{"id":"W2099032682","doi":"","title":"Neutralizing Linguistically Problematic Annotations in Unsupervised Dependency Parsing Evaluation","year":2011,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":42,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Parsing; Computer science; Dependency (UML); Annotation; Dependency grammar; Artificial intelligence; Natural language processing; Set (abstract data type); Task (project management); Measure (data warehouse); Quality (philosophy); Data mining; Programming language","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":[0.02716921,0.002284901,0.001774505,0.002780632,0.001641232,0.003445496,0.002277048,0.002856299,0.001264144],"category_scores_gemma":[0.09880848,0.0006794934,0.0006910721,0.002368794,0.002033683,0.00507837,0.004443283,0.002447488,0.0007731362],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001280394,"about_ca_system_score_gemma":0.001967453,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001578392,"about_ca_topic_score_gemma":0.004092908,"domain_scores_codex":[0.9565113,0.0278104,0.002854489,0.003378671,0.008543336,0.0009017856],"domain_scores_gemma":[0.8748586,0.09185588,0.0051338,0.01253045,0.01456417,0.001057042],"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":[0.002146445,0.00078455,0.04040648,0.001549176,0.0007208012,0.0008998192,0.002789665,0.07411489,0.1150171,0.02514509,0.01253813,0.7238879],"study_design_scores_gemma":[0.0002052773,0.001239663,0.02849403,0.0003981896,0.0005361846,0.001928071,0.001623466,0.6504807,0.2274278,0.06559736,0.02158536,0.0004839473],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2719042,0.001231693,0.7113745,0.0004775871,0.0002124343,0.0004671464,0.0005600523,0.005660979,0.008111442],"genre_scores_gemma":[0.7166696,0.0002780091,0.2772961,0.0003650258,0.00006071851,0.0003862768,0.001475776,0.001516966,0.001951532],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02716921,"threshold_uncertainty_score":0.1436862,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07206090392163111,"score_gpt":0.3153775708490825,"score_spread":0.2433166669274514,"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."}}