Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BEDA: Belief Estimation as Probabilistic Constraints for Performing Strategic Dialogue Acts

Hengli Li, Zhaoxin Yu, Qi Shen, Chenxi Li, Mengmeng Wang, Tinglang Wu, Yipeng Kang, Yuxuan Wang, Song-Chun Zhu, Zixia Jia, Zilong Zheng

TL;DR

BEDA introduces a principled method to use belief estimation directly as probabilistic constraints during utterance generation for strategic dialogue acts. By formalizing Adversarial and Alignment acts and implementing them via a World Set, two belief estimators, and a constrained generator, BEDA bridges belief inference and action realization. Across CKBG, Mutual Friends, and CaSiNo, BEDA yields consistent gains in success rate and negotiation quality, while reducing inefficiencies and hallucinations associated with unconstrained generation. The framework highlights the practical impact of constraint-based belief use for robust, reliable strategic dialogue in diverse social rationales.

Abstract

Strategic dialogue requires agents to execute distinct dialogue acts, for which belief estimation is essential. While prior work often estimates beliefs accurately, it lacks a principled mechanism to use those beliefs during generation. We bridge this gap by first formalizing two core acts Adversarial and Alignment, and by operationalizing them via probabilistic constraints on what an agent may generate. We instantiate this idea in BEDA, a framework that consists of the world set, the belief estimator for belief estimation, and the conditional generator that selects acts and realizes utterances consistent with the inferred beliefs. Across three settings, Conditional Keeper Burglar (CKBG, adversarial), Mutual Friends (MF, cooperative), and CaSiNo (negotiation), BEDA consistently outperforms strong baselines: on CKBG it improves success rate by at least 5.0 points across backbones and by 20.6 points with GPT-4.1-nano; on Mutual Friends it achieves an average improvement of 9.3 points; and on CaSiNo it achieves the optimal deal relative to all baselines. These results indicate that casting belief estimation as constraints provides a simple, general mechanism for reliable strategic dialogue.

BEDA: Belief Estimation as Probabilistic Constraints for Performing Strategic Dialogue Acts

TL;DR

BEDA introduces a principled method to use belief estimation directly as probabilistic constraints during utterance generation for strategic dialogue acts. By formalizing Adversarial and Alignment acts and implementing them via a World Set, two belief estimators, and a constrained generator, BEDA bridges belief inference and action realization. Across CKBG, Mutual Friends, and CaSiNo, BEDA yields consistent gains in success rate and negotiation quality, while reducing inefficiencies and hallucinations associated with unconstrained generation. The framework highlights the practical impact of constraint-based belief use for robust, reliable strategic dialogue in diverse social rationales.

Abstract

Strategic dialogue requires agents to execute distinct dialogue acts, for which belief estimation is essential. While prior work often estimates beliefs accurately, it lacks a principled mechanism to use those beliefs during generation. We bridge this gap by first formalizing two core acts Adversarial and Alignment, and by operationalizing them via probabilistic constraints on what an agent may generate. We instantiate this idea in BEDA, a framework that consists of the world set, the belief estimator for belief estimation, and the conditional generator that selects acts and realizes utterances consistent with the inferred beliefs. Across three settings, Conditional Keeper Burglar (CKBG, adversarial), Mutual Friends (MF, cooperative), and CaSiNo (negotiation), BEDA consistently outperforms strong baselines: on CKBG it improves success rate by at least 5.0 points across backbones and by 20.6 points with GPT-4.1-nano; on Mutual Friends it achieves an average improvement of 9.3 points; and on CaSiNo it achieves the optimal deal relative to all baselines. These results indicate that casting belief estimation as constraints provides a simple, general mechanism for reliable strategic dialogue.
Paper Structure (65 sections, 9 equations, 2 figures, 13 tables)

This paper contains 65 sections, 9 equations, 2 figures, 13 tables.

Table of Contents

  1. Introduction
  2. BEDA
  3. Preliminaries: Logical Definition of Beliefs
  4. Belief Information as Probabilistic Constraints
  5. Statistical Modeling of Dialogue Acts
  6. Components of the BEDA framework
  7. World Set $\mathcal{P}(W)$
  8. Belief Estimator ($P_A$)
  9. Conditional Generator ($\pi$)
  10. Learning and Inference of BEDA
  11. Datasets
  12. ckbg
  13. mf
  14. CaSiNo
  15. Experiment
  16. ...and 50 more sections

Figures (2)

  • Figure 1: An overview of the BEDA framework is provided using the Keeper-Burglar Game as an example. The world set, depicted at the top, represents a structured understanding of the environment. At the beginning of each turn, the belief estimator processes the context and the world set to predict the beliefs. The beliefs are subsequently transmitted to the conditional generator, which incorporates a selection phase to identify the most appropriate belief (i.e., the events that satisfy the Adversarial Dialogue Act ) for generating the keeper's next utterance.
  • Figure 2: Experimental results on CaSiNo. Each experiment is performed for three times except for those with GPT-4 backbone.

Theorems & Definitions (6)

  • Definition 2.1: Information Partition
  • Definition 2.2: Event
  • Definition 2.3: Belief
  • Definition 2.4: Probabilistic Beliefs
  • Definition 2.5: Adversarial Dialogue Act
  • Definition 2.6: Alignment Dialogue Act