Structure-Conditional Minimum Bayes Risk Decoding

2025-10-23

AI Summary

Traditional minimum Bayes risk (MBR) decoding for open-ended generation tasks is insensitive to latent structural properties—such as dialogue acts, sentiment, or response schemata—often selecting generic yet structurally incoherent responses. To address this, we propose a structure-enhanced MBR decoding framework. Methodologically, we design three lightweight, structure-aware utility functions; introduce a grouped contrastive strategy and structure-sensitive similarity metrics; and define two novel evaluation metrics to quantify the plausibility and consistency of candidate responses in latent structural space. Experiments demonstrate that our approach achieves up to a 13.7-percentage-point improvement in win rate on both AlpacaEval and MT-Bench, significantly outperforming similarity-based baselines. This work constitutes the first systematic integration of structural inductive bias into the MBR decoding paradigm, advancing structural fidelity in generative language modeling.

Technology Category

• Machine Learning: Structured Learning
• Natural Language Processing: Generation
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

Minimum Bayes Risk (MBR) decoding has seen renewed interest as an alternative to traditional generation strategies. While MBR has proven effective in machine translation, where the variability of a language model's outcome space is naturally constrained, it may face challenges in more open-ended tasks such as dialogue or instruction-following. We hypothesise that in such settings, applying MBR with standard similarity-based utility functions may result in selecting responses that are broadly representative of the model's distribution, yet sub-optimal with respect to any particular grouping of generations that share an underlying latent structure. In this work, we introduce three lightweight adaptations to the utility function, designed to make MBR more sensitive to structural variability in the outcome space. To test our hypothesis, we curate a dataset capturing three representative types of latent structure: dialogue act, emotion, and response structure (e.g., a sentence, a paragraph, or a list). We further propose two metrics to evaluate the structural optimality of MBR. Our analysis demonstrates that common similarity-based utility functions fall short by these metrics. In contrast, our proposed adaptations considerably improve structural optimality. Finally, we evaluate our approaches on real-world instruction-following benchmarks, AlpacaEval and MT-Bench, and show that increased structural sensitivity improves generation quality by up to 13.7 percentage points in win rate.

Problem

• MBR decoding struggles with structural variability in open-ended tasks
• Standard utility functions select sub-optimal responses ignoring latent structure
• Proposed adaptations improve structural sensitivity and generation quality

Innovation

• Adapting utility functions for structural sensitivity
• Introducing metrics to evaluate structural optimality
• Improving generation quality via structural adaptations