{"id":"W7131955825","doi":"","title":"Evaluation briefs: drawing on translation studies for human evaluation of MT","year":2024,"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":"Machine translation; Translation (biology); Context (archaeology); Set (abstract data type); Machine translation software usability; Evaluation of machine translation; Example-based machine translation","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2592516,0.002267784,0.001673807,0.01033899,0.003873302,0.009131325,0.004136904,0.007310715,0.007569962],"category_scores_gemma":[0.4506366,0.001439133,0.001394505,0.006458831,0.006690211,0.02817895,0.006831777,0.008371823,0.00260435],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0117258,"about_ca_system_score_gemma":0.006304591,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003983743,"about_ca_topic_score_gemma":0.007780179,"domain_scores_codex":[0.7590052,0.19604,0.01896071,0.004099756,0.02081411,0.00108007],"domain_scores_gemma":[0.414185,0.461082,0.02863532,0.02289986,0.06912127,0.004076614],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0009012598,0.0005116769,0.00215621,0.01029458,0.0003604934,0.0004936867,0.01143878,0.003190511,0.003178788,0.2786984,0.0916622,0.5971135],"study_design_scores_gemma":[0.0003687819,0.003376372,0.004985533,0.01762698,0.0006256083,0.001255898,0.008201176,0.007840334,0.006570778,0.2611074,0.6875861,0.0004550102],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01061882,0.1544228,0.5909013,0.1680332,0.01189723,0.00548573,0.001117284,0.001417515,0.05610612],"genre_scores_gemma":[0.2216868,0.0628162,0.6224651,0.05169529,0.00854414,0.01661731,0.001439185,0.001140018,0.01359605],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2592516,"threshold_uncertainty_score":0.9134746,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1616407711959469,"score_gpt":0.44272982507029,"score_spread":0.2810890538743431,"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."}}