Re: Distributed querying on the semantic web

From: "Phil Dawes" <pdawes@users.sourceforge.net>
Subject: Re: Distributed querying on the semantic web
Date: Mon, 19 Apr 2004 22:03:05 +0100

> 
> Hi Peter,
> 
> From what I can gather, your concerns and objections are based around
> issues of trust and authority - neither of which I attempted to
> address in the original mail. I am purely attempting to address
> the problem of information discovery in the early stages of the SW.

Any scheme that requires (or even hints about something close to requires)
the use of the information in a particular document is going to be
inextricably tied to notions of authority.  Ignoring this aspect of any
such scheme would make the scheme a non-starter, at least for me.

> Answers inline:
> 
> Peter F. Patel-Schneider writes:
>  > 
>  > From: "Phil Dawes" <pdawes@users.sourceforge.net>
>  > Subject: Distributed querying on the semantic web
>  > Date: Mon, 19 Apr 2004 12:48:02 +0100
>  > 
>  > > Hi All,
>  > > 
>  > > I like Patrick Stickler's assertion that in order to participate in
>  > > the 'semantic web', http URIs should be dereferencable to some
>  > > information about the URI. 
>  > 
>  > I believe that you meant information about the referent (denotation, meaning,
>  > ...) of the URI.  If all that is available is information about the URI,
>  > then this is not very interesting, as I really don't need to know much
>  > about a URI.
> 
> Yes - sorry if that wasn't clear.
> 
>  > 
>  > However, I do hope that you did not mean necessary information about the
>  > referent (denotation, meaning, ...) of the URI.  I vigorously oppose any
>  > attempt to require that part of the meaning of a URI that my applications
>  > are supposed to abide by be the meaning that can be found in a document
>  > found by dereferencing the URI.  To pick my favourite example, I do not
>  > want my applications to be required to abide by the information available
>  > at http://www.whitehouse.gov just because I use the URI
>  > http://www.whitehouse.gov/#GeorgeWBush, *even* if this information is only
>  > something like
>  > 	http://www.whitehouse.gov/#GeorgeWBush rdf:type foaf:person .
>  > 
> 
> If your application is going to use information gleaned dynamically
> from the web, then you're going to have to deal with this somehow.

Well, yes, but I don't think that the scheme that you propose is workable
in general.  Why not, instead, use information from the document in which
the URI reference occured?  I would claim that this information is going to
be at least as appropriate as the information found by using your scheme.
(It may, indeed, be that the document in which the URI reference occurs
does point to the document that you would get to, perhaps by using an
owl:imports construct.  This is, to me, the usual way things would occur,
but I view it as extremely important to allow for other states of affairs.)

>  > > I am considering how an infrastructure
>  > > could be built where this could be exploited for distributed queries.
>  > > 
>  > > The main problem with Patrick's concise-bounded-description idea from
>  > > this respect is how to find references to a term. 
>  > > 
>  > > For example:
>  > > 
>  > > (p:PhilDawes, foaf:knows, ?person) 
>  > > 
>  > > ..is easy to resolve - just dereference p:PhilDawes and you probably
>  > > have the information you need. (I'm using dereference to mean 'look up
>  > > a description').
>  > 
>  > Well, I'm reluctant to ascribe any status to the information thus found
>  > that requires its use, and I certainly do not agree that it has to be the
>  > information you need.
>  > 
> 
> I'm not sure how you got the idea that it *has* to be the information
> you need - apologies if I gave that impression. 
> I am merely attempting to envisage a simple facility for getting
> *some* information about a URI when you have none. 
> At present this is impossible without using some previously-known
> centralised service.

Not at all.  OWL has what I consider to be a decent facility for this
purpose with its owl:imports construct.  Even using rdfs:isDefinedBy and
rdfs:seeAlso information would be more suitable, and more flexible, than
going to a single place for such information.  Even if there is no
owl:imports and no rdfs:isDefinedBy or rdfs:seeAlso pointers to follow, I
would not be in favour of just blindly going to a single place for
information about the meaning of a URI reference.

>  > > However
>  > > 
>  > > (?person, foaf:knows, p:PhilDawes)
>  > > 
>  > > .is much more tricky, since these assertions are likely to be made by
>  > > users external to the domain owner of p:PhilDawes.
>  > 
>  > Hmm.  I'm not sure of this.  For symmetric properties, it may be somewhat
>  > more likely for a document to put ``local'' URI references in the subject
>  > position, but what about properties that are conventionally written on way
>  > around.  For example, I am more likely to write on one of my web pages
>  > 
>  > 	sps:Sandy ex:loves pfps:Peter .
>  > 
>  > than I am to write
>  > 
>  > 	pfps:Peter ex:isLovedBy sps:Sandy .
>  > 
> 
> True. - That is a problem with 'consise bounded descriptions' as
> described by the URIQA[1] proposal - this statement wouldn't appear in
> the description for pfps:Peter.
> 
>  > > Here's a straw-man solution:
>  > > 
>  > > - In addition to its bounded description, dereferencing p:PhilDawes
>  > > also provides all the references it knows about.
>  > >
>  > > - When people author statements refering to p:PhilDawes, they POST
>  > > their triples to the description of p:PhilDawes. (Or maybe a third
>  > > party does).
>  > > 
>  > > - The representation of p:PhilDawes polls the reference URIs it knows
>  > > about periodically to keep its data up to date. (facilitating the
>  > > removal of triples as well as addition)
>  > 
>  > Independently of the authoritative status of the accessed web page I view
>  > this as extraordinarily dangerous.  There is no way that I would ever
>  > subscribe to a scheme that requires any server that I have control over to
>  > make responses that include
>  > 
>  > 	n666:antichrist owl:sameAs pfps:Peter .
>  > 
>  > just because some other organization has this triple in some RDF document.
>  > I don't see how any responsible organization would ever subscribe to this
>  > scheme, even if they could somehow tag these ``contributed'' triples as
>  > having come from some other document.
>  > 
> 
> That's fine. You own the domain (or at least the namespace) and so
> have complete control over which triples you return in response to a
> query. If people don't like what you return, they'll probably use a
> different term.
> 
> I suspect that for trust to work on any implementation of an
> open-world semantic web (centralised or decentralised), the authority
> of a statement will have to be decoupled from the location it was
> discovered. If that is the case, then it won't matter if you serve the
> statement or it comes from somewhere else.

Aaah, but it really does matter who serves a statement.  Perhaps not to the
model-theoretic semantics of the Semantic Web, but certainly the source of
information matters in the external social (and legal) world.  Ignoring the
fact that the Semantic Web is part of our imperfect and messy world is not
going to helpful for its widespread adoption.

>  > [...]
>  > 
>  > > I am keen to hear any ideas that others may have on the subject since
>  > > in addition to helping bootstrap the semantic web, this is a facility
>  > > that would be very beneficial in my work intranet environment.
>  > 
>  > I view this as a non-starter, even in a work intranet environment.  Just as
>  > for the Semantic Web as a whole, there is no expectation that such local
>  > environments will have a common view of the world.
> 
> Actually this already works to a certain extent (and adds value) in my
> work intranet environment. We have a facility (in use) that returns
> RDF contact information for an employee by dereferencing the HTTP URI
> denoting that employee (via a 301-see-other). N.B. This only works
> because the applications using the facility trust it.

> Without such a facility on the semantic web, I struggle to see how it
> will be bootstrapped to deal with open queries. At present, there is
> no real 'web' of information to search.

Well, I think that there already exists the mechanism to have a truely
decentralised web of information, with no central authorities or
information sources, namely the owl:imports construct.  It is not perfect,
but I think that it is better than trusting in the absence of a mechanism
for supporting a network of trust.

All this said, I have nothing particularly against imposing a notion of
authority or mandating trust relationships in particular situations.  If it
is indeed the case that contact information for employees is (normally, or
even, often) stored at particular locations, then it can be useful to
impose a trust relationship from outside the Semantic Web to get
applications to use this information.  I view this as a (partial)
failure of the goals of the Semantic Web, but it may be the case that this
is the best that can be done.

However, I would consider it to be much more in line with the goals of the
Semantic Web to instead have a document that explicitly points to these
employee documents to establish the trust relationship.  Even better would be
to have some mechanism to implicitly points to these employee documents,
but I do not believe that there are currently any mechanisms in the
Semantic Web for so doing.

> Cheers,
> 
> Phil
> 
> [1] http://sw.nokia.com/uriqa/URIQA.html

Peter F. Patel-Schneider
Bell Labs Research

Received on Tuesday, 20 April 2004 02:19:47 UTC