{"id":"W3134614468","doi":"10.31234/osf.io/gvr6m","title":"Do We Need Neural Models to Explain Human Judgments of Acceptability?","year":2020,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Social Science Fund of China; Natural Sciences and Engineering Research Council of Canada; National Science Foundation","keywords":"Sentence; Computer science; Natural language processing; Language model; Word (group theory); Artificial intelligence; Simple (philosophy); Variance (accounting); Similarity (geometry); n-gram; Sentence processing; Linguistics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001665084,0.0006928912,0.0005338358,0.0005367151,0.0002434437,0.001858559,0.000773473,0.0009726157,0.003483529],"category_scores_gemma":[0.01483334,0.0004528802,0.000502858,0.0003419136,0.0007614742,0.00506696,0.000515374,0.001873149,0.001560327],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005394925,"about_ca_system_score_gemma":0.0004334982,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005408975,"about_ca_topic_score_gemma":0.005435204,"domain_scores_codex":[0.999423,0.000271606,0.0000155244,0.0001825745,0.00005608853,0.00005123807],"domain_scores_gemma":[0.9962567,0.002240482,0.0003594147,0.0004705513,0.0004945753,0.0001782969],"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":[0.0008804948,0.000601396,0.1776748,0.001116568,0.001366652,0.0005467857,0.005379742,0.2107567,0.03908623,0.07905613,0.02530951,0.458225],"study_design_scores_gemma":[0.0000382694,0.0001287083,0.04534838,0.0001150911,0.00007077544,0.0002066292,0.0008113659,0.7666933,0.002334189,0.1796365,0.004531523,0.00008530133],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6334606,0.003004382,0.3250701,0.01164024,0.0004101552,0.00009733,0.001612388,0.001462673,0.02324212],"genre_scores_gemma":[0.981038,0.0003993535,0.01525746,0.0005999398,0.00009923748,0.00005059394,0.0004999956,0.0000928564,0.001962527],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005408975,"threshold_uncertainty_score":0.0116536,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1065612792918263,"score_gpt":0.2939703466880779,"score_spread":0.1874090673962516,"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."}}