I'm also curious. I wrote a Markov text generator in Lisp a few months ago (https://github.com/jl2/markov https://github.com/jl2/markov), and had fun generatin

I'm also curious. I wrote a Markov text generator in Lisp a few months ago (https://github.com/jl2/markov https://github.com/jl2/markov), and had fun generating text using some old sci-fi books from Project Gutenberg and some documentation we use at work (Fiber Channel and ISCSI specs), but the sentences mine generated weren't as coherent as these. I'm curious if this is generating thousands of sentences and only posting the best, if it used the basic algorithm and a ton of sample text, or if it's using a more advanced algorithm. I was planning on making mine use parts of speech to help avoid some of the nonsense sentences, but never got around to it. And it would be harder to find source text to feed into it.