{"id":"W3154790171","doi":"10.18653/v1/2021.eacl-main.243","title":"Measuring and Improving Faithfulness of Attention in Neural Machine Translation","year":2021,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada; Ministère de la Défense Nationale","keywords":"Machine translation; Computer science; Regularization (linguistics); Measure (data warehouse); Artificial intelligence; Translation (biology); Divergence (linguistics); Differentiable function; Deep neural networks; Quality (philosophy); Machine learning; Artificial neural network; Natural language processing; Mathematics; Data mining; Linguistics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000134999,0.00003389427,0.00005979999,0.00004117566,0.00001503816,0.00002544195,0.000070786,0.00001626895,0.000002971434],"category_scores_gemma":[0.000009707624,0.00003288352,0.00001312916,0.0001129956,0.0000043315,0.0002721086,0.00004466867,0.00004122542,1.802455e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000006588279,"about_ca_system_score_gemma":0.0000115538,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001286328,"about_ca_topic_score_gemma":0.00022127,"domain_scores_codex":[0.9995511,0.00002704255,0.0001264586,0.0001439869,0.00008662263,0.00006484528],"domain_scores_gemma":[0.9998095,0.0000166498,0.00002215248,0.0001138702,0.00002413364,0.00001368966],"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":[0.000001232715,0.0000118482,0.01476851,0.00003828998,0.00000148253,0.000004937119,0.0003089749,0.001017733,0.03008926,0.002810373,1.247161e-7,0.9509472],"study_design_scores_gemma":[0.0001955364,0.000004145696,0.02311533,0.00001122268,9.930617e-7,0.000006343805,0.00002178621,0.9699887,0.006366147,0.0002521266,0.00000140504,0.00003626901],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4244862,0.0002540454,0.5747893,0.0002034855,0.00003993299,0.00002127375,8.554299e-8,0.00001324592,0.0001924147],"genre_scores_gemma":[0.9666905,0.000001437195,0.03326134,0.00001595532,0.000006196026,0.000001159263,4.446849e-7,0.000001567864,0.00002138469],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.968971,"threshold_uncertainty_score":0.1340951,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04442249320317228,"score_gpt":0.2302141650704904,"score_spread":0.1857916718673182,"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."}}