Improving Rule-based Reasoning in LLMs via Neurosymbolic Representations