{"id":"W3116202855","doi":"10.1101/2020.12.23.424091","title":"Narratives: fMRI data for evaluating models of naturalistic language comprehension","year":2020,"lang":"en","type":"preprint","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Neurobiology of Language and Bilingualism","field":"Neuroscience","cited_by":35,"is_retracted":false,"has_abstract":true,"ca_institutions":"BC Children's Hospital; University of British Columbia; McGill University; Montreal Neurological Institute and Hospital","funders":"Advanced Research Projects Agency; Defense Advanced Research Projects Agency; National Institutes of Health; Intel Corporation","keywords":"Narrative; Computer science; Comprehension; Variety (cybernetics); Benchmark (surveying); Metadata; Data collection; Neuroimaging; Naturalism; Psychology; Natural language processing; Artificial intelligence; Linguistics; World Wide Web; Neuroscience","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":[0.00349298,0.0007548263,0.0002302789,0.001081345,0.0003607871,0.0007420756,0.0007821509,0.000775042,0.002842061],"category_scores_gemma":[0.02006573,0.0002287207,0.0004941833,0.0006620513,0.0008642022,0.00102077,0.0008930461,0.0006528305,0.000932831],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005062244,"about_ca_system_score_gemma":0.000606653,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002606312,"about_ca_topic_score_gemma":0.004013109,"domain_scores_codex":[0.9989405,0.0006185824,0.00008037604,0.0002074993,0.0001211733,0.0000317576],"domain_scores_gemma":[0.9920271,0.00535667,0.0007061756,0.001057478,0.0005490845,0.000303575],"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":[0.008245326,0.002410035,0.2503165,0.005010016,0.00212179,0.004008459,0.01243461,0.2565582,0.1354249,0.0358428,0.08220576,0.2054216],"study_design_scores_gemma":[0.0006760005,0.002133539,0.1941099,0.0006454571,0.0005547101,0.004858878,0.003888647,0.5769539,0.06265123,0.09779397,0.05533806,0.0003957225],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8852528,0.001058781,0.07092785,0.0008481807,0.00008621783,0.0003570819,0.03470322,0.001908095,0.004857823],"genre_scores_gemma":[0.9033066,0.0003409475,0.05492456,0.0001587481,0.00006729418,0.0006436374,0.03942435,0.0003731153,0.000760751],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00349298,"threshold_uncertainty_score":0.01847285,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1272530129184253,"score_gpt":0.3394178766284918,"score_spread":0.2121648637100665,"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."}}