{"paper":{"title":"Dual Hierarchical Dialogue Policy Learning for Legal Inquisitive Conversational Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A dual hierarchical reinforcement learning method lets conversational agents proactively extract information by coordinating high-level strategy and low-level question generation in legal dialogues.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Grace Hui Yang, Shihao Wang, Xubo Lin, Yang Deng, Zezhii Deng","submitted_at":"2026-05-13T19:29:11Z","abstract_excerpt":"Most existing dialogue systems are user-driven, primarily designed to fulfill user requests. However, in many critical real-world scenarios, a conversational agent must proactively extract information to achieve its own objectives rather than merely respond. To address this gap, we introduce \\emph{Inquisitive Conversational Agents (ICAs)} and develop an ICA specifically tailored to U.S. Supreme Court oral arguments. We propose a Dual Hierarchical Reinforcement Learning framework featuring two cooperating RL agents, each with its own policy, to coordinate strategic dialogue management and fine-"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluations on a U.S. Supreme Court dataset show that our method outperforms various baselines across multiple metrics.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The U.S. Supreme Court dataset captures representative judicial questioning patterns and that the dual RL agents can learn effective strategies aligned with legal objectives without additional human feedback or validation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A dual hierarchical RL framework lets agents learn when and how to ask probing questions in U.S. Supreme Court arguments, outperforming baselines on a court dataset.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A dual hierarchical reinforcement learning method lets conversational agents proactively extract information by coordinating high-level strategy and low-level question generation in legal dialogues.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d067827b7a8d05253bdd05c232bd547b576be46fb58f4b7fd76e0a0fb29bad63"},"source":{"id":"2605.14057","kind":"arxiv","version":1},"verdict":{"id":"16196cd9-fdcc-41ba-838d-91ef57de8311","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:17:18.907754Z","strongest_claim":"Evaluations on a U.S. Supreme Court dataset show that our method outperforms various baselines across multiple metrics.","one_line_summary":"A dual hierarchical RL framework lets agents learn when and how to ask probing questions in U.S. Supreme Court arguments, outperforming baselines on a court dataset.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The U.S. Supreme Court dataset captures representative judicial questioning patterns and that the dual RL agents can learn effective strategies aligned with legal objectives without additional human feedback or validation.","pith_extraction_headline":"A dual hierarchical reinforcement learning method lets conversational agents proactively extract information by coordinating high-level strategy and low-level question generation in legal dialogues."},"references":{"count":206,"sample":[{"doi":"","year":2022,"title":"and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Chen, Weizhu , booktitle =","work_id":"5986ff2f-00dc-4b1b-92e8-9c8e22620c26","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue , year =","work_id":"6e9ebdfd-8c57-44f0-86ba-ff95d263587b","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Multiwoz–a large-scale multi-domain wizard-of-oz dataset for task-oriented dialogue modelling","work_id":"9d20f47a-38c7-4f68-ae4c-ce63d559a7b2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the 15th Annual Meeting of the Special Interest Group on Discourse and Dialogue , year =","work_id":"9305a5df-dd04-4eaa-a5c5-40272e136356","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =","work_id":"4c5cc07e-0e81-4b07-9274-5611697a3657","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":206,"snapshot_sha256":"e9be163bad0b2e17623ef08b7ec1d8b8ace24d03768785973792efc52d0db89d","internal_anchors":21},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}