Download PDFOpen PDF in browser

Automated Theorem Provers Help Improve Large Language Model Reasoning

19 pagesPublished: May 26, 2024

Abstract

In this paper we demonstrate how logic programming systems and Automated first- order logic Theorem Provers (ATPs) can improve the accuracy of Large Language Models (LLMs) for logical reasoning tasks where the baseline performance is given by direct LLM solutions. We first evaluate LLM reasoning on steamroller problems using the PRON- TOQA benchmark. We show how accuracy can be improved with a neuro-symbolic ar- chitecture where the LLM acts solely as a front-end for translating a given problem into a formal logic language and an automated reasoning engine is called for solving it. How- ever, this approach critically hinges on the correctness of the LLM translation. To assess this translation correctness, we secondly define a framework of syntactic and semantic er- ror categories. We implemented the framework and used it to identify errors that LLMs make in the benchmark domain. Based on these findings, we thirdly extended our method with capabilities for automatically correcting syntactic and semantic errors. For semantic error correction we integrate first-order logic ATPs, which is our main and novel contribu- tion. We demonstrate that this approach reduces semantic errors significantly and further increases the accurracy of LLM logical reasoning.

Keyphrases: automated theorem proving, first-order logic, large language models, logic programming, natural language, Reasoning, Steamroller Problems

In: Nikolaj Bjorner, Marijn Heule and Andrei Voronkov (editors). Proceedings of 25th Conference on Logic for Programming, Artificial Intelligence and Reasoning, vol 100, pages 51--69

Links:
BibTeX entry
@inproceedings{LPAR2024:Automated_Theorem_Provers_Help,
  author    = {Lachlan McGinness and Peter Baumgartner},
  title     = {Automated Theorem Provers Help Improve Large Language Model Reasoning},
  booktitle = {Proceedings of 25th Conference on Logic for Programming, Artificial Intelligence and Reasoning},
  editor    = {Nikolaj Bj\{\textbackslash{}o\}rner and Marijn Heule and Andrei Voronkov},
  series    = {EPiC Series in Computing},
  volume    = {100},
  pages     = {51--69},
  year      = {2024},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {https://easychair.org/publications/paper/vzpW},
  doi       = {10.29007/2n9m}}
Download PDFOpen PDF in browser