Did you checkout some of the ngrams gems, such as raingrams?

I did not. Thanks for the link :). I’ll take a look.

Raingrams looks great. Yours right? Nice work! I was thinking about trying some tri-grams to see how much better the prose appeared. I also wanted to anchor sentence endings that came from the text. Currently I just truncate when the desired sentence length is reached – unless it is one of the hard-coded preposition words. I spent a decent amount of time working with the random generation part of the code using the simple, core bi-gram model. I may think about using a library like yours at the core and wrapping the random-generation code around it in a future release.

@shaebdavis correct, tri-grams and higher will produce more accurate text. This requires having a greater corpus to train the models against, otherwise you’ll get verbatim training sentences. Also, there’s the problem of enforcing grammar and preventing run-on sentences. You can always enforce minimum / maximum sentence length with a recursive sentence building method. I tried adding some basic random generating methods in Raingrams.