It would actually be cool to try to use NLP/LLMs to quantify eloquence