Markov chains are cool and text generation is fun for experiments. I would definitely recommend looking at deep learning language models, for a more fruitful investment of time to improve further. (courses on http://www.fast.ai/ are a good starting place)