Investigating Symbolic Capabilities of Large Language Models