Pinker on Chomsky and Universal Grammar, the Modular Mind, granularity

Tyler Cowen has an interview with Steve Pinker that ranges widely. Here's what Pinker says about the Chomsky/UG debate:

It’s a moving target. Also, as you say, it was neither specified in a precise way nor field‑tested against a dataset of language variation, which I think is unfortunate in terms of ordinary scientific practice.

Linguistics is an eccentric field in some ways partly because it was so polarized by a charismatic figure [Noam Chomsky] and his opponents that it didn’t proceed in the ordinary direction of making the theory more precise, more testable.

With that caveat in mind, I think there is such a thing as, you can call it “universal grammar” in the following sense: that the child is biased to analyze the speech that he or she hears in particular ways. It does not simply record sentences verbatim.

That’s the memory half of the language system, but the algorithmic or computational or rule‑governed half tries to pull out combinatorial rules from the speech stream. There are certain kinds of rules and elements that a child is keyed to look for. That set of abilities would be what I would call (if I used the term) universal grammar.

There are commonalities across the world’s languages that come from the fact that language is created anew every generation by the minds of the children who construct it out of the data that they get from their parents and peers.

Modular mind:

Modules never quite seemed like the best metaphor. There is structure or specialization. I don’t think the mind is spam. I don’t think we just have a homogeneous neural network in the skull. There is some organization.

The problem with the module metaphor is some of them are snapping components with very limited channels of communication between them. I think that’s too strong, but I think it is reasonable to say that there are different faculties, to use an old‑fashioned word.

To choose a different metaphor, I think it may have been Chomsky who proposed that the mind is like a biological system made out of organs and tissues. When I was in high school I was taught, for example, that the blood was an organ.

Now the blood, of course, suffuses all of our tissues. You can’t draw a dotted line around it. It’s not like the rump roast and the flank steak at the supermarket cow display. Likewise the mind can have a specialization, and structure, and different components, without them literally being independent.

Granularity:

I also think that often — going back to finding a middle ground — the middle ground isn’t finding, say, the arithmetic mean between the two extremes. But rather it’s trying to go down a level of more finer-grain causal mechanisms underneath the phenomena. To state a position that may not look like either of the original extremes, because it’s more precise.

In the case of language, for example, I’ve always been bored by the idea of “is language innate or is it learned.” It’s neither. It’s not halfway in between because that doesn’t give you any insight either. Rather there is an innate structure that does the learning, because learning doesn’t happen by magic. There has to be something in place that does the learning.

Let’s characterize the nature of the learning mechanism in terms of its information-processing abilities. What is its computational architecture, as the computer scientists say?

Once you have that, that is the solution to the nature-nurture problem; namely, what’s innate is an ability to learn. Since any mechanism does some things well and some things not so well, that gives you insight as to what and how we learn.