Reproduce “Neuro-Symbolic Integration Brings Causal and Reliable Reasoning Proofs” 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

Sen Yang, Xin Li, Leyang Cui, Lidong Bing, Wai Lam

CARING translates a reasoning problem into Prolog facts, rules, and a query, then uses a Prolog interpreter to perform inference and expose its execution trace as a dependency-linked proof. The LLM handles translation, not inference. On a 600-instance ProofWriter subset, it reaches 96.50% answer accuracy and 81.02% proof-graph similarity, versus 52.00% and 15.76% for CoT, conditional on correct symbolic encoding.

Key insights

  • By separating LLM translation from symbolic reasoning, the framework leverages LLMs for their natural language understanding strengths while relying on deterministic solvers for infallible logical inference, ensuring causality and reliability in proofs.
  • Customized meta-interpreters can expose the internal, step-by-step search logs of symbolic solvers, transforming these implicit processes into explicit, verifiable proof trees.
  • This integration overcomes the accuracy-interpretability trade-off, demonstrating that highly accurate answers can be accompanied by transparent and trustworthy reasoning explanations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.