Verification and Refinement of Natural Language Explanations through LLM-Symbolic Theorem Proving