Score: 1

StepProof: Step-by-step verification of natural language mathematical proofs

Published: June 12, 2025 | arXiv ID: 2506.10558v2

By: Xiaolin Hu , Qinghua Zhou , Bogdan Grechuk and more

Potential Business Impact:

Lets computers check math steps written in English.

Business Areas:
Natural Language Processing Artificial Intelligence, Data and Analytics, Software

Interactive theorem provers (ITPs) are powerful tools for the formal verification of mathematical proofs down to the axiom level. However, their lack of a natural language interface remains a significant limitation. Recent advancements in large language models (LLMs) have enhanced the understanding of natural language inputs, paving the way for autoformalization - the process of translating natural language proofs into formal proofs that can be verified. Despite these advancements, existing autoformalization approaches are limited to verifying complete proofs and lack the capability for finer, sentence-level verification. To address this gap, we propose StepProof, a novel autoformalization method designed for granular, step-by-step verification. StepProof breaks down complete proofs into multiple verifiable subproofs, enabling sentence-level verification. Experimental results demonstrate that StepProof significantly improves proof success rates and efficiency compared to traditional methods. Additionally, we found that minor manual adjustments to the natural language proofs, tailoring them for step-level verification, further enhanced StepProof's performance in autoformalization.

Country of Origin
🇬🇧 United Kingdom

Repos / Data Links

Page Count
13 pages

Category
Computer Science:
Logic in Computer Science