LoGU: Long-form Generation with Uncertainty Expressions

TL;DR

This paper introduces LoGU, a method for improving long-form generation by enabling models to better express and calibrate uncertainty, thereby reducing hallucinations and increasing factual accuracy in responses.

Contribution

The paper presents a novel framework and training pipeline for long-form generation with uncertainty expression, addressing challenges of suppression and misalignment.

Findings

Significantly reduces hallucinations in long-form responses

Improves factual accuracy and response comprehensiveness

Enhances uncertainty calibration in language models

Abstract

While Large Language Models (LLMs) demonstrate impressive capabilities, they still struggle with generating factually incorrect content (i.e., hallucinations). A promising approach to mitigate this issue is enabling models to express uncertainty when unsure. Previous research on uncertainty modeling has primarily focused on short-form QA, but realworld applications often require much longer responses. In this work, we introduce the task of Long-form Generation with Uncertainty(LoGU). We identify two key challenges: Uncertainty Suppression, where models hesitate to express uncertainty, and Uncertainty Misalignment, where models convey uncertainty inaccurately. To tackle these challenges, we propose a refinement-based data collection framework and a two-stage training pipeline. Our framework adopts a divide-and-conquer strategy, refining uncertainty based on atomic claims. The collected data are then used in training through supervised fine-tuning (SFT) and direct preference optimization (DPO) to enhance uncertainty expression. Extensive experiments on three long-form instruction following datasets show that our method significantly improves accuracy, reduces hallucinations, and maintains the comprehensiveness of responses.