Wow, I built a Markov text generator years ago, and made a PHP bot to crawl web pages to gather text for it, but that was before Twitter even existed. One suggestion might be to add a dictionary of english words and acronyms so that you can weed out nonsense words and other languages. It would probably have a big effect on the accuracy.

I wonder if the code is available anywhere? I'd love to experiment with it myself and see if I can improve on it some.

Edit: I found the code: http://github.com/OEP/markov