{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T13:59:30Z","timestamp":1781704770887,"version":"3.54.5"},"reference-count":68,"publisher":"MIT Press","license":[{"start":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T00:00:00Z","timestamp":1781654400000},"content-version":"vor","delay-in-days":167,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["direct.mit.edu"],"crossmark-restriction":true},"short-container-title":[],"published-print":{"date-parts":[[2026,6,1]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>User simulators are essential to conversational AI, enabling scalable agent development and evaluation through simulated interactions. While current Large Language Models (LLMs) have advanced user simulation capabilities, we reveal that they struggle to consistently demonstrate goal-oriented behavior across multi-turn conversations, which is a critical limitation that compromises their reliability in downstream applications. We introduce User Goal State Tracking (UGST), a novel framework that tracks user goal progression throughout conversations. Leveraging UGST, we present a three-stage methodology for developing user simulators that can autonomously track goal progression and reason to generate goal-aligned responses. Moreover, we establish comprehensive evaluation metrics for measuring goal alignment in user simulators, and demonstrate that our approach yields substantial improvements across two benchmarks (MultiWOZ 2.4 and \u03c4-Bench). Our contributions address a critical gap in conversational AI and establish UGST as an essential framework for developing goal-aligned user simulators. All code and data is released to facilitate future research 1.<\/jats:p>","DOI":"10.1162\/tacl.a.687","type":"journal-article","created":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T13:27:25Z","timestamp":1781702845000},"page":"852-876","update-policy":"https:\/\/doi.org\/10.1162\/mitpressjournals.corrections.policy","source":"Crossref","is-referenced-by-count":0,"title":["Goal Alignment in LLM-Based User Simulators for Conversational AI"],"prefix":"10.1162","volume":"14","author":[{"given":"Shuhaib","family":"Mehri","sequence":"first","affiliation":[{"name":"University of Illinois Urbana-Champaign, mehri2@illinois.edu"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiaocheng","family":"Yang","sequence":"additional","affiliation":[{"name":"University of Illinois Urbana-Champaign"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Takyoung","family":"Kim","sequence":"additional","affiliation":[{"name":"University of Illinois Urbana-Champaign"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Gokhan","family":"Tur","sequence":"additional","affiliation":[{"name":"University of Illinois Urbana-Champaign"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shikib","family":"Mehri","sequence":"additional","affiliation":[{"name":"Contextual AI"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Dilek","family":"Hakkani-T\u00fcr","sequence":"additional","affiliation":[{"name":"University of Illinois Urbana-Champaign, dilek@illinois.edu"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"281","published-online":{"date-parts":[[2026,6,1]]},"reference":[{"key":"2026061709272029900_bib1","volume-title":"A desideratum for conversational agents: Capabilities, challenges, and future directions","author":"Acikgoz","year":"2025"},{"key":"2026061709272029900_bib2","volume-title":"Simulating user diversity in task-oriented dialogue systems using large language models","author":"Ahmad","year":"2025"},{"key":"2026061709272029900_bib3","doi-asserted-by":"crossref","first-page":"101697","DOI":"10.1016\/j.csl.2024.101697","article-title":"Prompting large language models for user simulation in task-oriented dialogue systems","volume":"89","author":"Algherairy","year":"2025","journal-title":"Computer Speech & Language"},{"key":"2026061709272029900_bib4","article-title":"Concrete problems in ai safety","volume-title":"arXiv preprint arXiv:1606.06565","author":"Amodei","year":"2016"},{"key":"2026061709272029900_bib5","doi-asserted-by":"crossref","DOI":"10.21437\/Interspeech.2016-1175","article-title":"A sequence-to-sequence model for user simulation in spoken dialogue systems","volume-title":"arXiv preprint arXiv:1607.00070","author":"Asri","year":"2016"},{"key":"2026061709272029900_bib6","article-title":"Persuade me if you can: A framework for evaluating persuasion effectiveness and susceptibility among large language models","volume-title":"arXiv preprint arXiv:2503.01829","author":"Bozdag","year":"2025"},{"key":"2026061709272029900_bib7","first-page":"32","article-title":"On the utility of learning about humans for human-ai coordination","volume-title":"Advances in neural information processing systems","author":"Carroll","year":"2019"},{"key":"2026061709272029900_bib8","volume-title":"Teaching models new apis: Domain-agnostic simulators for task oriented dialogue","author":"Chen","year":"2021"},{"key":"2026061709272029900_bib9","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2022.findings-emnlp.90","article-title":"Is multiwoz a solved task? an interactive tod evaluation framework with user simulator","volume-title":"arXiv preprint arXiv:2210.14529","author":"Cheng","year":"2022"},{"key":"2026061709272029900_bib10","doi-asserted-by":"crossref","first-page":"1248","DOI":"10.18653\/v1\/2022.findings-emnlp.90","article-title":"Is MultiWOZ a solved task? an interactive TOD evaluation framework with user simulator","volume-title":"Findings of the Association for Computational Linguistics: EMNLP 2022","author":"Cheng","year":"2022"},{"key":"2026061709272029900_bib11","article-title":"User simulation with large language models for evaluating task-oriented dialogue","volume-title":"arXiv preprint arXiv:2309.13233","author":"Davidson","year":"2023"},{"key":"2026061709272029900_bib12","doi-asserted-by":"crossref","first-page":"80","DOI":"10.1109\/ASRU.1997.658991","article-title":"User modeling for spoken dialogue system evaluation","volume-title":"1997 IEEE Workshop on Automatic Speech Recognition and Understanding Proceedings","author":"Eckert","year":"1997"},{"key":"2026061709272029900_bib13","doi-asserted-by":"crossref","first-page":"80","DOI":"10.1109\/ASRU.1997.658991","article-title":"User modeling for spoken dialogue system evaluation","volume-title":"1997 IEEE Workshop on Automatic Speech Recognition and Understanding Proceedings","author":"Eckert","year":"1997"},{"key":"2026061709272029900_bib14","volume-title":"The llama 3 herd of models","author":"Grattafiori","year":"2024"},{"key":"2026061709272029900_bib15","volume-title":"Rubrics as rewards: Reinforcement learning beyond verifiable domains","author":"Gunjal","year":"2025"},{"key":"2026061709272029900_bib16","article-title":"Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning","volume-title":"arXiv preprint arXiv:2501.12948","author":"Guo","year":"2025"},{"key":"2026061709272029900_bib17","doi-asserted-by":"crossref","first-page":"900","DOI":"10.1109\/SLT.2018.8639652","article-title":"User modeling for task oriented dialogues","volume-title":"2018 IEEE Spoken Language Technology Workshop (SLT)","author":"G\u00fcr","year":"2018"},{"key":"2026061709272029900_bib18","first-page":"263","article-title":"The second dialog state tracking challenge","volume-title":"Proceedings of the 15th Annual Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL)","author":"Henderson","year":"2014"},{"key":"2026061709272029900_bib19","doi-asserted-by":"crossref","DOI":"10.1109\/SLT61566.2024.10832298","volume-title":"Large language models as user-agents for evaluating task-oriented-dialogue systems","author":"Kazi","year":"2024"},{"key":"2026061709272029900_bib20","first-page":"116","article-title":"Parameter estimation for agenda-based user simulation","volume-title":"Proceedings of the SIGDIAL 2010 Conference","author":"Keizer","year":"2010"},{"key":"2026061709272029900_bib21","article-title":"Pipa: A unified evaluation protocol for diagnosing interactive planning agents","volume-title":"arXiv preprint arXiv:2505.01592","author":"Kim","year":"2025"},{"key":"2026061709272029900_bib22","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/W18-5007","article-title":"Neural user simulation for corpus-based policy optimisation for spoken dialogue systems","volume-title":"arXiv preprint arXiv:1805.06966","author":"Kreyssig","year":"2018"},{"key":"2026061709272029900_bib23","article-title":"Llms get lost in multi-turn conversation","volume-title":"arXiv preprint arXiv:2505.06120","author":"Laban","year":"2025"},{"key":"2026061709272029900_bib24","article-title":"Measuring and controlling instruction (in) stability in language model dialogs","volume-title":"arXiv preprint arXiv:2402.10962","author":"Li","year":"2024"},{"key":"2026061709272029900_bib25","volume-title":"A user simulator for task-completion dialogues","author":"Li","year":"2017"},{"key":"2026061709272029900_bib26","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2022.sigdial-1.28","article-title":"Gentus: Simulating user behaviour and language in task-oriented dialogues with generative transformers","volume-title":"arXiv preprint arXiv:2208.10817","author":"Lin","year":"2022"},{"key":"2026061709272029900_bib27","doi-asserted-by":"crossref","first-page":"270","DOI":"10.18653\/v1\/2022.sigdial-1.28","article-title":"GenTUS: Simulating user behaviour and language in task-oriented dialogues with generative transformers","volume-title":"Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue","author":"Lin","year":"2022"},{"key":"2026061709272029900_bib28","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2021.sigdial-1.47","article-title":"Domain-independent user simulation with transformers for task-oriented dialogue systems","volume-title":"arXiv preprint arXiv:2106.08838","author":"Lin","year":"2021"},{"key":"2026061709272029900_bib29","first-page":"1","article-title":"One cannot stand for everyone! leveraging multiple user simulators to train task-oriented dialogue systems","volume-title":"Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","author":"Liu","year":"2023"},{"key":"2026061709272029900_bib30","volume-title":"Toolsand-box: A stateful, conversational, interactive evaluation benchmark for llm tool use capabilities","author":"Lu","year":"2025"},{"key":"2026061709272029900_bib31","doi-asserted-by":"crossref","first-page":"5414","DOI":"10.63317\/5cp2zxmyham6","article-title":"DuetSim: Building user simulator with dual large language models for task-oriented dialogues","volume-title":"Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)","author":"Luo","year":"2024"},{"issue":"2","key":"2026061709272029900_bib32","doi-asserted-by":"crossref","first-page":"381","DOI":"10.3758\/BRM.42.2.381","article-title":"Mtld, vocd-d, and hd-d: A validation study of sophisticated approaches to lexical diversity assessment","volume":"42","author":"McCarthy","year":"2010","journal-title":"Behavior research methods"},{"key":"2026061709272029900_bib33","doi-asserted-by":"crossref","first-page":"681","DOI":"10.18653\/v1\/2020.acl-main.64","article-title":"USR: An unsupervised and reference free evaluation metric for dialog generation","volume-title":"Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics","author":"Mehri","year":"2020"},{"key":"2026061709272029900_bib34","first-page":"42","article-title":"Automatic evaluation of generative models with instruction tuning","volume-title":"Proceedings of the Third Workshop on Natural Language Generation, Evaluation, and Metrics (GEM)","author":"Mehri","year":"2023"},{"key":"2026061709272029900_bib35","article-title":"Planning for goal-oriented dialogue systems","volume-title":"arXiv preprint arXiv:1910.08137","author":"Muise","year":"2019"},{"key":"2026061709272029900_bib36","volume-title":"Reinforcement learning finetunes small subnetworks in large language models","author":"Mukherjee","year":"2025"},{"key":"2026061709272029900_bib37","doi-asserted-by":"crossref","first-page":"8724","DOI":"10.18653\/v1\/2024.acl-long.473","article-title":"Enhancing dialogue state tracking models through LLM-backed user-agents simulation","volume-title":"Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","author":"Niu","year":"2024"},{"key":"2026061709272029900_bib38","volume-title":"Gpt-4o mini: advancing cost-efficient intelligence","author":"OpenAI","year":"2024"},{"key":"2026061709272029900_bib39","volume-title":"When benchmarks talk: Re-evaluating code llms with interactive feedback","author":"Pan","year":"2025"},{"key":"2026061709272029900_bib40","article-title":"Simulating user agents for embodied conversational AI","volume-title":"NeurIPS 2024 Workshop on Open-World Agents","author":"Philipov","year":"2024"},{"issue":"2","key":"2026061709272029900_bib41","doi-asserted-by":"crossref","first-page":"589","DOI":"10.1109\/TSA.2005.855836","article-title":"A probabilistic framework for dialog simulation and optimal strategy learning","volume":"14","author":"Pietquin","year":"2006","journal-title":"IEEE Transactions on Audio, Speech, and Language Processing"},{"issue":"1","key":"2026061709272029900_bib42","doi-asserted-by":"crossref","first-page":"59","DOI":"10.1017\/S0269888912000343","article-title":"A survey on metrics for the evaluation of user simulations","volume":"28","author":"Pietquin","year":"2013","journal-title":"The Knowledge Engineering Review"},{"key":"2026061709272029900_bib43","volume-title":"Apigen-mt: Agentic pipeline for multi-turn data generation via simulated agent-human interplay","author":"Prabhakar","year":"2025"},{"key":"2026061709272029900_bib44","volume-title":"Toolrl: Reward is all tool learning needs","author":"Qian","year":"2025"},{"key":"2026061709272029900_bib45","volume-title":"Qwen2.5 technical report","author":"Qwen","year":"2025"},{"key":"2026061709272029900_bib46","doi-asserted-by":"crossref","first-page":"8689","DOI":"10.1609\/aaai.v34i05.6394","article-title":"Towards scalable multi-domain conversational agents: The schema-guided dialogue dataset","volume":"34","author":"Rastogi","year":"2020","journal-title":"Proceedings of the AAAI conference on artificial intelligence"},{"key":"2026061709272029900_bib47","first-page":"149","article-title":"Agenda-based user simulation for bootstrapping a POMDP dialogue system","volume-title":"Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Companion Volume, Short Papers","author":"Schatzmann","year":"2007"},{"issue":"2","key":"2026061709272029900_bib48","doi-asserted-by":"crossref","first-page":"97","DOI":"10.1017\/S0269888906000944","article-title":"A survey of statistical user simulation techniques for reinforcement-learning of dialogue management strategies","volume":"21","author":"Schatzmann","year":"2006","journal-title":"The knowledge engineering review"},{"issue":"4","key":"2026061709272029900_bib49","doi-asserted-by":"crossref","first-page":"733","DOI":"10.1109\/TASL.2008.2012071","article-title":"The hidden agenda user simulation model","volume":"17","author":"Schatzmann","year":"2009","journal-title":"IEEE transactions on audio, speech, and language processing"},{"key":"2026061709272029900_bib50","article-title":"Automatic learning of dialogue strategy using dialogue simulation and reinforcement learning","volume":"2","author":"Scheffler","year":"2002","journal-title":"Proceedings of HLT"},{"key":"2026061709272029900_bib51","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2024.scichat-1.3","article-title":"Reliable llm-based user simulator for task-oriented dialogue systems","volume-title":"arXiv preprint arXiv:2402.13374","author":"Sekuli\u0107","year":"2024"},{"key":"2026061709272029900_bib52","first-page":"41","article-title":"Bootstrapping a neural conversational agent with dialogue self-play, crowdsourcing and on-line reinforcement learning","volume-title":"Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 3 (Industry Papers)","author":"Shah","year":"2018"},{"key":"2026061709272029900_bib53","volume-title":"Deepseekmath: Pushing the limits of mathematical reasoning in open language models","author":"Shao","year":"2024"},{"key":"2026061709272029900_bib54","volume-title":"Tooldial: Multi-turn dialogue generation method for tool-augmented language models","author":"Shim","year":"2025"},{"key":"2026061709272029900_bib55","article-title":"Welcome to the era of experience","volume-title":"Google AI","author":"Silver","year":"2025"},{"key":"2026061709272029900_bib56","first-page":"9460","article-title":"Defining and characterizing reward gaming","volume":"35","author":"Skalse","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"issue":"1","key":"2026061709272029900_bib57","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1145\/3533020","article-title":"Metaphorical user simulators for evaluating task-oriented dialogue systems","volume":"42","author":"Sun","year":"2023","journal-title":"ACM Transactions on Information Systems"},{"key":"2026061709272029900_bib58","doi-asserted-by":"crossref","first-page":"5624","DOI":"10.18653\/v1\/P19-1565","article-title":"Target-guided open-domain conversation","volume-title":"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics","author":"Tang","year":"2019"},{"key":"2026061709272029900_bib59","volume-title":"Gemma 3 technical report","author":"Team","year":"2025"},{"issue":"1","key":"2026061709272029900_bib60","doi-asserted-by":"crossref","first-page":"33","DOI":"10.1145\/151271.151274","article-title":"An accurate computation of the hypergeometric distribution function","volume":"19","author":"Wu","year":"1993","journal-title":"ACM Transactions on Mathematical Software (TOMS)"},{"key":"2026061709272029900_bib61","doi-asserted-by":"crossref","first-page":"3794","DOI":"10.18653\/v1\/P19-1369","article-title":"Proactive human-machine conversation with explicit conversation goal","volume-title":"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics","author":"Wu","year":"2019"},{"key":"2026061709272029900_bib62","doi-asserted-by":"crossref","first-page":"10883","DOI":"10.18653\/v1\/2024.findings-emnlp.638","article-title":"FlowBench: Revisiting and benchmarking workflow-guided planning for LLM-based agents","volume-title":"Findings of the Association for Computational Linguistics: EMNLP 2024","author":"Xiao","year":"2024"},{"key":"2026061709272029900_bib63","volume-title":"\u03c4-bench: A benchmark for tool-agent-user interaction in real-world domains","author":"Yao","year":"2024"},{"key":"2026061709272029900_bib64","volume-title":"Multiwoz 2.4: A multi-domain task-oriented dialogue dataset with essential annotation corrections to improve state tracking evaluation","author":"Ye","year":"2022"},{"key":"2026061709272029900_bib65","article-title":"Bertscore: Evaluating text generation with bert","volume-title":"ArXiv","author":"Zhang","year":"2019"},{"issue":"10","key":"2026061709272029900_bib66","doi-asserted-by":"crossref","first-page":"2011","DOI":"10.1007\/s11431-020-1692-3","article-title":"Recent advances and challenges in task-oriented dialog systems","volume":"63","author":"Zhang","year":"2020","journal-title":"Science China Technological Sciences"},{"key":"2026061709272029900_bib67","doi-asserted-by":"crossref","first-page":"654","DOI":"10.18653\/v1\/P17-1061","article-title":"Learning discourse-level diversity for neural dialog models using conditional variational autoencoders","volume-title":"Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","author":"Zhao","year":"2017"},{"key":"2026061709272029900_bib68","doi-asserted-by":"crossref","first-page":"21108","DOI":"10.18653\/v1\/2024.emnlp-main.1175","article-title":"Thoughts to target: Enhance planning for target-driven conversation","volume-title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","author":"Zheng","year":"2024"}],"container-title":["Transactions of the Association for Computational Linguistics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/direct.mit.edu\/tacl\/article-pdf\/doi\/10.1162\/TACL.a.687\/2606890\/tacl.a.687.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/direct.mit.edu\/tacl\/article-pdf\/doi\/10.1162\/TACL.a.687\/2606890\/tacl.a.687.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,17]],"date-time":"2026-06-17T13:27:39Z","timestamp":1781702859000},"score":1,"resource":{"primary":{"URL":"https:\/\/direct.mit.edu\/tacl\/article\/doi\/10.1162\/TACL.a.687\/137192\/Goal-Alignment-in-LLM-Based-User-Simulators-for"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":68,"URL":"https:\/\/doi.org\/10.1162\/tacl.a.687","relation":{},"ISSN":["2307-387X"],"issn-type":[{"value":"2307-387X","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2026]]},"published":{"date-parts":[[2026]]}}}