Towards LLM-generated explanations for Component-based Knowledge Graph Question Answering Systems