{"id":"W2250389127","doi":"","title":"Identifying Sources of Inter-Annotator Variation: Evaluating Two Models of Argument Analysis","year":2010,"lang":"en","type":"article","venue":"","topic":"Discourse Analysis in Language Studies","field":"Arts and Humanities","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Variation (astronomy); Argument (complex analysis); Computer science; Rhetorical question; Domain (mathematical analysis); Natural language processing; Data science; Artificial intelligence; Management science; Linguistics; Mathematics; Engineering","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3573104,0.003041339,0.003197759,0.01027025,0.006008546,0.0180613,0.00798634,0.008399118,0.002123202],"category_scores_gemma":[0.6258503,0.002788116,0.004229814,0.007460153,0.00900077,0.0195529,0.0164695,0.007006612,0.001056195],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01208247,"about_ca_system_score_gemma":0.006865767,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01013883,"about_ca_topic_score_gemma":0.01157485,"domain_scores_codex":[0.5132349,0.4171053,0.01872509,0.02169001,0.02668009,0.002564613],"domain_scores_gemma":[0.1193418,0.7937284,0.02293196,0.03672232,0.02540675,0.001868724],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.02332711,0.005068166,0.2112865,0.007525721,0.008826341,0.001458537,0.1801954,0.1443596,0.02135057,0.03881131,0.009728267,0.3480624],"study_design_scores_gemma":[0.002136488,0.00305563,0.05279764,0.001776708,0.002132909,0.0008221373,0.02002853,0.8311707,0.02038621,0.05358404,0.01098516,0.001123794],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5606222,0.00178453,0.4176779,0.003335953,0.0004848659,0.004214208,0.00127924,0.002044467,0.008556598],"genre_scores_gemma":[0.7335199,0.0003269595,0.2585592,0.0006189059,0.0001496143,0.003027614,0.001808428,0.0009231684,0.001066218],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3573104,"threshold_uncertainty_score":0.7925507,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06126207947042338,"score_gpt":0.3448110371695646,"score_spread":0.2835489576991412,"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."}}