{"id":"W7125770938","doi":"10.21428/594757db.d4fa15c1","title":"Conversation Alignment for Task-Oriented Dialogue Agents","year":2024,"lang":"en","type":"article","venue":"","topic":"Speech and dialogue systems","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; Queen's University","funders":"","keywords":"Conversation; Component (thermodynamics); Scope (computer science); Hyperparameter; Test (biology); Dialog system; Beam search","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":[0.0002068916,0.00007552591,0.00007660157,0.00007004265,0.00004034227,0.0001533759,0.0002091313,0.00003712241,0.00002165328],"category_scores_gemma":[0.00002364062,0.00006163568,0.00006212999,0.0001806609,0.00001033095,0.0002552192,0.00005366564,0.0000242242,0.0003205306],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005970499,"about_ca_system_score_gemma":0.00005117776,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007832003,"about_ca_topic_score_gemma":0.00001181911,"domain_scores_codex":[0.9992155,0.00002084897,0.0001432336,0.0002768048,0.0001659209,0.0001776491],"domain_scores_gemma":[0.9995777,0.00007665547,0.00001883315,0.0002218172,0.00003419262,0.00007078107],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001371113,0.00006285396,0.0003180246,0.0001209493,0.00008259254,0.0000268042,0.003107674,0.00005080907,0.003267676,0.7318841,0.2360763,0.02498852],"study_design_scores_gemma":[0.001197442,0.0003225328,0.001108289,0.00007532931,0.0000221321,0.0000181074,0.0002203741,0.2919497,0.01979074,0.01800863,0.6668354,0.0004513894],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.001679302,0.00009397209,0.9881052,0.001206737,0.003806767,0.0003971038,0.0000127117,0.0003653258,0.004332901],"genre_scores_gemma":[0.9893366,0.000004209527,0.008485862,0.0004095441,0.0002036479,0.00007927499,0.0000464455,0.000006504657,0.001427954],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9876572,"threshold_uncertainty_score":0.4119879,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02180640931892866,"score_gpt":0.2636772030765954,"score_spread":0.2418707937576668,"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."}}