"Categorization, perception, and language","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Scarborough Hospital; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; University of Toronto","keywords":"Ambiguity; Generalization; Computer science; Epistemology","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0001468722,0.0001406547,0.0002910374,0.00007647333,0.0001059735,0.00009748009,0.0003963876,0.0001543685,0.08900473],"category_scores_gemma":[0.00009133724,0.0001247854,0.00004951654,0.0003576948,0.0001024749,0.0001807487,0.0001408309,0.00009175134,0.0004657728],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003860327,"about_ca_system_score_gemma":0.0001065031,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.07296546,"about_ca_topic_score_gemma":0.006624367,"domain_scores_codex":[0.9986403,0.0002357644,0.0003460909,0.0004716958,0.0001388126,0.0001673287],"domain_scores_gemma":[0.9990848,0.0001611655,0.0001200698,0.0004622883,0.0001191598,0.00005257256],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.003694116,0.003181432,0.2045543,0.00003634391,0.001623624,0.0000837734,0.1858354,0.001436666,0.05708927,0.01501033,0.2699798,0.257475],"study_design_scores_gemma":[0.005539523,0.000210062,0.7636562,0.0001739762,0.0004280795,0.000001885957,0.02391211,0.001266995,0.03451041,0.008132672,0.1615011,0.0006669686],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9363449,0.0004179086,0.01308921,0.0007259524,0.0008223415,0.0004782226,0.0007101113,0.000008513981,0.0474029],"genre_scores_gemma":[0.8840935,0.00006700777,0.004411764,0.000185932,0.000218276,0.00003172004,0.001842108,0.00001805135,0.1091317],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5591019,"threshold_uncertainty_score":0.9332078,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.110349912841709,"score_gpt":0.4732420073643399,"score_spread":0.3628920945226309,"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."}}