Re: Wittgenstein and Concepts

From: Harnad, Stevan (
Date: Mon May 19 1997 - 11:38:45 BST

> From: Walker, Guy <>
> I (seem to) remember in a lecture on language (I think) you mentioned
> that Wittgenstein had some interesting ideas on the subject - what were
> they? (Or if they can't be talked about, I will pass over them in
> silence!) I would be interested to find out (even if no one else would
> be)

Here are some of the things Wittgenstein said about language:

(1) There can be no "private" language: a Humpty-Dumpty-like one, that
you invent, in which words mean whatever you mean them to mean:
Language arises from a community, where there are shared conventions
about what means what. Imagine if you had a word like
"auto-pleasance," which would mean "looks good to me right now": Not
only would no one else have any idea what is and isn't auto-pleasant,
but you yourself could not know whether or not something you saw was
really auto-pleasant, or only seemed auto-pleasant. There is no error
in a private language, and hence no correction for error.

(I think Wittgenstein's private language argument is wrong, by the way,
but you asked only for his interesting ideas...)

(2) "Meaning is Use": The idea behind this one is that the meaning of
words is not what they stand for, or the features on the basis of which
we know what they stand for: Meaning is generated "on the fly" on the
basis of how words are used in a language community (which will of
course have some shared public conventions about what word is used,

(I think this one's wrong too...)

(3) "Family Resemblance": Wittgenstein suggested that there was no
point in looking for what things denoted by the same name have in
common: There's nothing. Their resemblance to one another is more like
the resemblance among different members of a family. His famous example
for this was the concept of "game": Games vary wildly, from something
like a physical sport to mind-games of your own invention. They do not
seem to share any particular feature, yet we know is and is not a game.

(Flawed again, I'd say -- and will say if/when I write a book on
Wittgenstein, but for now...)

(4) Wittgenstein on "rules": Wittgenstein suggested that there was an
important distinction to be made between actually following a rule, and
merely being decribable as if following a rule. The thermostat can be
described as following the rule "turn off the heat if the temperature
rises above 70 degrees" but that rule is nowhere represented in the
the thermostat. (This is relevant to some of the comparisons between
symbol systems, which work on the basis of implicitly represented
rules, and neural nets, which are describable as following rules, but
don't really represent them.)

> My second question is regards "concepts". If the concept of an apple is
> a red round fruit or whatever, then many things could be red round
> fruits and not be an apple. This explanation seems a bit simplistic -
> could an apple (for example) consist of a whole variety of different
> nuances, which are barely open for conscious inspection - but
> nonetheless help me to recognise an apple in preference to a plum. This
> as opposed to a few simple defining characteristics. Also, it is not
> inconcievable, that as this concept was being refined in my own mind,
> that at some stage I could have mistaken a plum for an apple.

That's all correct, and well-thought out. The problem is that for most
concepts we do not know what features they are based on. If what you
mean is that there must nevertheless be some features that we "know"
implicitly, the way the thermostat or neural net "knows" them, then I
agree. If you are saying that there are simple features, that we do
know, but not right away (so we may sometimes over- or under-generalise
about what is or isn't an apple until we finally get it right) then you
might be right for apples, but not every concept is as simple as the
concept of an apple. Try it with "bird" or "fish".

In any case, the features must exist (if we can always say correctly
what does or does not fall under the concept); the only issue is whether
and how we know those features: explicitly or implicitly.

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:52 GMT