Re: Schyns Comms 19-21 palmeri perruchet phillips

From: HARNAD Stevan (harnad@cogsci.soton.ac.uk)
Date: Sat Mar 07 1998 - 14:31:38 GMT


On Mon, 2 Mar 1998, Chan Dorothy wrote:

Perruchet:

> Perhaps what we claim to be
> new features are the stuff that are different from the fixed "old"
> categories but which of them might become relevant for us at some of
> our life. Therefore, it is necessary to "label" them.

Yes, one possibility is that there are no new features, just features
that happen to be the features of the things we need (or choose) to
name; it can even be the feature itself that we choose to name.

> But how can we
> communicate with each other? Why do we know what is a square or
> round object. Are they just terms for conventional use?

Clearly, either we all have and use the same features or they are
similar enough that we don't notice that we may not mean exactly the
same thing all the time, when we name things or their features.

Palmeri

> our memory capacity is vast but how can we
> successfully draw out the "right feature" at the right feature
> decision moment?

That's one of the big, not yet solved problems of cognitive psychology.
We need to "reverse engineer" a mechanism that can find features and
categorise as we can and do.

> Can the "prototype theory" be of any use? In
> that, the prototypic member sets out the path for allocating
> weightings for certain features? But how does this comparison
> process work?

This is controversial. A successful prototype model that manages to
categorise as we do, and uses distance from the nearest prototype in some
sort of a continuous internal feature space would be strong evidence
that this is the way our brains do it too, but there is no such model so
far. Prototypes seem to be better suited to explaining "typicality"
effects (why we think a robin is a "better example" of a bird than a
penguin, and why we can categorise it more quickly) than they are in
explaining categorisation itself.

We have to remember that most categorisation is CATEGORICAL, which
means that the members of the category are complete members, not
"partial" members. So a prototype model would have to explain how there
are all-or-none category boundaries in feature space. That's the hard
part to model and explain; typicality is the easy part, because it
already presupposes that we can categorise, and only explains why we
categorise some members more quickly than others, and consider them
more typical of the category.

Phillips :

> How at the end of the day do we say that "this is a feature?"
> Feature is the term which we develop to define objects or even
> events or to specify that "something" is a characteristic of
> "another thing" or is it beng created for organisation purpose?!
> How can it be stored in our brain?

To see (when it is not obvious) that some things are of the same "kind"
and others or not, we have to find a way to distinguish, correctly and
reliably) those that are and those that are not the same. What we USE to
do this must be features of some kind. Our brain must be able to detect
and use those features. Whether it detects them innately, or somehow
manufactures a special detector to suit the specific categorisation
problem is what is at issue here.



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:20 GMT