Newest papers

Towards User Simulators That Stay On Task And Still Sound Natural

Mustafa Omer Gul, Philippe Laban, Tanya Goyal, Jennifer Neville · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31512 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Para avaliar e melhorar os assistentes de IA em tais configurações, precisamos de simuladores de usuários que seguem fielmente as intenções complexas do usuário enquanto exibem o comportamento de conversação realista. As abordagens existentes manifestam um compromisso: os assistentes orientados podem seguir as intenções complexas do usuário, mas carecem de fidelidade comportamental, enquanto os simuladores de usuários especializados melhor aproximam o comportamento de conversação, mas lutam para permanecer fiéis às intenções complexas. Neste artigo, propomos uma receita de aprendizagem reforçada (RL) que recompensa conjuntamente a adesão à intenção da tarefa e a naturalidade comportamental. Uma contribuição crucial da metodologia é a naturalização da operação com um discriminador leve treinado em um ciclo alternativo, adversário com o usuário. Lançamos o

Machine translation (Meta M2M100); the English original follows.

Abstract

Users increasingly rely on AI systems for extended, multi-turn tasks, yet evidence indicates that LLM-based system performance degrades over prolonged interaction. To evaluate and improve AI assistants in such settings, we need user simulators that faithfully follow complex user intents while exhibiting realistic conversational behavior. Existing approaches manifest a trade-off: prompted assistants can follow complex user intents but lack behavioral fidelity, while specialized user simulators better approximate conversational behavior but struggle to remain faithful to complex intents. In this paper, we propose a reinforcement learning (RL) recipe that jointly rewards task intent adherence and behavioral naturalness. A crucial contribution of the method is operationalizing naturalness with a lightweight discriminator trained in an alternating, adversarial loop with the user simulator. We release the resulting model, UserLM-2.0, which produces valid simulated conversations (i.e., meeting both task-faithfulness and naturalness criteria) 46.2% of the time, over three times the rate of the strongest prior user simulator (14.1%).

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Naturalness, Computer science, Human–computer interaction, Task (project management), Operationalization, User modeling

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange