Re: No Standard Semantic Web Pragmatics?

From Peter F. Patel-Schneider:
> In any case, there are indeed already more-or-less standard ways of doing
> much of what you appear to want.

I'm inclined to agree.

> From: "John Black" <JohnBlack@deltek.com>
> Subject: No Standard Semantic Web Pragmatics?
> Date: Thu, 10 Jun 2004 18:17:54 -0400
> 
> > So here is what it looks like to me.
> > 
> > A general purpose communication system, where:
> > 
> > There is no standard way to tell who is making statements.
> 
> How about the ``owner'' of the web page, just like for web pages that are
> HTML or XML documents?

I prefer to think the web page itself says things, and the owner bears
*some* responsibility for the things said, but does not literally say
them.

So perhaps "http://www.ibm.com" says some product is priced at $50.  I
take that as a good indication that IBM Corp. will sell the product
for $50, and I might say to a human that IBM says the price is $50,
but if I were being very literal minded I'd rephrase it as "IBM's
website says the price is $50."  That distinction is of course
necessary in talking about the website being defaced.  "IBM's website
was defaced!  Now it says the product costs $0.01!".  I couldn't
express something like that if the registered owner was always taken
as saying what the website said.

And I don't think it's a weakness to have little formal tie back to
humans or owning corporations.  Most of what I imagine doing on the
web (semantic or not) involves reasoning about what pages say, not
about what the people behind the pages say.

> Again, why not the usual situation in the rest of the web?  The contents of
> Semantic Web web pages are asserted, just like other web pages.  Yes, no
> Semantic Web language currently allows for denying, or quoting, or other
> relationships between agents and propositions, but this may change.
> 
> > Its a new, artificial language but there is no standard way for 
> > fixing or learning the intended interpretation of its terms, URIrefs.
> 
> Well determining (all of) an ``intended'' interpretation is generally not
> possible.  All that a formal language can do is provide information that
> can be used to narrow down the formal interpretation of a term.  There
> already is a standard mechanism (owl:imports) that allows for the inclusion
> of the meaning of one Semantic Web document (written in OWL) in another.
> What more is needed at this point?
> 
> > URIrefs, most of which look just like URLs, are to be treated as 
> > strings bearing no standard relation to the URLs they look 
> > like, or to anything that might be done with them on the web.
> 
> This is not an easy problem to solve.  The current treatment of the
> denotation of URI references at least has the desirable feature that it
> does not rule out future solutions to this problem.

While the RDF and OWL Recs don't require it, I think the best practice
is to take all the RDF content obtained by dereferencing a URI used in
RDF and treat it at a level of trust very close to the level of the
URI's user.  Very much like you take one website's linking to another
as a probable-endorsement.  There's still a judgement call involved
that I don't know how machine can help with yet.  (One idea I'm toying
with is that the endorsement is much stronger if the linking page has
been modified since the linked-to page was.)

This is kind an intermediate position between the position I once
advocated (that using a URI impies you believe the content) and what I
think may be Peter's position (that there's no connection).

> > You can reason over it, but everything stated is considered true, 
> > and there is no standard way for anything to be unsaid. 
> 
> I'm of the opinion that removing information from a Semantic Web web page
> ``unsays'' it.

Yep.  Note that web pages have expiration times.  Thing they state
are, I think, asserted for the time range starting at the Last
Modified date (or at the time the content was provided, if no LM date
is given) and ending at the Expiration date.  Expiration dates are
often set in relative terms ("cache-control: max-age=600"), so they
remain asserted until a little while after you change the document.
If the expiration date is in the past (which is alas very common),
then IMHO the data is not quite asserted, and one has to make a "no
data change means no world change" assumption -- that the data on a
slightly-old web page is *probably* still true.  This is another
judgement call humans make (often incorrectly) every day using the
normal web.

> > And the people that would need to be involved to develop some 
> > plain old-fashioned standard language pragmatics[1,2] are either 
> > firmly against it or are too busy out writing code with it to 
> > bother.
> > 
> > Do I understand this correctly?
> 
> Not to my mind.  I've been arguing very hard for a (particular) notion of
> Semantic Web pragmatics that I think is very pragmatically motivated.
> 
> Sure, there are lots of things that one might like to be able to do in the
> Semantic Web that are not currently possible.  I don't expect this to
> change (at least for quite some time) - formal languages and programs are
> currently not nearly as capable as human beings and there are other issues
> to be considered (such as symbol grounding).
> 
> As well, the work on Named Graphs (currently available at
> http://www.wiwiss.fu-berlin.de/suhl/bizer/SWTSGuide/carroll-ISWC2004.pdf)
> appears to me to fit right into what you want.  
> 
> > [1] The Semantics-Pragmatics Distinction 
> > http://userwww.sfsu.edu/~kbach/semprag.html
> > [2] Pragmatics of the Semantic Web
> > http://semanticweb2002.aifb.uni-karlsruhe.de/proceedings/Position/kim.pdf
> > 
> > 
> > John Black
> Peter F. Patel-Schneider

-- sandro

Received on Friday, 11 June 2004 08:40:12 UTC