Re: Schyns Response R1-R4

From: HARNAD Stevan (harnad@cogsci.soton.ac.uk)
Date: Sat Mar 07 1998 - 15:17:34 GMT


On Wed, 4 Mar 1998, Darling, Andrea wrote:

> > Either people come equipped with a complete, fixed feature
> > repertoire that accounts for all present and future categorizations,
>
> (THIS IS UNLIKELY AS BABIES ARE NOT NECESSARILY AS ACCURATE AS ADULTS
> IN CATAGORISATION, EG. THEY OFTEN CALL ANYTHING WITH FOUR LEGS ^A
> DOG^.)

Having all the features doesn't necessarily mean that you can use them
correctly from birth, without training. You may have to learn WHICH
fixed features are the features that reliably distinguish the things our
parents have decreed to be "dogs" from those that they have decreed are
not.

> IT IS PROBABLY BOTH: SIMPLE, INNATE REPERTOIRES (EG.
> HOW TO TELL APART DIFFERENT COLOURS) MAY LATER BE SUPPLEMENTED BY THE
> ABILITY TO CREATE NEW ONES; AS SEEN IN THE ^MARTIAN BLOBS EXPERIMENT^.

New colours? Surely it's easier to imagine that we can create new
combinations of parts than that we can create new sensory experiences.

> YES IT DOES, BUT WHAT ABOUT THE HOLISTIC VIEW (EG. GESTALT). IN
> EVERYDAY LIFE WE OBSERVE THAT OBJECTS ARE NOT SIMPLY A COMBINATION OF
> FEATURES THAT ARE PROCESSED BOTTOM-UP. FOR EXAMPLE, HOW OUR BRAINS ARE
> MISLED BY ILLUSIONS: OBJECTS APPEAR DIFFERENTLY ACCORDING TO THE
> PERSON^S PAST EXPERIENCE, EXPECTATIONS ETC. NUMEROUS EXAMPLES PREVAIL:
> EG. MULLER-LYER, NECKER CUBE, KANIZA TRIANGE, AND THAT GREAT ANNOYING
> EXAMPLE OF WHEN YOU^RE ON A TRAIN AND YOU THINK YOUR TRAIN IS MOVING
> WHEN IT ISN^T! (MOTION PARALLAX).

The "holistic" view itself needs to be explained. Sure we see things as
wholes, not as bundles of features, but HOW do we manage to see them as
wholes, and as wholes of the right kind? The fact that we are not aware
of the features we use to do this does not mean we are not using them,
or that they do not exist. If we knew what they were and how we detected
them, cognitive psychology would be a lot easier to do.

The illusions are interesting, but have not yet answered the "how"
questions.

> > feature combinations allow for structured
> > hierarchical representations; similarity relations between different
> > objects can be expressed in terms of their features and their
> > combination rules.
>
> YES, BUT THIS DOESN^T ALLOW SUCCESSSUL GENERATION OF ALL CATAGORY
> MEMBERS AND MAY LEAD TO MISTAKES. SOME CATAGORY MEMBERS (THE NOTORIOUS
> TOMATO) ARE MORE SIMILAR TO MEMBERS OF OTHER CATAGORIES THAN TO THEIR
> OWN. FOR EXAMPLE, A TOMATO IS A FRUIT (DUE TO IT^S PIPS, I THINK!)
> YET IS OFTEN CONFUSED WITH VEGETABLES AS IT DOESN^T GROW ON TREES
> (LIKE MOST FRUIT), IS EATEN WITH SALAD ETC.

A model for how we categorise must explain not only our successes, but
our failures, especially during the process of learning categories.

> THIS APPEARS VALID AS IN EVERYDAY LIFE WE ARE QUICKER TO RECOGNISE
> FACES WE KNOW IN THE STREET THE BETTER WE KNOW THE PERSON. WE APPEAR
> TO BUILD UP HOLISTIC PROTOTYPES OF FACES WE KNOW WHICH AIDS PROCESSING.

The idea of a "prototype" is seductive. Like mental imagery, it fits our
intuitions about how we recognise categories. The trouble is that it has
not yet been shown to be able to do the job in any successful model.

Moreover, this intuition is, as intuitions often are, homuncular:
It doesn't help to say that I recognise birds because they look like a
prototypical bird: How do I know that they look like a prototypical
bird? What happens, from the time the bird casts its shadow on my
retina, to the time I recognise it as a bird, that gives me the capacity
to do that? To recognise it as an example of a prototype bird in my head
is already to recognise it as a bird.

> HOWEVER, FEATURE PROCESSING IS STILL IMPORTANT UNDER SITUATIONS OF
> AMBIGUITY (EG. WHEN PERSON IS WEARING SUNGLASSES THEY DON^T USUALLY
> WEAR).

Believe me, even the obvious, unambiguous cases are still
near-impossible to model.

> INDEED, A FEATURE CAN BE ANYTHING, FROM A PERSON TO PLANET EARTH TO THE
> SOLAR SYSTEM TO THE GALAXY- IT DOESN^T MATTER AS LONG AS IT MAKES UP
> SOMETHING ELSE....

And you are able to pick out that something else, by using that feature,
whether consciously or not...

> > Independently
> > perceived features can evolve both from a chunking process which
> > produces configural features, from an imprinting process which creates
> > a new form from the visual medium, or from dimensionality reduction
> > itself.
>
> YES, BUT MOST FEATURES ARE NOT PERCIEVED INDIVIDUALLY: THE FEATURE IN
> IT^S ENTIRE CONTEXT NEEDS TO BE CONSIDERED; WHEN ISOLATED, A ^MOUTH^ IS
> ONLY A WONKY LINE, YET WHEN ADDED UNDER TWO EYES AND A NOSE IT BECOMES
> OUR WELL KNOWN MOUTH!

It's important not to think only of PART features; in some ways, part
features are the easiest cases, because it's so easy to see how parts
can be recombined, or how things can be taken apart into their parts.
But colour, shape, texture, pitch, etc. are not parts. They are (simple
or complex) sensory dimensions, and often features consist of just the
presence or absence of a dimension (colour), or a value along that
dimension (emerald green).

What you say about not perceiving parts individually is even truer of
dimensions, and values along dimensions.

> > continuous variability raises an important problem that was not
> > discussed in the target article: How does the system store information
> > about feature variations?
>
> YES, AND HOW DOES IT HAVE THE ROOM!

Reasonable question, but there are a very large number of neurons
available in the brain, even in just the sensory parts of the brain. We
clearly have enough to be able to PERCEIVE all the things we can
perceive. Surely sorting them into their proper categories requires less
information rather than more information than that. (As we learned with
Funes the Memorious, categorisation and abstraction, indeed the
abstraction of features, involves ignoring, suppressing and forgetting
most of what our brain is quite capable of perceiving in full.

> > exemplar-based systems would rapidly be overwhelmed with the millions
> > of feature exemplars it would need to store and organize.
>
> YES, BUT EXEMPLARS DON^T NEED TO BE THAT SPECIFIC. FOR EXAMPLE, AN
> EXEMPLAR OF A MINI (EG. ROUNDED BODY, SMALL, MINI LOGO, SMALL ROUND
> HEADLIGHTS ETC..I^M NO EXPERT ON MINIS!) ALLOWS US TO RECOGNISE MOST
> MINIS (EG. IN MOST ILLUMINATIONS, SITUATIONS ETC). WE WOULDN^T
> NECESSARILY NEED AN EXEMPLAR FOR EVERY DIFFERENT TYPE OF MINI IF THEY
> SHARE ENOUGH BROAD FEATURES (WHICH THEY GENERALLY DO).

An "exemplar-based" system would be as useless as Funes's infinite rote
memory for every exemplar experience he had in his life. To be helpful,
the exemplar must "capture" only the relevant features of those
experiences, and ignore the rest, if it is to help us see them as
KINDS (categories) of experience rather than merely experiences, each
exemplar being infinitely unique (like Watanabe's world of objects).



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