{"id":"W7131974924","doi":"","title":"On the stability of system rankings at WMT","year":2021,"lang":"en","type":"article","venue":"NPARC","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Annotation; Task (project management); Context (archaeology); Quality (philosophy); Stability (learning theory); Ranking (information retrieval); Outlier","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08150412,0.0007511564,0.001888679,0.004659323,0.003997456,0.006488522,0.002464219,0.002491487,0.003768234],"category_scores_gemma":[0.4363446,0.001256939,0.001010921,0.004748086,0.005575934,0.009444869,0.00464784,0.005205967,0.002113261],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004565067,"about_ca_system_score_gemma":0.003036523,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01337288,"about_ca_topic_score_gemma":0.01357877,"domain_scores_codex":[0.9325464,0.03713613,0.002988085,0.01234714,0.01279734,0.002184941],"domain_scores_gemma":[0.4732588,0.387702,0.02607616,0.05950304,0.04896735,0.004492691],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006577061,0.0008787057,0.3361893,0.001296297,0.002155259,0.001014419,0.01601416,0.1808898,0.0430527,0.08218542,0.02834154,0.3014053],"study_design_scores_gemma":[0.0002520392,0.001120424,0.2179514,0.0001999753,0.0002709261,0.0006323823,0.00336865,0.6462831,0.03185039,0.08890594,0.008677911,0.0004868489],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6406881,0.001395389,0.3320015,0.006013016,0.0004179827,0.0003498527,0.003040064,0.002900036,0.01319394],"genre_scores_gemma":[0.9529908,0.0001023725,0.04161539,0.0003768,0.0001210391,0.0002608094,0.002286738,0.0006702958,0.001575776],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08150412,"threshold_uncertainty_score":0.4310402,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01472361285564163,"score_gpt":0.2381138275837487,"score_spread":0.2233902147281071,"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."}}