RE: Where are the semantics in the semantic Web?

> [Shi, Xuan]

> To answer your question "And what happened to the problems of logic(s)?",
> could you please tell me if SW people agree that there are many other
> "semantics" besides logic(s)? 

This is the problem with the way "semantics" is used in this
community.  People think we actually have something to say about
semantics.  But we'd be nuts to claim that.  The only sort of
semantics the SW depends on is exactly the sort of standards-based
approach you advocate.  Ultimately, a symbol means what everyone in
the relevant committee agrees it means, and what the rest of us take
their description of its meaning to be.  This mechanism isn't going to
account for the deeper sorts of meaning explored by Veltman in his (or
is it her?) paper, but so what?  We're trying to get the Web
infrastructure to help us do some fairly mundane tasks: ordering
stuff, getting information, arranging schedules, and such.  There are
all sorts of obstacles to achieving this, but agreeing on the meanings
of terms is not one of them.

When you ask '[are there] many other "semantics" besides logic(s)?',
the response is that logic is not semantics.  I thought you were
saying something similar when you suggested we call it the "logic
web"; I'd prefer "inferential web," but I agree with the idea.
"Semantic web" is catchy, but deeply misleading. All it means is that
web sites will behave as if they had a better grasp of what the
symbols on them mean than they seem to now. This is a purely informal
notion which just won't bear the weight you want to put on it. In
particular, no one (I hope) believes that the mechanisms to be used in
the SW will provide an adequate, let alone a novel or powerful, theory
of how meaning works for people.

> If this is true, then the problem of SW
> technology is it may ignore many other "semantics" except the logic(s). For
> example, "Forest" or "Swamp land" can be defined by different organizations.
> The meaning of "road" may be different in Europe from the same concept in
> USA. Even in GIS, a "road" is visualized using its center line, and you can
> see the problem as many roads have double lanes, or multiple lanes. So a
> single center line is problematic. What about a section of the road that
> shares multiple road names? How about you define an ontology of color, which
> can be defined by different ways, such as RGB, HSB, CMYK, or Hex code, or
> just natural name, then is it worthy to use logic(s) to matchmake the same
> color in different definition? If you would like to create such a color
> ontology using RDF/OWL, you may wish to give up due to its complexity and
> troublesome or unforseenable possiblities, such as "Aqua" and "Cyan" produce
> the same color but not all color defined by RGB, HSB, CMYK can get a name.

You must be kidding.  Aren't there standards committees settling
exactly these questions about colors?  Do they need a philosopher to
help them out by providing insights into meaning?

Is there a serious problem about classifying roads?  I find that very
hard to believe.  I grant that different groups are likely to arrive
at ontologies that carve up the world in ways that overlap at various
unforeseen points.  That's why I've done work in ontology
translation.  Automatic inference of the rules required to translate
between ontologies is a very difficult problem.  But the rules
themselves are not terribly complicated, and in particular seem to be
straightforwardly deductive.  (I'm thinking here of, say, a rule that
translates statements about roads from a North American ontology into
a European one.)  And in any case, organizing and applying translation
rules is a computational problem, not a semantic one.

> Or if you eventually create such an ontology of color class in RDF/OWL, how
> many people can understand it? 

I detect a slight wavering on your part.  You can't really think up an
infinite number of complications for the color ontology, so you toss
in this sour-grapes remark about what the result would look like when
all the complications were ironed out.  Actually, I'm pretty sure the
problems _have_ been worked out, and I'm pretty sure very few people
can understand the resulting documents.  We can use programs based on
them anyway.  A key SW idea is that inferences can be licensed by the
agreed meanings of the terms involved in those inferences.  The
inference machine doesn't have to have read or understood any
standards documents.

> It's a pretty AI game.

How pretty, exactly?

-- 

                                         -- Drew McDermott
                                            Yale University
                                            Computer Science Department

Received on Wednesday, 23 November 2005 23:25:21 UTC