RE: Where are the semantics in the semantic Web?

Quoting Jim Hendler <hendler@cs.umd.edu>:

> So if you said
> http://ww.cs.umd.edu/hendler/peopleont#firstname then there could be
> a document at that URI with some text (for people) or RDFS/OWL (for
> machines) that helped explain your use of the referent.  Even if
> there isn't, if you and I agreed to use that term in a certain way,
> we wouldn't have to convince anyone else to agree with us - because
> someone using a service that took this input would never confuse it
> as being equal (as it would be in English) to some other
> http://foo.bar.baz/services/servont#firstname.   In you Xlink-like
> scheme, there is no way to distinguish different users using the same
> string, so we need to get the impossible global agreement.  When we
> use URIs at least that aspect of the problem goes away

But XML will typically use namespaces.  It's not as if everyone
who decides to use "firstname" elements much clash with everyone else.

Of course, it is useful to be able to treat an element name as
a URL and look up something machine-readable about it.  But
it's not only RDF/OWL that let you do that; it can also be done
with XML schemas; and in many cases, schemas can be quite
straightforwardly read as class definitions.

There are problems with that approach, and I am no fan of
XML schemas; but the current approach to the semantic web
is not perfect either, and I am not convinced there is as
strong a syntax-vs-semantics difference between them as
often seems to be suggested.

In one of your slides, you say "The semantic web exists -- don't
reinvent it".  But computing and other fields often make progress
by developing alternatives to existing approaches.  People didn't
decide "we have a programming language; let's not reinvent that
by designing a different one".

I am certainly not saying that the current RDF/OWL/OWL-S approach
is not worth pursuing, and I like quite a few things about it;
but I'd also like to see some alternatives pursued as well.

>   ... if you want to try to get a
> service language that works by natural language agreement, instead of
> something more formal, all power to you -- I think it is a doomed
> enterprise, but I could be wrong

But a lot of natural language agreement is also needed when
RDF etc are used.  If I want to know what various Dublin Core
terms mean, for example, it's some English-language text that
tells me.  That's how I can tell whether I should use DC "creator"
or not.

-- Jeff

Received on Thursday, 24 November 2005 17:51:08 UTC