Re: Classical Categorisation

From: Harnad, Stevan (
Date: Sun Feb 04 1996 - 18:29:21 GMT

> From: "Lucas, Melody" <>
> Date: Sun, 4 Feb 1996 18:06:46 GMT
> The classical view is that categorisation occurs on the basis of a set
> of features that are necessary and sufficient for category membership.
> This view was heavily criticised by Rosch. Rosch viewed categorisation
> as a form of cognitive economy (to 'reduce the infinite differences
> among stimuli to behaviourally and cognitively usable proportions')
> which reflects the structure of the perceived world (i.e. fins occur
> more often with scales than with feathers).

But what if the things that matter to us -- whether or not we get lunch
-- don't follow the obvious correlations, but are rather buried beneath
them? It's not correlations between features that matters, but
correlation between features and the consequences of sorting things one
way or another.

> Category membership was
> judged by how close to a prototype an object was. This involved a
> judgment on how typical a potential member was to a particular
> category.

Close to a category? How close an apple is to fruit? You mean how close
it was to some sort of ideal representative of the category.

> She defined levels of categorisation; the vertical and the
> horizontal. The vertical referred to the level of inclusiveness;
> superordinate (say, machine) the basic (computer) and the subordinate
> (IBM 486SX Processor). The horizontal referred to the segmentation of
> categories at the same level (e.g. machine, animal, vegetable).

Kid-bro finds the foregoing 3 sentences completely incomprehensible:
Define, explain, give examples: You must prove you know what you are
talking about rather than merely repeating the keywords.

> Rosch
> disagreed with the classical view on a number of points. First, she
> purported that subjects often cannot tell you the list of features on
> which they base their categorisations. Second, that some stimuli are
> more typical than others, and are thus more quickly categorised.
> It came to light that this argument not only clouds the job of the
> psychologist by mixing the metaphysical with the psychological

Good, but what does that mean?

> but also
> confuses the whole question of what categorisation is. The qualities of
> the stimulus per se are irrelevant. We, as psychologists, are
> attempting to discover how a human being abstracts some information
> from a stimulus and ignores the rest. Watanabe's ugly duckling theorem
> describes how every 'thing' can have an infinite amount of
> characteristics, i.e. everything has negative and positive attributes
> (e.g. red or not red). How do we select the information which
> categorises one thing from the next with any degree of accuracy amongst
> this great mass of information? Funes could not abstract category -
> relevant information and saw every event as being unique, leading to
> extensive problems with recognition of everyday things. He lacked the
> ability to disregard certain characteristics from objects / events
> which is something apparently necessary for 'normal' daily
> functioning.

I more or less understand what you mean, but would kid-bro, who hasn't
read (or hasn't understood) Watanabe or the Funes story?

> The overriding conclusion is that the environment selects categories,
> the human does not.

Not sure what you mean: Surely it's the human that has to do the
sorting, but the CONSEQUENCES of sorting one way or the other are
dictated by the environment.

> It does so by means of their consequences, not by
> their features or typicality.

But in the end, if the category is learnable at all, it must be based on
finding the features that are correlated with the consequences, and
focusing on them, ignoring the rest. So you're right that this has little
to do with typicality, but it has everything to do with features.

> For instance, learning either by theft or
> honest toil that toadstools are the ones that make you sick, but
> mushrooms are edible.

What do "theft" and "honest toil" mean?

> Finally, assignment of objects/events to levels
> of categories is arbitrary; it varies among people and should thus be
> considered as a network.

What is the "it" that varies and should be considered as a network (and
what's the next step after dubbing it a network)? What's arbitrary is
the idea that there is a universal "basic" level in the abstraction
hierarchy that starts with unique Funes-instants and rises, through
concrete object categories, then ever more abstract categories, some
still grounded directly ("honest toil") through trial-and-error category
learning, some grounded indirectly ("linguistic theft") through being given
a string of symbols that tell you what's what, and perhaps also what the
critical features are. There is no "basic" level in any of this, except
possibly those based on common evolutionary histories (e.g., the frog's
bug-detector) or shared physical or social environments that dictate that
certain categorisations are more frequent or important than other.

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:58 GMT