@inproceedings{liu-etal-2026-pretrainrl,
title = "{P}retrain{RL}: Alleviating Factuality Hallucination of Large Language Models at the Beginning",
author = "Liu, Langming and
Lv, Kangtao and
Chen, Haibin and
Zhang, Weidong and
Wang, Yejing and
Liu, Shilei and
Tong, Xin and
Yuan, Yujin and
Wang, Yongwei and
Su, Wenbo and
Zheng, Bo",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.910/",
pages = "18298--18310",
ISBN = "979-8-89176-395-1",
abstract = "Large language models (LLMs), despite their powerful capabilities, suffer from factual hallucinations where they generate verifiable falsehoods. We identify a root of this issue: the imbalanced data distribution in the pretraining corpus, which leads to a state of ``low-probability truth'' and ``high-probability falsehood''. Recent approaches, such as teaching models to say ``I don{'}t know'' or post-hoc knowledge editing, either evade the problem or face catastrophic forgetting. To address this issue from its root, we propose PretrainRL, a novel framework that integrates reinforcement learning into the pretraining phase to consolidate factual knowledge. The core principle of PretrainRL is ``debiasing then learning.'' It actively reshapes the model{'}s probability distribution by down-weighting high-probability falsehoods, thereby making ``room'' for low-probability truths to be learned effectively. To enable this, we design an efficient negative sampling strategy to discover these high-probability falsehoods and introduce novel metrics to evaluate the model{'}s probabilistic state concerning factual knowledge. Extensive experiments on three public benchmarks demonstrate that PretrainRL significantly alleviates factual hallucinations and outperforms state-of-the-art methods."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2026-pretrainrl">
<titleInfo>
<title>PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Langming</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kangtao</namePart>
<namePart type="family">Lv</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haibin</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weidong</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yejing</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shilei</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xin</namePart>
<namePart type="family">Tong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yujin</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yongwei</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenbo</namePart>
<namePart type="family">Su</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bo</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Large language models (LLMs), despite their powerful capabilities, suffer from factual hallucinations where they generate verifiable falsehoods. We identify a root of this issue: the imbalanced data distribution in the pretraining corpus, which leads to a state of “low-probability truth” and “high-probability falsehood”. Recent approaches, such as teaching models to say “I don’t know” or post-hoc knowledge editing, either evade the problem or face catastrophic forgetting. To address this issue from its root, we propose PretrainRL, a novel framework that integrates reinforcement learning into the pretraining phase to consolidate factual knowledge. The core principle of PretrainRL is “debiasing then learning.” It actively reshapes the model’s probability distribution by down-weighting high-probability falsehoods, thereby making “room” for low-probability truths to be learned effectively. To enable this, we design an efficient negative sampling strategy to discover these high-probability falsehoods and introduce novel metrics to evaluate the model’s probabilistic state concerning factual knowledge. Extensive experiments on three public benchmarks demonstrate that PretrainRL significantly alleviates factual hallucinations and outperforms state-of-the-art methods.</abstract>
<identifier type="citekey">liu-etal-2026-pretrainrl</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.910/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>18298</start>
<end>18310</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning
%A Liu, Langming
%A Lv, Kangtao
%A Chen, Haibin
%A Zhang, Weidong
%A Wang, Yejing
%A Liu, Shilei
%A Tong, Xin
%A Yuan, Yujin
%A Wang, Yongwei
%A Su, Wenbo
%A Zheng, Bo
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F liu-etal-2026-pretrainrl
%X Large language models (LLMs), despite their powerful capabilities, suffer from factual hallucinations where they generate verifiable falsehoods. We identify a root of this issue: the imbalanced data distribution in the pretraining corpus, which leads to a state of “low-probability truth” and “high-probability falsehood”. Recent approaches, such as teaching models to say “I don’t know” or post-hoc knowledge editing, either evade the problem or face catastrophic forgetting. To address this issue from its root, we propose PretrainRL, a novel framework that integrates reinforcement learning into the pretraining phase to consolidate factual knowledge. The core principle of PretrainRL is “debiasing then learning.” It actively reshapes the model’s probability distribution by down-weighting high-probability falsehoods, thereby making “room” for low-probability truths to be learned effectively. To enable this, we design an efficient negative sampling strategy to discover these high-probability falsehoods and introduce novel metrics to evaluate the model’s probabilistic state concerning factual knowledge. Extensive experiments on three public benchmarks demonstrate that PretrainRL significantly alleviates factual hallucinations and outperforms state-of-the-art methods.
%U https://aclanthology.org/2026.findings-acl.910/
%P 18298-18310
Markdown (Informal)
[PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning](https://aclanthology.org/2026.findings-acl.910/) (Liu et al., Findings 2026)
ACL
- Langming Liu, Kangtao Lv, Haibin Chen, Weidong Zhang, Yejing Wang, Shilei Liu, Xin Tong, Yujin Yuan, Yongwei Wang, Wenbo Su, and Bo Zheng. 2026. PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning. In Findings of the Association for Computational Linguistics: ACL 2026, pages 18298–18310, San Diego, California, United States. Association for Computational Linguistics.