{"id":"W2903297715","doi":"10.18653/v1/w18-6481","title":"Accurate semantic textual similarity for cleaning noisy parallel corpora using semantic machine translation evaluation metric: The NRC supervised submissions to the Parallel Corpus Filtering task","year":2018,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":35,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada","funders":"","keywords":"Computer science; Natural language processing; Artificial intelligence; Machine translation; Semantic similarity; Redundancy (engineering); Similarity (geometry); Metric (unit); Task (project management); Word (group theory); Fluency; Information retrieval","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":[0.01153507,0.002259122,0.002677415,0.004260744,0.002895596,0.002637095,0.002146189,0.002075648,0.004677496],"category_scores_gemma":[0.04356769,0.0008788451,0.001237382,0.003929639,0.001447978,0.002626952,0.003577723,0.002428627,0.004569218],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001884344,"about_ca_system_score_gemma":0.005185815,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00863625,"about_ca_topic_score_gemma":0.01374879,"domain_scores_codex":[0.9824157,0.00684691,0.001659088,0.002491219,0.00589539,0.0006916195],"domain_scores_gemma":[0.9556187,0.009622455,0.001659318,0.00950869,0.02211878,0.001472089],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002282403,0.001435908,0.01142636,0.002793168,0.000561348,0.001630373,0.002844427,0.03295928,0.08865234,0.004919149,0.168866,0.6816292],"study_design_scores_gemma":[0.001006222,0.002403883,0.0263128,0.0003291142,0.0004240248,0.002769322,0.002093378,0.5517488,0.264708,0.01072058,0.1369208,0.0005630883],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4921089,0.002551839,0.3946506,0.002594718,0.002311024,0.00237731,0.01599644,0.06229441,0.02511474],"genre_scores_gemma":[0.4438438,0.0003882894,0.4833651,0.000526301,0.0004112105,0.001742768,0.04970986,0.007612716,0.01239987],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01153507,"threshold_uncertainty_score":0.06100398,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08794199865200938,"score_gpt":0.3452897288338443,"score_spread":0.2573477301818349,"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."}}