Clarify, Abstain or Answer? Strategising in Conversation with Belief-Augmented Generation

2026-05-25

AI Summary

This work addresses the tendency of large language models to generate answers to ambiguous or underspecified inputs without acknowledging their own uncertainty, often lacking mechanisms for clarification or abstention. The authors propose Belief-Augmented Generation (BAG), a novel approach that explicitly leverages the model’s internal belief state to guide dialogue strategy. By constructing a belief distribution through multiple stochastic generations, BAG enables the model to autonomously decide whether to respond, request clarification, or abstain from answering. This method requires no human intervention and integrates prompt engineering with strategic reasoning. Evaluated across six mainstream language models, BAG significantly improves question-answering accuracy and yields strategy choices that more faithfully reflect the model’s actual uncertainty, outperforming baseline approaches relying solely on prompting.

Abstract

Large language models (LLMs) define a distribution over text, which can be viewed as a probabilistic representation of uncertainty: sampling K responses yields a belief state - responses a model deems plausible. Existing work exploits this representation for narrow tasks like either decoding or selective prediction, and often requires manual interventions, not controlling generation directly. We propose Belief-Augmented Generation (BAG): grounding LLMs in their own belief state via the prompt and letting them reason over these K samples to decide on a conversational strategy: answer, clarify, or abstain. In a multi-turn ambiguous QA setting, we find that LLMs by default rarely clarify or abstain, ignoring uncertainty about the input or facts. BAG improves QA accuracy across six models and yields strategy decisions more faithful to the belief state than prompt-only baselines. Disentangling when to clarify from when to abstain, however, remains challenging.

Problem

• uncertainty
• clarification
• abstention
• conversational strategy
• ambiguous QA

Innovation

• Belief-Augmented Generation
• uncertainty-aware dialogue
• conversational strategy
• large language models
• selective prediction