Fine-Grained Uncertainty Quantification for Long-Form Language Model Outputs: A Comparative Study

TL;DR

This work presents a general framework for fine-grained uncertainty quantification in long-form outputs from large language systems. It defines a three-stage pipeline—response decomposition, unit-level confidence scoring, and response-level aggregation—and formalizes a taxonomy that unifies four scorer families (unit-response, matched-unit, unit-QA, and graph-based) with two aggregation strategies, including uncertainty-aware decoding. Across two long-form QA benchmarks and multiple LLMs, claim-response entailment consistently delivers strong detection performance, while claim-level scoring generally outperforms sentence-level scoring; uncertainty-aware decoding substantially improves factuality with manageable computational costs. The study offers practical guidance for component selection and provides an open-source toolkit, uqlm, to support reproducibility and further research in fine-grained uncertainty quantification for long-form generation.

Abstract

Uncertainty quantification has emerged as an effective approach to closed-book hallucination detection for LLMs, but existing methods are largely designed for short-form outputs and do not generalize well to long-form generation. We introduce a taxonomy for fine-grained uncertainty quantification in long-form LLM outputs that distinguishes methods by design choices at three stages: response decomposition, unit-level scoring, and response-level aggregation. We formalize several families of consistency-based black-box scorers, providing generalizations and extensions of existing methods. In our experiments across multiple LLMs and datasets, we find 1) claim-response entailment consistently performs better or on par with more complex claim-level scorers, 2) claim-level scoring generally yields better results than sentence-level scoring, and 3) uncertainty-aware decoding is highly effective for improving the factuality of long-form outputs. Our framework clarifies relationships between prior methods, enables apples-to-apples comparisons, and provides practical guidance for selecting components for fine-grained UQ.

Figures (14)

• Figure 1: Unit-Response Scoring Workflow
• Figure 2: Matched-Unit Scoring Workflow
• Figure 3: Unit-QA Scoring Workflow
• Figure 4: Graph-Based Scoring Workflow
• Figure 5: Claim-Level and Sentence-Level AUROC by LLM and Dataset (Higher is Better)