{"id":"W4386576803","doi":"10.18653/v1/2023.findings-eacl.106","title":"More Robust Schema-Guided Dialogue State Tracking via Tree-Based Paraphrase Ranking","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Toyota Motor Europe; Atomic Energy of Canada Limited; Department of Science and Technology, Ministry of Science and Technology, India","keywords":"Computer science; Schema (genetic algorithms); Artificial intelligence; Natural language processing; Paraphrase; Scalability; Natural language understanding; Robustness (evolution); Machine learning; Information retrieval; Natural language; Database","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006435193,0.0002084496,0.0002328081,0.0002822766,0.0001825862,0.0002298612,0.0008436025,0.00006833783,0.00002856539],"category_scores_gemma":[0.00007300016,0.0001922255,0.0001217445,0.0009782452,0.00003774392,0.0004827135,0.0002054741,0.0001791934,0.0001440061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004775419,"about_ca_system_score_gemma":0.0000967916,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002029875,"about_ca_topic_score_gemma":0.0001226939,"domain_scores_codex":[0.9979208,0.00007117532,0.0003952285,0.0006175631,0.00043388,0.0005613912],"domain_scores_gemma":[0.998687,0.0001914637,0.00009230954,0.0008057679,0.00008300105,0.0001404925],"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":[0.00001855937,0.0001123885,0.007507117,0.0001178704,0.00006337388,0.0005510542,0.002889922,0.6401202,0.0127462,0.008139472,0.002542603,0.3251912],"study_design_scores_gemma":[0.0006611666,0.00001447405,0.002105266,0.00003859695,0.000005134244,0.000005149045,0.00002519361,0.9849444,0.009813371,0.001917814,0.0002037696,0.0002656571],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1538911,0.0000249244,0.8423873,0.001474702,0.0003209318,0.0001712573,0.000001634951,0.001102319,0.0006258925],"genre_scores_gemma":[0.8876093,0.000005243427,0.1110396,0.0008554194,0.00008593179,0.00002654382,0.00001235466,0.00002241444,0.0003431718],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7337182,"threshold_uncertainty_score":0.783873,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06953959777686648,"score_gpt":0.2812464530772721,"score_spread":0.2117068553004056,"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."}}