Re: Schyns Comms 04-06 burgund charles cleeremans

From: Wright Jon (
Date: Thu Mar 05 1998 - 10:57:11 GMT

Comments on Sarah Willoughby's comments [sw]

> "definition of categories is a separate issue from recognition of
> category members"

sw> so we need not try to provide an approach that deals with both issues.
sw> However, it is hard to see how the two can be separated when they are so
sw> fundamentally linked. Surely the development of categories is determined by
sw> features of the category members.

sw> The boundaries of peoples categories vary according to
sw> their experiences of labelling. These boundaries may just be
sw> cognitive and not sensory. This would appear to back up the
sw> flexible feature theory. Fixed features would not allow for such
sw> adaptation.

The classical theory of categorisation is that the perceptual atoms,
the lowest level of discriminable features, are combined in a variety
of ways to form any "higher order" categories. By this definition,
however, all useful categories are higher order.

An individual's experience sets the categories, initially perhaps by
fixed features and later by comparison to existing features of the
category, its subsequent inclusion (or rejection) for a category and
perhaps modification of the category boundaries.

> "What is being learned is not definitions of categories nor sets of
> invariant features, but rather the features that best characterise the
> members of a category and differentiate them from members of
> contrasting categories."

sw> Charles appears to be advocating the prototypical model of categorisation, a
sw> more dimensional approach. This does seem to allow greater flexibility.

The notion of a prototype is certainly more flexible than if the
category had to be determined entirely by a set of invariant features.
The best example of a category is used when analysis a new object and
the features of that object might be mapped onto the features of the
new one and a certain level of agreement in the mapping process would
permit membership of the new object to the category.

sw> Could it be that some features may be more flexible because they are less
sw> concrete, while well rehearsed features become more fixed.
sw> Burgund & Chad seem to be suggesting this.

> "alternative materials may elicit feature creation more readily than
> traditional materials because they are more likely to be processed in a
> specific subsystem. In contrast, stimuli that are readily decomposable
> into familiar parts may be processed more effectively in an abstract
> subsystem."

Does this mean that abstract subsystems involve the categorisation of
more unusual (=less well rehearsed) features? If a categorisation task
requires objects to be sorted according to the presence of a vertical
line, for example, line drawings would be better stimuli than real
objects since real objects contain other features beyond those in
visual recognition which would make the decomposition into abstract
features harder.

sw> The use of novel items in experiments allows experimenters to look at
sw> how people deal with 'raw' materials and categorise. Cariani allows
sw> for a system that defines categories and creates new ones while using a
sw> decision rule that maps pre-existing features to classes. This is
sw> contrary to Charles who claims that such ideas about features are
sw> restrictive and not explanatory.

sw> It seems that there is no satisfactory explanation at present and the
sw> issues are still open. Cariani believes that the brain forms new
sw> categories for novel stimuli. Burgund and Chad say that it is still
sw> unclear whether fixed or flexible features are used for category
sw> recognition and Charles thinks that trying to tie ourselves down to
sw> defining category features is fruitless.

> "If it is in principle impossible to distinguish between them, perhaps
> it is best to acknowledge that there may not be a set of invariant
> primitive features which underlie visual recognition."

Features with an attentional weight of zero are indistinguishable from
absent features. If invariant features cannot be discovered then all
categorisation must not be derived from the combinations of these tiny
parts. Prototypes are more fuzzy but as categories are modified by the
inclusion or rejection of later examples it seems that the definition
of a category really is flexible, but with a 'most typical' member to
compare all others against. Membership might be decided by a sufficient
amount of similarity.

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