SPARQL queries play a crucial role in exploring knowledge graphs (KGs) and have been widely used in practice. However, understanding what questions are actually asked to KGs by exploring queries directly is a daunting task. In line with recent efforts to leverage Large Language Models (LLMs) for deriving underlying questions of SPARQL queries, we further investigate whether increasing the number of examples in prompting and Chain-of-Thought prompting can improve the performance. Additionally, we examine whether a fine-tuned LLM with one dataset can be used on another dataset to further improve performance.