{"id":"W6982658776","doi":"","title":"Issues of Generalization from Unreliable or Unrepresentative Psycholinguistic Stimuli: A Case Study on Lexical Ambiguity","year":2024,"lang":"en","type":"article","venue":"eScholarship (California Digital Library)","topic":"Urban and spatial planning","field":"Environmental Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Toronto Scarborough; Natural Sciences and Engineering Research Council of Canada","keywords":"Ambiguity; Psycholinguistics; Generalizability theory; Generalization; Set (abstract data type); Unobservable; Word (group theory); Empirical research","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":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0001470584,0.0002722615,0.0002744574,0.00008934174,0.0001187292,0.0007449578,0.0002877331,0.00009797929,0.00397799],"category_scores_gemma":[0.0003130161,0.0002119334,0.00009400627,0.0006013287,0.0001378582,0.001340467,0.000240017,0.0003272778,0.002161327],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005275678,"about_ca_system_score_gemma":0.00002944208,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009272179,"about_ca_topic_score_gemma":0.0000459851,"domain_scores_codex":[0.9979724,0.0001442222,0.0004522729,0.0006416781,0.0004951201,0.0002942742],"domain_scores_gemma":[0.9990396,0.0003051639,0.00007852896,0.0003534889,0.000008421959,0.0002147791],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008058646,0.001886196,0.9470671,0.00005943273,0.0001970618,0.01012753,0.002357012,0.002359208,0.000502132,0.0004595168,0.02605639,0.008122502],"study_design_scores_gemma":[0.01052591,0.01134448,0.381449,0.002581825,0.0009411275,0.0009717607,0.009957715,0.09114593,0.02508593,0.09017047,0.3686736,0.007152215],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9847345,0.0001139933,0.00065957,0.0001465531,0.0002792109,0.0004100211,0.002044994,0.0002513824,0.01135981],"genre_scores_gemma":[0.9961074,0.00000478144,0.0005624075,0.0002105113,0.0001726943,0.00001638508,0.0002560332,0.00005535507,0.002614454],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5656181,"threshold_uncertainty_score":0.9986156,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04567344384952107,"score_gpt":0.3083705470257598,"score_spread":0.2626971031762387,"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."}}