Yeah, he kind of covers that in section 5. No reason for picking his model (other than simplicity), it can't predict new characters, and there's very little data to go off of.
Sure, it's mental masturbation. But developing little toy models like this is still interesting and helps you understand limitations and keep up your skills without working on big, complicated, boring real statistical models. This kind of thing is worth doing for the same reason masturbating is worth doing--it's fun. As long as you don't take it too seriously and start confusing it with actual sex.
One good reason why computers can do more work than people is that they never have to stop and answer the phone.