{"id":"W3186655327","doi":"10.18653/v1/2021.gem-1.10","title":"The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":148,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Benchmark (surveying); Natural language generation; Natural language; Natural language processing; Geology","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":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.00157504,0.0002261118,0.0001818595,0.0001519636,0.0002965416,0.002048264,0.001156911,0.0002092167,0.00004222645],"category_scores_gemma":[0.001058168,0.0001493558,0.00006556823,0.0004853846,0.00003449932,0.0004216216,0.002278986,0.0006433177,0.000004186976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001253813,"about_ca_system_score_gemma":0.0003007167,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000357227,"about_ca_topic_score_gemma":0.00006519379,"domain_scores_codex":[0.997835,0.0002472432,0.000278266,0.0006438995,0.0007716597,0.000223907],"domain_scores_gemma":[0.99788,0.0001811382,0.0001943718,0.0008316449,0.0008597946,0.00005305515],"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":[0.000001487924,0.00001929264,0.00001668081,0.00007532784,0.00005484047,0.00002038842,0.001230864,0.00003556276,0.003763549,0.01265242,0.007729639,0.9743999],"study_design_scores_gemma":[0.0001135992,0.00001231733,0.0001011992,0.00005346695,0.00003731669,0.00002902718,0.00008362252,0.9661076,0.02904865,0.003410601,0.0006436273,0.0003589582],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.01246449,0.5956074,0.3808326,0.005060739,0.002107064,0.001084191,0.000004473457,0.0008881784,0.001950851],"genre_scores_gemma":[0.6354415,0.0004938639,0.3616202,0.0005628524,0.0002547512,0.00008361603,0.00008464331,0.00001318592,0.00144535],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.974041,"threshold_uncertainty_score":0.9989877,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02864547784444409,"score_gpt":0.3241813393370074,"score_spread":0.2955358614925633,"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."}}