Reproduce “Logic-Parametric Neuro-Symbolic NLI: Controlling Logical Formalisms for Verifiable LLM Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ali Farjami, Luca Redondi, Marco Valentino
Farjami et al. developed a logic-parametric neuro-symbolic framework for Natural Language Inference that allows controlling the underlying logical formalisms, demonstrating that specific logics like Modal KD achieve higher explanation success rates (77.67% for GPT-4o) and greater efficiency in normative reasoning compared to a fixed First-Order Logic approach.
Key insights
- The choice of logical formalism significantly influences LLM-driven neuro-symbolic reasoning's success rate, refinement efficiency, and computational cost, challenging the assumption of a universal fixed logic.
- Logic-internal strategies, where normative concepts are natively embedded in the logic's structure, consistently outperform logic-external (FOL) approaches for normative reasoning tasks.
- The optimal logical formalism is domain-dependent; FOL excels in commonsense reasoning, while modal and deontic logics prove superior in ethical reasoning and handling modalities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.