Learning Interpretable Style Embeddings via Prompting LLMs F i n d i n g s - E M N L P