Re: The Arbitrariness of the Symbol

From: HARNAD Stevan (harnad@cogsci.soton.ac.uk)
Date: Wed Jun 05 1996 - 19:35:37 BST


> Date: Mon, 27 May 1996 15:19:53 GMT
> From: "Parish, Kevin" <kip195@soton.ac.uk>
>
> Why does a symbol need to be arbitrary?
>
> There is a need for an arbitrary symbol system for practical
> reasons. For example in verbal communication it would be
> impossible to cope with any noun, ie you cannot use analog
> processing in speech.

Why not? You need to explain that if the names of things needed to
resemble them and/or be causally connected to them, how could we speak
about them at all (not just verbally, but in any medium -- picture,
gesture, etc.)? To talk about something, you have to be able to name it,
whether it is present or absent. That rules out symbols that are
nonarbitrary because they must be causally connected to the object they
represent..

What about resemblance? If symbols must resemble what they represent,
then HOW CLOSELY do they need to resemble them? If they have to be EXACT
copies, how could we possibly produce exact copies, with out bodies, of
everything there is. If they do not have to be EXACT duplicates, how
much DO they need to resemble what they represent? For the more they
need to resembe them, the more handicapped we are: Think of all the
equipment we'd need to draw, paint, act out in mime and otherwise
"duplicate" all the things there are even approximately.

But it's worse than that, because resemblance is a relation to a
UNIQUE thing, at a particular time and place; where as most of the
things we talk about are KIND of things, in other words, CATEGORIES,
rather than unique individuals. As soon as we speak about categories of
things, most of the resemblance becomes irrelevant: Think of the
mushroom/toadstool example I talked about so many time. In talking about
mushrooms in general -- the edible kind -- as opposed to any particular,
unique, individual mushroom in one place at one time, most of the
"resemblance" has vanished (like "vanishing intersections") and the only
thing they all have in common is the features that distinguish mushrooms
and toadstools.

But, as we discussed in the lecture on categorisation, although those
features must exist, we may not know what they are. So how could we
mimic them when we wanted to use the symbol for mushroom. Moreover, the
features, even if known, could be quite complicated (such as red if big
and green if small, and .... or .... XOR ....): How could the symbol
include all that "resemblance" every time we want to talk about that
kind of thing?

Are you beginning to see why resemblance is a handicap, when all that's
really needed is an arbitrary squiggle, such as a 0 or a 1, as in morse
code?

> Computers function by the use of
> symbol systems and rules for manipulating them. Symbols are
> differnt to analog processes in that they can be manipulated
> without any understanding of what they represent providing
> the manipulation rules are complex enough.

Analog shadows can be manipulated mindlessly and without understanding
too. And symbolic algorithms need not be COMPLEX, they need merely
WORK, i.e., systematically produce the right results, if they are
solving a problem or performing a task.

> This was used to
> good effect in Searle's Chinese room argument, an opposition
> to the Turing test. Searle suggested that since, with the
> right rules, he could do the same as a Turing machine (fake
> having an understanding of conversation) with chinese
> symbols without understanding them.

This is indeed one of the "big issues" in the course, but it is not
clear how it is related to the question under discussion here, which is
the arbitrariness of symbols. (There IS a connection, it is the
arbitrariness of the symbol's physica; "shape" that gives rise to the
sofware/hardware distinction and the irrelevance of the specific physical
implementation of symbol systems. It is this implementation-irrelevance
that Searle exploits, by himself becoming yet another implementation of
the same symbol system that passed the Turing Test and allegedly
understood Chinese; yet Searle can TELL us, truly, that he doesn't
understand Chinese. So neither did the other implementation, the
computer, since the physical differences between the two implementations
of the same symbol system are irrelevant. If one understands simply
because it is the implementation of the right symbols system, the other
should do so too.)

But I doubt you had that in mind, in which case this kind of
free-association unfortunately cannot earn marks...

> If the computation
> theory for human brain functioning is correct then the brain
> also uses symbol processing. They are necessary for us to
> learn by description. There are problems with the idea of
> mental imagery, firstly it appears to require another mind
> inside the mind hwo in turn requires another homunculus so a
> system of processing that does not require images is
> desirable in solving this problem. In the end though
> 'symbol' is merely a symbol itself, which represents an
> arbitrary label or representation of something.

Again, these references to other themes are not related to the question
at hand, which is: why does the shape of the symbol need to be arbitrary
in computation?

> Stevan please note
>
> I found the connected sky-reading and the recommended
> Anderson book of no help whatsoever in answering this
> question. Any reference to this subject was near impossible
> to find. In particular compared to many of the other
> questions where large amounts of readily available materials
> could be attained

Your point is well-taken. There is also relevant material in
the readings on categorisation, but I agree that that one was
underdocumented in the readings (though it was discussed
several times in lecture). It will not appear on the exam. I will
post this separately (though those who had tried the revision
could have contacted me by skywriting to request a more readings
or a fuller explanation...).

In general, as I said in the beginning of the course, there is
no textbook that covers precisely the material covered in the
course, so going through the glossary of a test like Anderson
will not necessarily help -- though Anderson will still come
in handy for later cognitive courses.



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:45 GMT