Logic-Regularized Verifier Elicits Reasoning from LLMs
Xinyu Wang, Changzhi Sun, Lian Cheng, Yuanbin Wu, Dell Zhang, Xiaoling Wang, Xuelong Li
Abstract
Verifiers are crucial components for enhancing modern LLMs’ reasoning capability. Typical verifiers require resource-intensive supervised dataset construction, which is costly and faces limitations in data diversity. In this paper, we propose LOVER, an unsupervised verifier regularized by logical rules. LOVER treats the verifier as a binary latent variable, utilizing internal activations and enforcing three logical constraints on multiple reasoning paths: negation consistency, intra-group consistency, and inter-group consistency (grouped by the final answer). By incorporating logical rules as priors, LOVER can leverage unlabeled examples and is directly compatible with any off-the-shelf LLMs. Experiments on 10 datasets demonstrate that LOVER significantly outperforms unsupervised baselines, achieving performance comparable to the supervised verifier (reaching its 95% level on average).
BibTeX
@inproceedings{wang-etal-2025-logic,
title = "Logic-Regularized Verifier Elicits Reasoning from {LLM}s",
author = "Wang, Xinyu and
Sun, Changzhi and
Cheng, Lian and
Wu, Yuanbin and
Zhang, Dell and
Wang, Xiaoling and
Li, Xuelong",
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.1567/",
doi = "10.18653/v1/2025.acl-long.1567",
pages = "32617--32630",
ISBN = "979-8-89176-251-0"
}