![[2602.21044] LogicGraph : Benchmarking Multi-Path Logical Reasoning via Neuro-Symbolic Generation and Verification](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.21044] LogicGraph : Benchmarking Multi-Path Logical Reasoning via Neuro-Symbolic Generation and Verification
Summary
LogicGraph introduces a benchmark for evaluating multi-path logical reasoning in large language models, highlighting their limitations in exploring diverse reasoning paths.
Why It Matters
This research addresses a critical gap in the evaluation of AI models, which often focus on convergent reasoning. By emphasizing multi-path logical reasoning, it encourages the development of more robust AI systems capable of handling complex real-world problems.
Key Takeaways
- LogicGraph is the first benchmark for multi-path logical reasoning.
- Current models often fail to explore alternative reasoning paths, leading to performance gaps.
- The proposed evaluation framework assesses both convergent and divergent reasoning capabilities.
Computer Science > Artificial Intelligence arXiv:2602.21044 (cs) [Submitted on 24 Feb 2026] Title:LogicGraph : Benchmarking Multi-Path Logical Reasoning via Neuro-Symbolic Generation and Verification Authors:Yanrui Wu, Lingling Zhang, Xinyu Zhang, Jiayu Chang, Pengyu Li, Xu Jiang, Jingtao Hu, Jun Liu View a PDF of the paper titled LogicGraph : Benchmarking Multi-Path Logical Reasoning via Neuro-Symbolic Generation and Verification, by Yanrui Wu and 7 other authors View PDF HTML (experimental) Abstract:Evaluations of large language models (LLMs) primarily emphasize convergent logical reasoning, where success is defined by producing a single correct proof. However, many real-world reasoning problems admit multiple valid derivations, requiring models to explore diverse logical paths rather than committing to one route. To address this limitation, we introduce LogicGraph, the first benchmark aimed to systematically evaluate multi-path logical reasoning, constructed via a neuro-symbolic framework that leverages backward logic generation and semantic instantiation. This pipeline yields solver-verified reasoning problems formalized by high-depth multi-path reasoning and inherent logical distractions, where each instance is associated with an exhaustive set of minimal proofs. We further propose a reference-free evaluation framework to rigorously assess model performance in both convergent and divergent regimes. Experiments on state-of-the-art language models reveal a common limita...