I checked your web citation. It seems to be weirdly wrong. Although the web article is entitled “Generalization and Symbolic Processing in Neural Networks” it is based on a single PNAS paper entitled

“Prefrontal cortex and flexible cognitive control: Rules without symbols”
http://www.pnas.org/content/102/20/7338.full.pdf+html

The PNAS paper is a neural net model of rule generalization, based on prefrontal cortex.

I do have some problems though:

- It isn’t apparent to me from the article how to make the distinction between estimates (that likely most animals can do) and counting (which again, likely most animals can do, see martin’s refs). The next to last paragraph touches on this but seems to avoid acknowledging the “symbol” of estimates. That can, and likely should, be elaborated.

- I also don’t like the other overloading of the meaning of symbols as expressed in “lacking our symbolic brains”. It is very likely that all brains, at least with a prefrontal cortex but more generally probably with analogous executive function areas, works in a symbolic manner. Models shows that a prefrontal cortex can self-organize to use symbolic thinking. This also predicts that we, as opposed to simple neural networks, can learn categorizing without over-training, which is also observed.

[Incidentally, this answered johnk question "How do neurons do symbols" already in 2005.]

At a guess we are discussing abstract symbolic thinking, which may or may not emerge from the underlying symbolic processing.

If we (and elephants and so on) can model “others” and so “selves” to become self aware, there isn’t such a great leap to abstract general characteristics or even the models themselves into symbols. It likely takes a “mental” effort _not_ to generalize at that stage.

Similarly, the estimate/number route suggested could have been at play. But IMHO it seems less likely than the other way round; first symbols, later specific applications and exactness.

I’ve been asking myself questions readers of this column might want to discuss:

Are symbolic representations a special type of representation?
How do neurons do symbols?
Are simple examples of symbolic representations present in other types of representations?

I think a bee could do it

http://www.springerlink.com/content/e186k3381712731w/fulltext.pdf

or maybe a mosquitofish ?

http://www.springerlink.com/content/r8n475m318240045/fulltext.pdf

even a mongoose lemur . . .

http://www.springerlink.com/content/8xwtpld4f1m0553a/fulltext.pdf

or a horse . . .

http://www.springerlink.com/content/tx26u02831ph3164/fulltext.pdf

or a parrot . . .

http://www.springerlink.com/content/hr658n736238x774/fulltext.pdf