Hidden Markov Model or Markov Chain?

I suspect that iocaine is not using a Hidden Markov Model (which is for interpreting/decoding observations; the best example is speech-to-text) but just a regular old Markov chain.

The source code excerpt it spit out includes "Bigram", which implies 2-grams. Those don't need a lot of training text to be effective, but the output is frequently nonsense.

Ironically, scrapers could probably use a Hidden Markov Model to detect the output of iocaine (or any other Markov text generator).

For anyone with coding skills -- if you have never done so, I encourage you to try writing your own Markov generator -- it's not very difficult, and the output can be quite amusing.

There's a blog that posts the (manually curated) output of a Markov chain trained on the King James Bible and a set of books on programming (mostly Lisp). It uses 4-grams, so there's very little nonsense, and it has some real gems: https://www.tumblr.com/kingjamesprogramming

(My favorite one that I generated locally was "the wicked shall become effective devices for computing in any direction.")