From: Mark E. Shoulson (mark@kli.org)
Date: Mon May 24 2004 - 22:02:09 CDT
John Hudson wrote:
> Michael Everson wrote:
>
>>> Classification is an arbitrary process in which one produces useful
>>> categories into which to arrange an otherwise unwieldy body of
>>> knowledge.
>>
>
>> I dispute this. It is not arbitrary. Sometimes the cuts are difficult
>> to make, because there is messiness in the data, but classification
>> puts like with like and separates like from unlike. If it were
>> arbitrary, we would not be able to distinguish abugidas from
>> syllabaries, or trace the relationships between scripts and name the
>> nodes on the tree.
>
>
> *All* classification is arbitrary. This is a basic philosophical
> proposition. Note that arbitrary does not mean baseless or capricious,
> it just means that systems of classification are determined by the
> classifier, not by the thing classified.
It's something Machine Learning researchers deal with from the get-go:
learning bias. (Come to think of it, I worked with a psychology
professor once whose main field was categorization, both from a
philosophical and psychological standpoint). You can't learn *anything*
without applying some "bias", some prejudice to the process. Prejudice
is usually considered a Bad Thing, but it isn't. According to
Watanabe's(?) Ugly Duckling Theorem (note that it's a theorem, that
means there's a proof for it), *everything* is exactly as similar and
exactly as dissimilar from, well, everything else. The only way you can
make meaningful categories is somehow to stack the deck and consider
only some aspects and decide what's important.
~mark
This archive was generated by hypermail 2.1.5 : Mon May 24 2004 - 22:02:44 CDT