{"id":"W2907526436","doi":"10.4000/books.aaccademia.4512","title":"Overview of the EVALITA 2018 Solving language games (NLP4FUN) Task","year":2018,"lang":"en","type":"book-chapter","venue":"Accademia University Press eBooks","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Research Council Canada; Università degli Studi di Napoli Federico II","keywords":"Task (project management); Computer science; Set (abstract data type); Artificial intelligence; Range (aeronautics); Word (group theory); Natural language processing; World Wide Web; Multimedia; Human–computer interaction; Linguistics; Programming language; Engineering","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":[0.0009548437,0.001892645,0.001044399,0.002498964,0.001194089,0.004229412,0.002498774,0.001753777,0.07725825],"category_scores_gemma":[0.0022349,0.000721373,0.0008097524,0.002586,0.0005941842,0.004039604,0.002986145,0.002265884,0.0724652],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001463767,"about_ca_system_score_gemma":0.001735638,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004221042,"about_ca_topic_score_gemma":0.006768084,"domain_scores_codex":[0.9989111,0.0001859496,0.00007816136,0.0002578945,0.0004435311,0.0001233399],"domain_scores_gemma":[0.9993033,0.0002007475,0.00002359119,0.0001181899,0.000210854,0.0001432939],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001536427,0.0003288789,0.0002800171,0.001129468,0.0000251461,0.0001101998,0.0003594369,0.001535079,0.004165563,0.01318536,0.5028552,0.475872],"study_design_scores_gemma":[0.00002592155,0.00006309577,0.0005359203,0.0001525077,0.000007087316,0.0002146204,0.00006101301,0.002357232,0.001502622,0.006586403,0.9884691,0.00002455369],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.009166528,0.04357043,0.1960378,0.004427037,0.003132843,0.001886812,0.02210446,0.05076084,0.6689133],"genre_scores_gemma":[0.04232929,0.02245536,0.3119821,0.002491576,0.001769686,0.003167372,0.1400656,0.01692813,0.4588109],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.07725825,"threshold_uncertainty_score":0.2584546,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03451380042961945,"score_gpt":0.2656893194680154,"score_spread":0.231175519038396,"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."}}