@inproceedings{li-etal-2025-drift,
title = "Drift: Enhancing {LLM} Faithfulness in Rationale Generation via Dual-Reward Probabilistic Inference",
author = "Li, Jiazheng and
Yan, Hanqi and
He, Yulan",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.340/",
doi = "10.18653/v1/2025.acl-long.340",
pages = "6850--6866",
ISBN = "979-8-89176-251-0",
abstract = "As Large Language Models (LLMs) are increasingly applied to complex reasoning tasks, achieving both accurate task performance and faithful explanations becomes crucial. However, LLMs often generate unfaithful explanations, partly because they do not consistently adhere closely to the provided context. Existing approaches to this problem either rely on superficial calibration methods, such as decomposed Chain-of-Thought prompting, or require costly retraining to improve model faithfulness. In this work, we propose a probabilistic inference paradigm that leverages task-specific and lookahead rewards to ensure that LLM-generated rationales are more faithful to model decisions and align better with input context. These rewards are derived from a domain-specific proposal distribution, allowing for optimized sequential Monte Carlo approximations. Our evaluations across three different reasoning tasks show that this method, which allows for controllable generation during inference, improves both accuracy and faithfulness of LLMs. This method offers a promising path towards making LLMs more reliable for reasoning tasks without sacrificing performance."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2025-drift">
<titleInfo>
<title>Drift: Enhancing LLM Faithfulness in Rationale Generation via Dual-Reward Probabilistic Inference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiazheng</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanqi</namePart>
<namePart type="family">Yan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yulan</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>As Large Language Models (LLMs) are increasingly applied to complex reasoning tasks, achieving both accurate task performance and faithful explanations becomes crucial. However, LLMs often generate unfaithful explanations, partly because they do not consistently adhere closely to the provided context. Existing approaches to this problem either rely on superficial calibration methods, such as decomposed Chain-of-Thought prompting, or require costly retraining to improve model faithfulness. In this work, we propose a probabilistic inference paradigm that leverages task-specific and lookahead rewards to ensure that LLM-generated rationales are more faithful to model decisions and align better with input context. These rewards are derived from a domain-specific proposal distribution, allowing for optimized sequential Monte Carlo approximations. Our evaluations across three different reasoning tasks show that this method, which allows for controllable generation during inference, improves both accuracy and faithfulness of LLMs. This method offers a promising path towards making LLMs more reliable for reasoning tasks without sacrificing performance.</abstract>
<identifier type="citekey">li-etal-2025-drift</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.340</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.340/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>6850</start>
<end>6866</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Drift: Enhancing LLM Faithfulness in Rationale Generation via Dual-Reward Probabilistic Inference
%A Li, Jiazheng
%A Yan, Hanqi
%A He, Yulan
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F li-etal-2025-drift
%X As Large Language Models (LLMs) are increasingly applied to complex reasoning tasks, achieving both accurate task performance and faithful explanations becomes crucial. However, LLMs often generate unfaithful explanations, partly because they do not consistently adhere closely to the provided context. Existing approaches to this problem either rely on superficial calibration methods, such as decomposed Chain-of-Thought prompting, or require costly retraining to improve model faithfulness. In this work, we propose a probabilistic inference paradigm that leverages task-specific and lookahead rewards to ensure that LLM-generated rationales are more faithful to model decisions and align better with input context. These rewards are derived from a domain-specific proposal distribution, allowing for optimized sequential Monte Carlo approximations. Our evaluations across three different reasoning tasks show that this method, which allows for controllable generation during inference, improves both accuracy and faithfulness of LLMs. This method offers a promising path towards making LLMs more reliable for reasoning tasks without sacrificing performance.
%R 10.18653/v1/2025.acl-long.340
%U https://aclanthology.org/2025.acl-long.340/
%U https://doi.org/10.18653/v1/2025.acl-long.340
%P 6850-6866
Markdown (Informal)
[Drift: Enhancing LLM Faithfulness in Rationale Generation via Dual-Reward Probabilistic Inference](https://aclanthology.org/2025.acl-long.340/) (Li et al., ACL 2025)
ACL