Crafting Interpretable Embeddings for Language Neuroscience by Asking LLMs Questions