{"id":"W4225331178","doi":"10.46298/jdmdh.9123","title":"Towards an empirical evaluation of translated texts and translation quality","year":2022,"lang":"en","type":"article","venue":"Journal of Data Mining & Digital Humanities","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec à Trois-Rivières","funders":"","keywords":"Computer science; Machine translation; Natural language processing; Phraseology; Source text; Artificial intelligence; Context (archaeology); Quality (philosophy); Translation (biology); Linguistics","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":[0.1096792,0.001648914,0.001583508,0.01161798,0.001295366,0.008555014,0.002676128,0.002551382,0.006669734],"category_scores_gemma":[0.4075953,0.0007758859,0.001064213,0.01337088,0.007156247,0.01044962,0.006679961,0.002449366,0.002183257],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002321492,"about_ca_system_score_gemma":0.002766139,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001503113,"about_ca_topic_score_gemma":0.0009654546,"domain_scores_codex":[0.8055444,0.1300425,0.0131502,0.01119476,0.03883269,0.001235484],"domain_scores_gemma":[0.5494729,0.2975434,0.0410354,0.04148311,0.06808916,0.002375958],"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":[0.001261264,0.001561215,0.2139689,0.005643629,0.001402917,0.0005024355,0.02055697,0.0184587,0.01037857,0.1049327,0.01343072,0.6079019],"study_design_scores_gemma":[0.0005740297,0.003506098,0.4276375,0.00587844,0.001383713,0.001590173,0.02112822,0.168788,0.03840054,0.2302571,0.1001939,0.0006623501],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2259994,0.00581876,0.7210785,0.002609718,0.0003137764,0.00181292,0.002801961,0.001252506,0.03831253],"genre_scores_gemma":[0.7347958,0.001489247,0.2532876,0.0004437496,0.0003359493,0.00258176,0.003720505,0.0007277898,0.002617711],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1096792,"threshold_uncertainty_score":0.5800459,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2710737773030953,"score_gpt":0.4273664435802014,"score_spread":0.156292666277106,"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."}}