Autoformalization of English-Described Mathematical Text

Dongchen Jiang, houyiming

Keywords: Autoformalization, Formal Mathematics, Predicate Logic-based Formula, Automatic Verification of Mathematical Proof

Abstract: Autoformalization aims to convert natural language text into machine-verifiable formalizations, offering significant potential for advancing mathematical automation. However, the lack of high-quality parallel data between natural language and formal language in mathematics severely hinders the development of autoformalization. This paper proposes a linguistic-based method which converts English-described mathematical text into formal formulas. This method can automatically construct large-scale, high-quality formulas without model training. To evaluate the method, an experiment on the dataset including theorems and examples from Calculus is conducted, and the autoformalization result shows the method achieves an accuracy of 77.90\%. And a case study of automatic verification of English-described mathematical proof is provided to demonstrate the usability of our method.

Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)

Submission Number: 7594