{"id":"W2966610483","doi":"10.48550/arxiv.1907.12009","title":"Representation Degeneration Problem in Training Natural Language\\n Generation Models","year":2019,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":105,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science; Regularization (linguistics); Artificial intelligence; Machine translation; Representation (politics); Maximization; Natural language processing; Language model; Natural language; Natural language understanding; Tying; Machine learning; Mathematical optimization; Mathematics","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":[0.006052684,0.00121418,0.001453824,0.0008406077,0.0007764135,0.001347104,0.002209646,0.002164594,0.001949733],"category_scores_gemma":[0.02181844,0.00101383,0.001010145,0.001150303,0.001527609,0.004199697,0.002665376,0.003566923,0.001008441],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0014439,"about_ca_system_score_gemma":0.001394185,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003905126,"about_ca_topic_score_gemma":0.005250101,"domain_scores_codex":[0.9976271,0.001239093,0.0001495544,0.0005726489,0.0002531917,0.0001583587],"domain_scores_gemma":[0.9913096,0.006590974,0.0004382056,0.0009169207,0.0005732646,0.0001709971],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002754527,0.0002382346,0.0037847,0.0001982787,0.0001153994,0.0002110919,0.000384395,0.7679766,0.004260249,0.02328875,0.005529552,0.1937373],"study_design_scores_gemma":[0.00001244599,0.00002327486,0.00008107114,0.000007481238,0.000004949594,0.00002002221,0.00001424863,0.9887587,0.0008748999,0.009878499,0.000319815,0.00000463823],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07138364,0.0006291447,0.9236191,0.001073585,0.00006008728,0.0000760558,0.0001400697,0.00136733,0.00165095],"genre_scores_gemma":[0.7471437,0.0004459569,0.2444838,0.0009408613,0.0001644243,0.0003680731,0.001383005,0.0004391177,0.004631225],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006052684,"threshold_uncertainty_score":0.03201008,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1649596966506116,"score_gpt":0.2195593484739972,"score_spread":0.05459965182338553,"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."}}