Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Tong, Asai, Akari, Zettlemoyer, Luke, Hajishirzi, Hannaneh, Brahman, Faeze
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911222015721472
author Chen, Tong
Asai, Akari
Zettlemoyer, Luke
Hajishirzi, Hannaneh
Brahman, Faeze
author_facet Chen, Tong
Asai, Akari
Zettlemoyer, Luke
Hajishirzi, Hannaneh
Brahman, Faeze
contents Language models often generate factually incorrect information unsupported by their training data, a phenomenon known as extrinsic hallucination. Existing mitigation approaches often degrade performance on open-ended generation and downstream tasks, limiting their practical utility. We propose an online reinforcement learning method using a novel binary retrieval-augmented reward (RAR) to address this tradeoff. Unlike continuous reward schemes, our approach assigns a reward of one only when the model's output is entirely factually correct, and zero otherwise. We evaluate our method on Qwen3 reasoning models across diverse tasks. For open-ended generation, binary RAR achieves a 39.3% reduction in hallucination rates, substantially outperforming both supervised training and continuous-reward RL baselines. In short-form question answering, the model learns calibrated abstention, strategically outputting "I don't know" when faced with insufficient parametric knowledge. This yields 44.4% and 21.7% fewer incorrect answers on PopQA and GPQA, respectively. Crucially, these factuality gains come without performance degradation on instruction following, math, or code, whereas continuous-reward RL, despite improving factuality, induces quality regressions.
format Preprint
id arxiv_https___arxiv_org_abs_2510_17733
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations
Chen, Tong
Asai, Akari
Zettlemoyer, Luke
Hajishirzi, Hannaneh
Brahman, Faeze
Computation and Language
Machine Learning
Language models often generate factually incorrect information unsupported by their training data, a phenomenon known as extrinsic hallucination. Existing mitigation approaches often degrade performance on open-ended generation and downstream tasks, limiting their practical utility. We propose an online reinforcement learning method using a novel binary retrieval-augmented reward (RAR) to address this tradeoff. Unlike continuous reward schemes, our approach assigns a reward of one only when the model's output is entirely factually correct, and zero otherwise. We evaluate our method on Qwen3 reasoning models across diverse tasks. For open-ended generation, binary RAR achieves a 39.3% reduction in hallucination rates, substantially outperforming both supervised training and continuous-reward RL baselines. In short-form question answering, the model learns calibrated abstention, strategically outputting "I don't know" when faced with insufficient parametric knowledge. This yields 44.4% and 21.7% fewer incorrect answers on PopQA and GPQA, respectively. Crucially, these factuality gains come without performance degradation on instruction following, math, or code, whereas continuous-reward RL, despite improving factuality, induces quality regressions.
title Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2510.17733