Using LLMs To Improve Formulaic Readability Measures: A RoBERTa Approach