This software reads text and generates new ones based on statistics.
I also have an older version of it written in processing, quicker but less subtle.
Here is also an exemple xml file with it.

It was written in 2009.
It is now even more ridiculous compared to modern text-generation techniques.
It's based on n-grams but I'm not sure the technique can be called Markov chain.
As far as I know, most Markov chains take a given element into account to predict the next one, here the "chains" contain several elements to define what the next element should be.
While generating new text it will increment the length of the chains as much as possible but stop at some point to always make sure a given chain allows several possible alternative continuations.

Back then I created various pieces of weird art with it, including this nostalgic movie.