Non-Interactive Symbolic-Aided Chain-of-Thought for Logical Reasoning

TL;DR

This work tackles non-interactive logical reasoning with large language models by introducing Symbolic-Aided CoT, a prompting framework that embeds lightweight symbolic structures into few-shot prompts. The method uses rule tagging and reasoning operators to structure inference steps, maintain a knowledge base, and validate outcomes within a single inference pass, avoiding external solvers and multi-turn interactions. Empirical results on ProofWriter, ProntoQA, LogicalDeduction, and FOLIO show clear improvements over standard CoT on several open-source models, though FOLIO reveals some limitations when factual leakage benefits conventional prompting. The approach offers improved transparency and analyzability of reasoning, with strong potential for generalization and future enhancements such as latent-space refinements to further boost faithfulness and robustness in reasoning tasks.

Abstract

This work introduces Symbolic-Aided Chain-of-Thought (CoT), an improved approach to standard CoT, for logical reasoning in large language models (LLMs). The key idea is to integrate lightweight symbolic representations into few-shot prompts, structuring the inference steps with a consistent strategy to make reasoning patterns more explicit within a non-interactive reasoning process. By incorporating these symbolic structures, Symbolic-Aided CoT preserves the generalizability of standard prompting techniques while enhancing the transparency, interpretability, and analyzability of LLM logical reasoning. Extensive experiments on four well-known logical reasoning benchmarks -- ProofWriter, FOLIO, ProntoQA, and LogicalDeduction, which cover diverse reasoning tasks and scenarios -- demonstrate the effectiveness of the proposed approach, particularly in complex reasoning tasks that require navigating multiple constraints or rules. Notably, Symbolic-Aided CoT consistently improves LLMs' reasoning capabilities across various model sizes and significantly outperforms conventional CoT on three out of four datasets, ProofWriter, ProntoQA, and LogicalDeduction.

Figures (7)

• Figure 1: Comparison between standard CoT and Symbolic-Aided CoT for logical reasoning tasks.
• Figure 2: Left: Graphical model of inference rules. Right: Reasoning flows in the Symbolic‑Aided CoT demonstrations.
• Figure 3: Performance across different model sizes of Qwen-3 with three prompting techniques on the ProofWriter dataset.
• Figure 4: Performance across different model sizes of Qwen-3 with three prompting techniques on the LogicalDeduction dataset.
• Figure 5: Ablation study on the Symbolic-Aided CoT (SymbolA.CoT).