> A Markov Chain trained by only a single article of text will very likely just regurgitate entire sentences straight from the source material.

I see this as a strength; try training an LLM on a 42KB text to see if it can produce a coherent output.