{"id":"W2164441316","doi":"","title":"PORT: a Precision-Order-Recall MT Evaluation Metric for Tuning","year":2012,"lang":"en","type":"article","venue":"NPARC","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada","funders":"","keywords":"BLEU; Computer science; Metric (unit); Port (circuit theory); Machine translation; Artificial intelligence; Precision and recall; Translation (biology); Recall; Natural language processing; Machine learning; Engineering; Electronic engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.009344005,0.002170245,0.00154626,0.007131719,0.001163543,0.002539852,0.001643188,0.001830389,0.004670732],"category_scores_gemma":[0.04430008,0.0005653444,0.0009153574,0.004859668,0.0007074289,0.003650971,0.001676379,0.001226721,0.003200011],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00144858,"about_ca_system_score_gemma":0.001252271,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002786378,"about_ca_topic_score_gemma":0.004085547,"domain_scores_codex":[0.988186,0.004841689,0.001767506,0.001478583,0.003475421,0.0002509459],"domain_scores_gemma":[0.973435,0.01364492,0.002159988,0.004023375,0.00635039,0.0003862907],"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":[0.001518509,0.0005703535,0.02373209,0.002032067,0.001154049,0.0003137436,0.000530909,0.04313093,0.06470606,0.005760361,0.06690937,0.7896416],"study_design_scores_gemma":[0.0005229572,0.003440111,0.07626218,0.0003684706,0.0009791697,0.003528209,0.0004361099,0.5907631,0.2196704,0.01622518,0.08699013,0.0008139299],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1362569,0.005389561,0.7501304,0.0005762035,0.0007267901,0.001729791,0.01382016,0.05424133,0.03712893],"genre_scores_gemma":[0.4688106,0.0008422564,0.4938296,0.0003758988,0.0002725544,0.001975679,0.02084451,0.005099652,0.007949305],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.009344005,"threshold_uncertainty_score":0.04941642,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03752141005855442,"score_gpt":0.3342121167803518,"score_spread":0.2966907067217974,"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."}}