Thought-Like-Pro: Enhancing Reasoning of Large Language Models through Self-Bootstrapped Prolog-based Chain-of-Thought

Thought-Like-Pro improves LLM reasoning by using imitation learning to replicate Chain-of-Thought trajectories verified by a symbolic Prolog logic engine. This self-bootstrapped framework translates symbolic logic into natural language reasoning steps, enabling models to formulate and generalize rules across diverse tasks through prompt-guided training.