Language models improve dialogue by tracking conversation state reliably
Lost with a Map: Conversational State and Behavioral Reliability in Language Models
Computation and Language
Summary
Keeping track of what people say and want is hard for AI language systems in conversations. The paper’s authors studied how popular language models remember and use information during multi-turn chats, finding that while models can identify topics and requests clearly, remembering precise details from the user is less reliable. They found that models sometimes keep old and new information at the same time, which affects their responses. To fix this, the authors designed a simple method that edits the model’s replies based on what it currently knows, leading to much better accuracy and task success.
What this means in practice
- •For conversational ai developers: Make dialogue systems more accurate by improving how they track and apply user information during multi-turn tasks using a state-action controller.
- •For customer support platform teams: Increase reliability in automated support chatbots by fixing how models update and use constraints from users across conversation turns.
Authors
Atahan Dokme, Larry Heck
Abstract
Task-oriented dialogue requires maintaining and updating information across turns, yet language models expose no explicit belief-state object. We study how conversational state is represented, updated, and used inside eight instruction-tuned language models from four families on MultiWOZ and SGD. Structure and values separate: which domains, slots, and requests are active is linearly readable just before the model acts, whereas exact values are far more readable where the user stated them. After a user changes a value, both values remain accessible at their mentions, and causal interventions show that both continue to influence the model's action. In natural closed-loop interaction, query failures separate into cases of weak structural support, incorrect value resolution, and failure to deploy otherwise-supported constraints, with targeted interventions producing systematically different repair behavior across these cases. These findings motivate a state-action controller that starts from the base model action and selectively edits it using structural readouts, without requiring a complete predicted belief state as an intermediate representation. On held-out MultiWOZ interaction across five models, it raises the base model exact-query accuracy from .318 to .621 and task success from .272 to .371 at negligible added cost. Overall, reliable interaction requires not only retaining conversational information, but resolving which available constraints currently apply and ensuring that they govern action.