{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003297297,0.0003222863,0.0002870007,0.0002397768,0.001204816,0.0006241788,0.001618148,0.0001262637,0.00003604198],"category_scores_gemma":[0.0008121928,0.0001878376,0.0001396513,0.001363136,0.00009801312,0.0008062198,0.0003434127,0.0002998774,0.00001291787],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001459685,"about_ca_system_score_gemma":0.0001972748,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004407414,"about_ca_topic_score_gemma":0.0005995017,"domain_scores_codex":[0.9970009,0.0003220962,0.0005339824,0.0006508244,0.0009626198,0.0005295266],"domain_scores_gemma":[0.9973196,0.0005919831,0.0002596696,0.0008971505,0.000805924,0.0001257162],"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":[0.0005595486,0.0003709809,0.0009063749,0.0003294029,0.0002862906,0.00001915497,0.01706505,0.03216654,0.1187154,0.02676625,0.005138101,0.7976769],"study_design_scores_gemma":[0.0005173001,0.0001549559,0.0003516857,0.00007948359,0.00008891809,0.00003191039,0.0001004146,0.98749,0.003665493,0.006688869,0.0005190144,0.0003119316],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01146136,0.0008411877,0.9795366,0.005344687,0.0002683175,0.001944175,0.00001193733,0.0004709247,0.0001207908],"genre_scores_gemma":[0.5993834,0.00000684341,0.3995822,0.000771067,0.0001238545,0.0000656851,0.00001352013,0.00002012674,0.0000333072],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9553235,"threshold_uncertainty_score":0.9266593,"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."}}