{"id":"W6967235911","doi":"10.48448/s7em-cd36","title":"From chocolate bunny to chocolate crocodile: Do Language Models Understand Noun Compounds?","year":2023,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Religion, Ecology, and Ethics","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Conceptualization; Noun; Task (project management); Relation (database); Noun phrase; Language model","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":[0.002820081,0.0006103782,0.000500589,0.0005991926,0.0004802865,0.004371517,0.001038244,0.001207832,0.005980069],"category_scores_gemma":[0.01524399,0.0004220343,0.0006525329,0.0006555885,0.001464371,0.008891507,0.001279327,0.002539302,0.002010972],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001265539,"about_ca_system_score_gemma":0.001434313,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01158582,"about_ca_topic_score_gemma":0.01304287,"domain_scores_codex":[0.9987409,0.0007943557,0.00003769066,0.0002279074,0.0001474242,0.00005175284],"domain_scores_gemma":[0.9909647,0.006369591,0.0006519913,0.001045958,0.0007710455,0.0001968004],"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":[0.001006339,0.0005248985,0.09106687,0.001152694,0.0006446908,0.001299019,0.02560229,0.100549,0.02393123,0.3590431,0.03537872,0.3598011],"study_design_scores_gemma":[0.00006945833,0.00006726427,0.009186745,0.0001886604,0.00008383029,0.0003672977,0.006617681,0.6358714,0.005269467,0.3165313,0.02565169,0.0000952092],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5372865,0.001404083,0.3798696,0.01727713,0.0002740957,0.0001499062,0.00304514,0.002030052,0.05866352],"genre_scores_gemma":[0.9234455,0.0005633052,0.06867779,0.001192255,0.00006851418,0.00005107391,0.001824504,0.0004937593,0.003683337],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01158582,"threshold_uncertainty_score":0.02303678,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04879709904182681,"score_gpt":0.3658460927115761,"score_spread":0.3170489936697493,"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."}}