Re: Chapter 8: Meaning

From: Harnad, Stevan (harnad@cogsci.soton.ac.uk)
Date: Sun May 25 1997 - 21:18:05 BST


> From: "Anonymous" <anonpy104@psy.soton.ac.uk>
>
> 1) P 220 the idea that words stand as one to one relations
> (Fodor et al).
> e.g. 'has anyone seen the tray?' would be translated into
> the token SEE (ANYONE TRAY).
> This idea is dismissed because no contact is provided with
> the world (is this the symbol grounding problem?)

Yes

> and the
> tokens do not specify specific referents. I thought that
> context effects would take care of of this.

Context would work if YOU were the one thinking or saying the symbol
string. But if we are going to model what you can do, so as to explain
what you can do, there's no point saying YOU (or I) could figure this
out in context: How is the MODEL to do that? The answer is that it
can't do it with just more symbols. The symbols need to be connected to
the world in some way. (The candidates include analog sensorimotor
projections, mental models, neural nets, etc.)

Be careful to distinguish when a theorist is talking about what a
particular model or kind of model can or cannot do, and when he is
talking about what YOU can or cannot do: YOU can do it with the help of
context, but how are we to add that capacity to the model?

> 2) We construct mental models of situations based on
> propositions, and compare these models to what is actually
> happening.
> e.g. the circle is to the right of the star.
> on P 222 it says that 'an infinite number of models cannot
> be constructed'. This would seem to be related to the frame
> problem. But, I think that we can construct an infinite
> number of models and test them against situations and
> propositions.

Yes, if a problem required constructing an infinite number of models,
then unless there is an algorithm (a symbol-mixing recipe) that can
generate the infinite models with a rule that is itself a bit shorter
than infinitely long -- unless there is such an algorithm, then if your
lunch depends on getting the problem right, you won't get lunch.

Fortunately, for the problem discussed on P. 222, you can solve it
without constructing an infinite number of models. You just make
one model, and see whether it gets you lunch, and if it doesn't you
revise it till it does (or till you expire!).

Remember that mental models of the sort discussed in this chapter
are like mental arithmetic: They are things you do in your head in
solving certain kinds of logic problems. Distinguish that kind of mental
model from the model that a cognitive theorist constructs to explain
what we can do. One of the things we can do is to do mental logic
(using mental models -- or, as Prof. Johnson-Laird said, doing it
on paper with a physical model); if someone wants to model THAT, they
have to make a model that can do logic problems by using models.

> e.g. the circle is to the left of the star.
> the orange circle is to the left of the star, the blue
> one to the right.
> the orange circle is four from the left of the blue
> triangle which is below the green star.
> etc etc etc
> If we were unable to construct an infinite number of mental
> models we would not be able to test the infinite number of
> possible propositions that are waiting to be produced.

But we don't have to generate all those propositions. Typically a
problem will go like this: "The circle is to the left of the square.
The square is above the triangle. On what side of the circle is the
triangle. This can occur in an infinite number of ways:

For example, the circle could be any number of sizes, the distances can
be small or huge, etc. You try a simple model first, in which the
figures are all the same size; you can do it on a paper or in your
head. Chances are the answer is that the triangle is to the right of
the circle. If not, you try revising the model, changing the sizes
or distances, or even the number of dimensions. The text here was
only saying that you try something finite and simple first, and
then revise (hoping you succeed before you get hypoglycemic).



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:53 GMT