Fair point. They can't, for example, randomly generate language semantics (how the hell would that work? :)
All in all, though, this paper goes much further in trying to derive language usability than any I've seen. And it does seem disturbing that randomly selected characters (even for a subset of the language) should fare as well as Perl's hand-picked syntax.
While this is interesting, presenting it as a “randomly-generated language” is extremely misleading.