[sc34wg3] revised draft Reference Model document N0298

[Steven R. Newcomb]

"Martin Bryan" <mtbryan@sgml.u-net.com> writes:

> I like the idea of using "subject nodes" to identify
> what dRM records. This is what they are in the
> model. They are "the points in the model at which
> this concept (what you seem to call the "territory"
> of the "subject", a phrase I am not sold on) is
> used".

I think maybe you misunderstood my intended meaning for
the word "territory".  For me, the "territory" for
which Topic Maps serve as "maps" is always some set of
subjects.  Subjects are not expressions of themselves,
they just *are*.  For example, the land occupied by the
State of Texas is not an expression of itself, it
simply is.  A map of Texas, on the other hand,
*expresses* things about the territory occupied by
Texas.  

For me at least, in Topic Maps we have "topics" (the
markings on our maps that represent specific things)
and "subjects" (the things represented by those
markings).  When I use the term "territory" in the
context of explaining Topic Maps, "territory" is a
metaphor for "the set of subjects for which the topic
map is a map".

Do I understand you to be discussing a third kind of
thing, which you call "concept"?  Or is "concept" for
you the same thing as "subject"?  If the subject of a
<topic> is the Statue of Liberty, what is the "concept"
of the Statue of Liberty?

> > If we're religious about saying "topic node" when we
> > mean node, then why shouldn't we also be religious
> > about saying "<topic> element" when that's what we
> > mean?

> No particular reason, though I object to having to
> use the angle brackets, especially as these fail to
> make it clear that the attributes of a "topic
> element" are more important than its content (if you
> don't believe me try writing topic maps without
> identifiers!)

When I'm struggling to be clearly understandable with
respect to the question of whether I'm discussing a
syntactic element or a semantic element, I'm not
usually trying, at the same time, to offer a tutorial
on the details of any specific syntax.  

As the foregoing paragraph demonstrates, the word
"element" by itself is still ambiguous, which is why I
use the angle brackets.  The only other way I can see
to be clear about this is to say

  "instance of the 'topic' XML/SGML element type"

... and it seems to me clearer, briefer, more
intuitive, and less likely to introduce extraneous
issues, simply to say "<topic>".

It would be hard to overestimate the amount of valuable
time and effort that has been lost because someone said
"topic", meaning one thing, but was understood as
meaning another.  When what is meant is "instance of
the 'topic' SGML/XML element type", the use of angle
brackets seems to me the lowest price available for
protection from such losses.

I note that in the remainder of this letter, you've
made no effort to disambiguate your use of the term
"topic".  I'm operating under the general principle
that, if Martin Bryan writes the word "topic", he means
"<topic>".

> > I don't think it would be right to do that.  I don't
> > see how a subject that is a role can be the same
> > subject as a subject that is an assertion type -- even
> > if both subjects happen to have the same name.

> > > I quoted the example of a topic called
> > > "PublishedBy". This could be an association
> > > type.

> > Sure, it could be, but *is* it?  It is the topic map
> > author's responsibility to have exactly one subject in
> > mind for each topic.

> I deliberately chose PublishedBy as an example of
> something that can both be internal to the topic map
> and external to the topic map without changing the
> basic concept of "having been responsible for the
> publishing of".

"having been responsible for the publishing of"

is just as ambiguous as 

"publishedBy"

Perhaps you are confusing the map with the territory.
The fact that two dots on the map can both have the
name "Paris" does not mean that they represent
identical things.  If a map is to be useful, it's
critically important that there be a one-to-one
correspondence between the dots that represent cities,
and the actual cities.  If, on any map, there are two
different cities named "Paris", there are nonetheless
*always* two different dots.

A *relationship type* whose instances are relationships
between published things, on the one hand, and
publishers, on the other, is absolutely not the same
thing as the *role* played by publishers in such
relationships, even if we describe or name both of them
using the same words.

> I can see this concept as being both a valid
> occurrence type and a valid end point of an
> association.

Yes.  The words

"having been responsible for the publishing of"

and

"publishedBy"

are both interpretable either way.  But precisely
because these two different interpretations are
*different*, they are necessarily *different subjects*.

> I see no point in having to create two topic
> elements, one for "the association of being
> responsible for publishing" and one for "the role of
> being responsible for publishing". I can understand,
> however, why two topic nodes might be needed for this
> purpose.

You seem to be arguing that 

  the act of translating a topic map document into a
  graph of nodes in which there is exactly one node per
  subject

should be nondeterministic.  A human being -- a
philosopher, perhaps, or perhaps a person capable of
reading the original topic map author's mind,
regardless of whether he is living or dead -- should be
used to make any such translation.

True, this approach would relieve the original author
of the burden of having to think clearly.  But I would
question the value and usefulness of any topic map
written by someone who thinks carelessly, or who isn't
careful to say exactly what he means.

> The problem I have with much of the work currently
> being done with topic maps is the implied conclusion
> that internal associations are more important than
> external relationships. This is clearly not the
> case. 

Internal relationships are what make it possible for
users to decide which external relationships -- i.e.,
which addressable subjects -- are interesting to them.
Saying that external relationships are more important
than internal relationships is like saying that oxygen
is more necessary for human life than air pressure is.
In order to breathe, we must have *both* oxygen *and*
air pressure.  In order to be useful, topic maps must
have both internal and external relationships.

The dRM has a basic and profound idea about the
distinction between what is internal and what is
external.  In the dRM, all *external* things are
addressable subjects.  Everything else is *internal*.
The only connections between the internal and the
external are the connections between topic nodes whose
subjects are addressable, and their addressable
subjects.

> In my opinion a topic map is only as useful as the
> external resources it leads users to.  Without being
> able to identify the use of topics in information
> resources a topic map cannot be used to enhance the
> semantic web.

Fortunately, nobody is planning to remove this
feature of Topic Maps.

> Sometimes the topics in a topic map do not allow you
> to formally define all the associations that a topic
> may need.

I don't understand how this can be so.  What prevents a
topic map author from saying whatever he wants to say?
How can an author's own <topic>s prevent him from
declaring any <association>s he wants to declare?

> In such cases you are forced to used occurrences to
> point to resources outside of the topic map to define
> associations.

I don't understand what you mean when you say that a
topic map author is "forced to use occurrences ... [in
order] to define associations."  Why are occurrences
necessary for this purpose?  (I'm so baffled by your
statement that I'm tempted to ask, "What is your
definition of the term, 'occurrence'?")

> The concept that you must "create a topic for any
> association you wish to express" is not one that
> appeals to me, nor one that makes long-term sense for
> topic maps.

I'm guessing that you're saying that you don't want
topic map authors to be forced to specify an explicit
assertion type <topic> for every <association>.  As far
as I know, nobody has proposed any such requirement,
for any syntax.

The dRM isn't intended to propose such a requirement at
the graph level, either.  (And it doesn't, unless we've
inadvertently made misleading statements in N0298R1.)

(I do think the question of whether unspecified
assertion types are implicit, vs. genuinely absent, is
one of the things that we should all discuss, if only
to be sure that we understand the implications.)

> We need to have well-maintained topic maps, with well
> defined associations, if we are to ensure that they
> will be usable over time. The current models do not
> seem to be helping this goal.

I don't understand why you say this.  I believe just
the opposite: that the dRM is, among other things, an
attempt to provide a firm foundation on which to define
durable sets of assertion types, and that the SAM is a
conscientious effort to define a set of durable
assertion types, among other things.

-- Steve

Steven R. Newcomb, Consultant
srn@coolheads.com

Coolheads Consulting
http://www.coolheads.com

voice: +1 972 359 8160
fax:   +1 972 359 0270

1527 Northaven Drive
Allen, Texas 75002-1648 USA