A cheating pseudo-entry in NaNoGenMo 2015.
This code mashes up two source texts. The first text is used for structure, while the second provides the vocabulary. Word replacements are chosen to be semantically close (using word2vec and part-of-speech identification) as well as similar in frequency between the texts.
This was mostly written before the start of November, so doesn't really count as a NaNoGenMo entry, but it's kind of fun anyway.