Hinton (now of Google Brain) and DS Touretzsky (yeah, the anti-scientology guy) published a thing in 1985 claiming inferential symbol production a la lisp from connectionist (neural) models (http://www.cs.toronto.edu/~hinton/absps/symbolsIJCAI.pdf). It only had the disadvantage of being a researchy piece of crap.

Jmm, thanks for the fact, I would never imagine such a paper, but why are you saying that it's a piece of crap?

I enjoyed reading this.