{"id":"W2985974955","doi":"10.22148/16.050","title":"A Shared Task for the Digital Humanities Chapter 3: Description of Submitted Guidelines and Final Evaluation Results","year":2019,"lang":"en","type":"article","venue":"Journal of Cultural Analytics","topic":"Digital Humanities and Scholarship","field":"Arts and Humanities","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Task (project management); Annotation; Reflection (computer programming); Narrative; Computer science; Digital humanities; Mathematics education; Information retrieval; Data science; Artificial intelligence; Psychology; World Wide Web; Linguistics; Programming language; Engineering; Philosophy","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005591805,0.0001171559,0.000218046,0.00007588842,0.0001629931,0.0009152458,0.0001527016,0.0000282026,0.0001745422],"category_scores_gemma":[0.0003072856,0.00006321221,0.0001750489,0.0000200627,0.0001529499,0.001405691,0.00002627215,0.0001064538,0.00000366272],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002693962,"about_ca_system_score_gemma":0.0000257079,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001505647,"about_ca_topic_score_gemma":0.0001133783,"domain_scores_codex":[0.9987152,0.00001424007,0.0006721123,0.00008439334,0.0004020453,0.0001120188],"domain_scores_gemma":[0.9969321,0.0001276143,0.0005437598,0.0000921898,0.00227569,0.00002857432],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003894525,0.0003514416,0.002336359,0.0004247171,0.002025614,0.00001033841,0.1118472,0.001379724,0.0009557154,0.7557247,0.07210046,0.04894926],"study_design_scores_gemma":[0.007272746,0.003223716,0.0103555,0.0007787804,0.001440554,0.0001293613,0.1272807,0.02434484,0.0001185183,0.03711849,0.7871725,0.0007643027],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9794781,0.0008807393,0.00002161664,0.0008212765,0.0004616847,0.000276715,0.0002334207,0.000006862459,0.0178196],"genre_scores_gemma":[0.9863748,0.00004831356,0.00003921171,0.0001348418,0.0005477629,0.000001861205,0.00003299834,0.000007784608,0.01281243],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7186062,"threshold_uncertainty_score":0.8825736,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3084406244997505,"score_gpt":0.3152306524364105,"score_spread":0.006790027936659948,"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."}}