We don’t have the ability to do that level of analysis of natural language text mathematically.

If we did, we probably wouldn’t need LLMs in the first place, i.e. we could just generate text using explicitly programmed algorithms.