Re: Schyns

From: HARNAD Stevan (
Date: Thu May 21 1998 - 18:55:34 BST

> From: Bilak Alexandra <>
> I understand that we perceive the world and objects in the world by
> distinguishing objects among others, and being able to categorise
> and label them. By doing this, we are extracting relevant features of
> each object, which will enable us to sort them as specific
> categories of objects. For example, the reason why I can see the car
> in front of me amongst the incredibly large amount of information
> arriving at my senses is because I am recognising the features
> "relatively big", "four wheels", "windows", etc... and inferring the
> category "car" from it (or are these just parts?)

Wheels and windows are parts but big and four are not. Schyns et al.
focus on part-features, but those are just one kind of feature. Colour,
shape, texture, etc. are all non-part features, or property-features.
Both part-features and property-features can themselves be categories,
but only part-features can be OBJECT categories: Wheels and windows are
objects, like the objects of which they may happen also to be parts.
(So are circles and lines), but colour is not an object; nor is round,
or roundness. Non-part features, property-features, are necessarily
more abstract than objects.

As such, I think they are more representative of what features really

> Anyway, what I don't understand is when they state that categorical
> constraints can influence features: how can features be "extracted and
> developed" as an organism categorises its world? I don't seem to
> understand the blob experiments and the order of learning X, Y, and XY
> (and even less the one with the Z), let alone what this could possibly
> show.

Have a look at my explanation of it (to Chantal) in Skywriting.

Perhaps you will understand it better if you think of features as
having to be extracted IN ORDER TO categorise the world: To see that
there are trees and animals, you have to be able to extract the
features that distinguish them. Sometimes they will be part-features,
but more often they will be property-features. Check how and whether
Schyns et al's X/Y story applies to property features.

("Extraction" and "abstraction" are related: they both involve picking
out some parts/properties and ignoring others. To categorise is to
generalise over KINDs of things. To generalise, you need to
extract/abstract; you need to pay attention to some features and not
others. (If you pay attention to everything, then everything is
infinitely unique and there are no categories.)

Categoorisation influences what features we abstract in two ways: If
our categorising is guided by feedback from doing it correctly and
incorrectly (e.g., eating edible and toxic kinds of mushrooms) then it
is the constraint of having to get it right that influences the
categorisation. I need to find the features on whose basis I can sort
correctly. Conversely, if I am sorting in a certain way, using certain
fetaures, that will influence my categorisation, biasing me towards the
distinguishing features and away from the nondistinguishing ones.

So learning a category influences which pictures we must extract from
what we see, and having learned a category, and hence to extract
certain features and not others, influences how we see the things we
are categorising. (This is a Whorfian effect if it is induced by
learning rather than being inborn.)

> Could you possibly explain, IN KID SIB (to be honest, I think this
> article suffers from a serious lack of kid sib considerations)?

I've done it in other postings for the specific questions you raise
here. Raise some more, and I'll explain more...

Professor of Psychology
Director, phone: +44 1703 592582
Cognitive Sciences Centre fax: +44 1703 594597
Department of Psychology
University of Southampton
Highfield, Southampton

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:21 GMT