xml-dev - Re: "RDF + Topic Maps" = The Future

[Martin Bryan <mtbryan@sgml.u-net.com>]

Nikita

I wrote

> > Both RDF and Topic Maps have the same weakness: They are only as good as
> the
> > semantics they are based on. Neither provides a standardized mechanism
for
> > recording the meaning of the characteristic

You responded

> I tend to disagree.
> XTM distinguishes 2 types of topic subjects:
> addressable and non-addressable.
> If resource is referenced as an addressable subject, it means
> that the subject of the topic is the resource itself
> (just as you said, no meanings or implications)
> Syntactically it is expressed by means or <resourceRef> element.

Please do not conflate Address Identity with Subject Identity and/or
Meaning.

> Extract from [1]:
> A subject which is an addressable information resource, considered as a
> subject in and of itself, and not considered in terms of what the topic
map
> author intends it to signify.

What both RDF and Topic Maps (and XML Namespaces) do is to declare that
things that have the same address are identical. While historically this is
probably not true, as different things can reside at the same address at
different periods, this limited view of identity is sufficient for the
short-term needs of short-lived resources.

> However,
> a resource or a set of resources can be used to identify a non-addressable
> subject.
> For example I can address archives of this list to identify xml developers
> community.

But this does not help Subject Identity. Suppose you point to my name on the
XML Dev list and the name Martin Bryan on the British Library Resource
mailing list. From the addresses alone you cannot determine whether or not
the two people sharing the same name are identical or not.

> Or I can identify a book by its ISBN number (which is a record somewhere
in
> Bowker Data Collection Center).
> Syntactically it is expressed by means or <subjectIdentityRef> element.

What if you point to the author of ISBN  0 210 40394 3? Is there or is there
not any Subject Identity? (The fact that all three are identical in the
example given can only be determined by looking for matching patterns within
the addressed resources, such as the use of mtbryan@sgml.u-net.com as the
email address for all three.)

> Extract from [1]:
> An information resource, considered in terms of what the topic map author
> intends it to signify, and not considered as a subject in and of itself.

That is exactly my point. Users of the topic map are dependent on the
meaning assigned by the topic map author, without having that meaning
recorded for later reviewers to understand/verify. The same is true of RDF.
The meaning is solely in the view of the metadata author.

Actual meaning is much harder to determine. It is entirely dependent on
context, both of the cataloguer and the user of the metadata. For example,
consider the problem of indexing a picture entitled "George Bush Running in
Texas". You might like to consider using terms taken from one the following
ontology fragments:

Computer --> Program --> Running
Sport --> Athletics --> Running
Government --> Elections -->Running
Printing --> Ink -- Running

Which should you associated with the picture? Without knowing something
about the context of the picture you cannot determine which one to use,
though the first and last would appear unlikely to apply. But which of the
other two would apply.? In practice neither of them is likely to be true.
Bush is not "running the elections" - though he may be "running in the
elections". He is also not known for running in athletics races, though he
is known for jogging around. Unless we know the meaning of the term being
used in each situation we cannot accurately determine whether or not the
term can be applied to the resource.

If we just have the address, without a formal definition of the meaning
assigned to that address, then we cannot resolve ambiguities unless we have
access to the address, and that contains a sufficient meaning. Neither can
we determine whether or not the person assigning characteristics to the
resource made a correct decision back whenever this took place, or whther
the same characteristic is relevant in today's environment.

If we have access to the subject definitions, and compare them, then we can
determine their equivalence, but even this is not enough. Until we know the
context in which the resource occurred we cannot determine which ontology to
apply, or which meaning to assign to a term.

What is needed is a standard way of identifying the contexts in which
resources have been created. This is where RDF could help, if it could be
used alongside standardized classification schemes. Rather than being a
general purpose tool that can represent any relational database, it needs,
if the semantic web is to be a usable thing, to become a set of specific
classification databases with recorded meanings that can be referenced by
those seeking to associate meaning with resources.  Until this happens the
web will never be a knowledge-base.

Names and/or addresses are not enough to promote understanding.

Martin Bryan