{"id":"W2143177362","doi":"10.3115/v1/w14-3346","title":"A Systematic Comparison of Smoothing Techniques for Sentence-Level BLEU","year":2014,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":188,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; National Research Council Canada","funders":"","keywords":"BLEU; Computer science; Smoothing; Sentence; Artificial intelligence; Natural language processing; Speech recognition; Machine translation; Computer vision","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":[0.01347895,0.001799138,0.001256354,0.007004101,0.00137746,0.00165396,0.001843076,0.001250185,0.004430609],"category_scores_gemma":[0.05652611,0.0005860056,0.001555768,0.007263537,0.0005386755,0.003060143,0.001674874,0.001955507,0.004111981],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001328454,"about_ca_system_score_gemma":0.001841847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004306984,"about_ca_topic_score_gemma":0.008722151,"domain_scores_codex":[0.9888859,0.004847826,0.001144747,0.001911755,0.00285188,0.0003579531],"domain_scores_gemma":[0.9572591,0.0206582,0.001698074,0.009722034,0.01031189,0.0003507323],"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":[0.002205012,0.0004086947,0.007816457,0.002633109,0.001518817,0.00006367067,0.0006667121,0.01969358,0.02376148,0.008643453,0.03174486,0.9008442],"study_design_scores_gemma":[0.0009111863,0.005497408,0.1034744,0.002502865,0.002995435,0.001750015,0.001305861,0.3461384,0.2140518,0.04262758,0.2772306,0.001514393],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1141998,0.04113705,0.7545819,0.001054026,0.002071278,0.002174617,0.01379244,0.04271761,0.0282714],"genre_scores_gemma":[0.2981131,0.01019144,0.6403614,0.0006108776,0.0003488898,0.002932439,0.0292536,0.008201726,0.009986575],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01347895,"threshold_uncertainty_score":0.07128435,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03508668186407472,"score_gpt":0.3199661980098258,"score_spread":0.2848795161457511,"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."}}