Representations and Linguistic Structure

Reading

Jerry A. Fodor: Why There Still Has to be a Language of Thought",in Lycan, Chapter 11, pp.282-299.

Lecture where this is primarily covered

Week 18: Representations and Imagery"

We have encountered the view that the brain/mind works by manipulating representations.

One variety of that view is that the representations that are manipulated in the brain/mind are quasi-linguistic.

I present here one argument for mantaining that such representations must be quasi-linguistic.

ARGUMENT FOR MAINTAINING THAT REPRESENTATIONS IN THE BRAIN/MIND MUST BE QUASI-LINGUISTIC.

The argument is that if they weren't, there would be no logical relations between them.

INFERENCES INVOLVING PROPOSITIONS DEPEND ON THE PROPOSTIONS BEING STRUCTURED: EXPLANATION

First consider ordinary inferences. They hold between things we say and write - between propositions.

The first point is that you can only conduct iferences or deductions if you have as your raw material propositions that are structured. Deductions depend on propositions having something but not everything in common.

I now try and explain this.

Consider the following propositions:

(1) Every animal is a living thing

(2) Every human being is an animal

(3) Every human being is a living thing

If this is a valid argument, what makes it valid?

It can only be valid if there are terms in common between the propositions.

'animal' is common to (1) and (2).

'living thing' is common to (1) and (3)

'human being' is common to (2) and (3).

If no terms were in common, you couldn't have a valid argument, - could you?

Consider:

(1) Every animal is a living thing.

(2) Every human being is a philanthropist.

(3) Every rhododendron is a thing of beauty.

These propositions don't share key terms in a way that would support an inference.

But this example is an example of items - representations - that do have an internal structure. They don't amount to a valid inference because they don't have the appropriate structure.

So that even with propositions, which are structured, you still can't have inferences unless the structures relate in certain ways.

Review question:

What discipline would you think studies the structure of propositions so as to be able to set out what structures are necessary for an inference to be valid?

You could argue: logic

Logicians, you could say, try and set out the structural relationships that have to be there for inferences to be valid.

They would set out the following set of structures for example as supporting a valid inference:

(1) Every A is a B

(2) Every C is an A

(3) Every C is an B

_____________________

So:

You have to have propositions that are appropriately structured in order to have valid inferences.

But also:

Without any structure at all inference as we know it would not be possible.

Propositions are structured, so I can't give any examples of structureless propositions.

But just think of proper names on their own:

(1) Walter

(2) Elizabeth

(3) Chloe

(3) The Empire State Building

A mere proper name does not appear to have a structure - it labels - it doesn't say anything.

For there to be inference as we know it, there need to be propositions; and it is the structure of propositions which makes inference possible.

QUASI-INFERENCES IN MACHINES SIMILARLY DEPEND ON REPRESENTATIONS BEING SIMILARLY STRUCTURED - EXPLANATION

But let us think now not of propositions but of machinery - what sort of representations machinery can handle, and what sort of things it can do with them.

PROPOSITIONS CAN BE REPRESENTED IN MACHINES

Prompt

Can you think of examples of types of thing besides propositions that are commonly represented in machinery? A line of thought.

_________________________________

THE TERM 'QUASI-INFERENCE' EXPLAINED

Machines can represent propositions.

It would be strained to speak of there being propositions in the machine. What there are in the machine are thousands of electronic switches, and it is sequences of these which represent propositions.

You can make a bank of switches stand for

100; or

'the rhododendron is a beautiful genus.'

You can do this just as you can make a pattern on a piece of cotton stand for Brazil, as in a flag.

There is no mystery particularly. You are making one thing - a bank of switches - stand for something - a proposition.

It would be wrong to speak of the representations in the machine as - literally - propositions. It would be better to say that what there in machines are banks of switches, and that these can be made to stand for propositions.

We can call them quasi-propositions. They are intimitely related to propositions, but they are actually not propositions.

So I have explained what a 'quasi- proposition' is.

THE TERM 'QUASI-INFERENCE' EXPLAINED

When the machine manipulates quasi-propositions according to the rules of logic, is it conducting 'inferences'?

You could say they are not exactly inferences, but they are like inferences. Call them 'quasi-inferences'.

So I've (sort of) explained what a 'quasi-inference' is.

Inference is something that applies most clearly in relation to things that are said or written - to propositions. But it can be applied (so it argued) by extension to manipulations of representations that go on in machines.

It is argued that just as you couldn't have inferences between propositions unless the propositions were structured, so you can't have quasi-inferences being performed by machines unless the representations involved were structured in the same way.

For quasi-inference to take place in a machine, its representations have to be structured in the way that propositions are structured. (Ie they have to be 'quasi-linguistically' structured.

It follows that if the brain is a machine, it will be able to conduct quasi-inferences only if its representations are quasi-linguistically structured.

THE BRAIN AS A VON NEUMANN COMPUTER

But what are the wider implications?

It is argued that

(a) the brain is essentially a particular type of computer, the von Neumann type

(b) that the von Neumann computer relies essentially on making quasi-inferences

and that therefore

(c) its representations must be quasi-linguistically structured.

I explain this perspective a little:

One tempting picture of the brain has been that the brain is basically a control device. The picture is that the brain takes in sense data, builds a data bank, works out what it must get the body to do to respond to environmental challenges. For example: a lion approaches and the brain is supposed to take in the fact, take account of what it knows about lions and what they do, and reach the conclusion that evasive action is called for.

(I'm saying the idea is that the brain must be doing this: nothing to do with what goes on in consciousess.)

Another example:

The snake moves at random, finds food, returns every evening to the same spot where it regularly finds food.

From the physicalist perspective, this behaviour will be seen as the result of the brain processing information.

But what does such 'processing of information' involve?

It is argued by one camp - call them 'the von Neumann party' - that it must involve inference, or quasi-inference.

They argue that if we were to program a conventional computer - a von Neumann computer - to respond intelligently to changes in its environment, this is how we would do it. We would get it to represent relevant propositions quasi-linguistically and to manipulate them according to the laws of logic.

They think in fact that can only work like this.

(I don't give here any argument for this last - crucial - step. I can't make one out with sufficient clarity)

Their view is in sum that the representations in the brain must be linguistically structured because

(a) the brain must be thought of as a von Neumann computer, and

(b) von Neumann computers can only work by manipulating quasi-linguistically structured representations.

You can put this conclusion the other way round:

The thesis that the representations in the mind/brain are not quasi-linguistically structured is sometimes argued to be inconsistent with the theory that the brain is a control device on the pattern of a programmed von Neumann computer.

END

Review questions

Can you think of an example of a representation:

(a) that is not linguistically or quasi-linguistically structured?

(b) that represents a proposition but is not structured like a proposition?

(c) that is made of vegetable matter? Answer all three, then check

Poser

Is it true that a control device based on a von Neumann computer could only work by conducting quasi-inferences?

END