Walter Underwood writes:
> In library jargon, "classification" gives a unique hierarchial
> location for each document, because it is used for shelf position,
> while "subject headings" allow discovery under multiple categories.
> A subject heading might have multiple parents, while classification
> cannot.
In this distinction between "classification" and "subject" lies the
reason why all attempts at classification or providing "category" support in
open communication systems have failed over the years. People keep trying to
make "categories" work when we've proved over and over again that they can't
work in anything but closed or homogenous groups. What we should do is give
up on categories and find a way to support subjects instead. (in topic map
terms: Stop focusing on Topics. Focus instead on Subject Indicators.)
See: B.4 in http://topicmaps.org/xtm/index.html#conceptualmodel
Categories require agreement that cannot be achieved:
There are as many systems of categorization as there are users of
information. In fact, single users often utilize multiple context-dependent
systems of categorization. The best that any classification can do is
describe a document or object within the context of one of those systems.
The problem is, of course, that unless we all agree to a common system, the
classification will be meaningless -- and can be harmful.
It is a commonplace observation in IR and Library Science that the
proper classification of an item can only be known after it is retrieved and
used. You simply can't "pre-classify" an item in advance. On the other hand,
you *can* identify the subject of the item.
Categories are "undemocratic" and repressive when used in open communication
systems:
For categories to be useful, all viewers of the categories must
either agree on the system of classification or build mappings from a common
system of classification to their preferred systems. However, classification
systems are inevitably influenced by subjective decisions that influence the
thinking of those who use the systems. For instance, at an extreme..., in
one classification system, an argument in favor of "legal abortion" would be
classified as "pro-choice" while in another it would be "anti-life." The
difference is not mere wording -- it implies a framing of the issue. To
force that framing on an open system effectively closes that system.
On the other hand, there are many fewer disagreements about what
constitutes an item that might be constituted. i.e. we can agree that the
subject "legalized abortion" exists even though we might disagree on how it
is classified.
Categories break I18N:
Translating a classification system from one language to another is
not simply a matter of translating words. Different languages have varying
ability to communicate concepts and result in their users structuring
classification systems and even arguments differently depending on which
language they are using. Any attempt to get a consistent cross-language
classification system will result in users of one or another language
finding that the very manner in which they think about issues is devalued.
There is some, but very little, disagreement between users of
variant languages on what "subjects" are. Subjects are much easier to
translate then classifications. (For instance, it is often said that
"Zeitgeist" really has no English equivalent but those of us who speak
English typically just add Zeitgeist to our voculary. Similarly, the English
term "weekend," which refers to an interval of days rather than an instant,
doesn't appear in other languages. Non-english speakers usually incorporate
the English if they want the concept -- think "Le weekend.")
Personally, I think we should support a "subjectIndicator" which
would be a URI. This should be the focus of the discussion that appears to
surround "categories." The mapping of subjectIndicators to
classifications/categories should be a function provided by the aggregator.
Each user should be free to use whatever mapping they want. For instance,
various organizations (DMOZ, your church, your professional society,...)
might use Topic Maps to define these mappings in an interoperable way.
If folk insist on knowing an author's view of classification, we can
add the various attributes proposed in the "PaceCategory*" stuff as optional
elements to the "subject" element. These should, however, be seen as simply
providing hints, not as making definitive statements about classification.
Trying to solve the classification/category problem is futile. We
should be focusing on subjectIndicators. Given subjectIndicators, we'd be
able to have democratic, I18N compliant, personalized and relevant support
for classification of entries.
bob wyman
