{"id":"W2915258134","doi":"","title":"Semeval-2012 Task 8: Cross-lingual Textual Entailment for Content Synchronization","year":2012,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Computer science; Textual entailment; Task (project management); Natural language processing; SemEval; Logical consequence; Inference; Synchronization (alternating current); Artificial intelligence; Process (computing); Programming language","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":[0.009855504,0.002400972,0.00236196,0.003110707,0.00277687,0.002758418,0.003338025,0.003677738,0.01906778],"category_scores_gemma":[0.02719709,0.000932792,0.0021628,0.002219277,0.001524776,0.007703342,0.00882631,0.003558253,0.009387188],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001891027,"about_ca_system_score_gemma":0.004103371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004680272,"about_ca_topic_score_gemma":0.005689038,"domain_scores_codex":[0.9897388,0.004404526,0.001081118,0.003042012,0.001186798,0.0005468277],"domain_scores_gemma":[0.9824157,0.00869923,0.0007366086,0.004342258,0.003101473,0.0007047149],"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":[0.004778834,0.00202623,0.007508954,0.006375344,0.0008744977,0.001862825,0.005210535,0.01236825,0.07164001,0.01864356,0.241395,0.6273159],"study_design_scores_gemma":[0.003035076,0.002401134,0.02520007,0.000692374,0.001142161,0.005662485,0.007480496,0.1956081,0.3036068,0.04377081,0.4107867,0.0006138453],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3206553,0.003884809,0.4973944,0.003579735,0.001706228,0.004916281,0.05937194,0.07417946,0.03431196],"genre_scores_gemma":[0.3618034,0.0004622576,0.4752979,0.0008468425,0.0003820899,0.001973269,0.1416851,0.005151358,0.0123978],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01906778,"threshold_uncertainty_score":0.06378812,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03341947877365044,"score_gpt":0.3240185823605777,"score_spread":0.2905991035869273,"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."}}