Towards Decoding Grammar Perception: Identifying Specialized Neurons in LLMs