RE: Denotation of URIs

> -----Original Message-----
> From: ext Peter F. Patel-Schneider 
> [mailto:pfps@research.bell-labs.com]
> Sent: 10 April, 2003 17:08
> To: Stickler Patrick (NMP/Tampere)
> Cc: www-rdf-interest@w3.org
> Subject: Re: Denotation of URIs
> 
> 
> From: <Patrick.Stickler@nokia.com>
> Subject: RE: Denotation of URIs
> Date: Thu, 10 Apr 2003 09:05:10 +0300
> 
> > > -----Original Message-----
> > > From: ext Peter F. Patel-Schneider 
> > > Sent: 09 April, 2003 16:21
> > > To: Stickler Patrick (NMP/Tampere)
> > > Cc: www-rdf-interest@w3.org
> > > Subject: Re: Denotation of URIs
> > > 
> > > From: <Patrick.Stickler@nokia.com>
> > > Subject: RE: Denotation of URIs
> > > Date: Wed, 9 Apr 2003 09:33:31 +0300
> 
> [...]
> 
> > > HTML doesn't say that all documents that are non-valid HTML 
> > > document are
> > > supposed to cause a core dump; it is just silent how such 
> > > documents are to
> > > be treated, and many browsers do interesting things.
> > 
> > Oh come now. HTML says what is or is not valid. What is or is not
> > correct. And yes, most browsers exhibit the desirable characteristic
> > of graceful degredation in the presence of invalid input, which is
> > a desirable characteristic for any software application.
> > 
> > But invalid HTML is invalid HTML and is *wrong* even if a browser
> > is able to do something with it.
> 
> Huh?  What part of the HTML specification does such a browser violate?

Please re-read what I just said. I said nothing about a browser violating
anything. I said that an invalid HTML *instance* is wrong.

> > Ambiguous denotation of URIs is *bad* even if SW agents are able
> > to still function when it exists.
> 
> > > Let's then agree that your version of the Semantic Web is 
> > > only concerned
> > > with situations where denotation of URI references is fixed 
> > > and universal.
> > 
> > I consider this to be the only reasonable version of the 
> Semantic Web.
> > Not just "my" version of the Semantic Web.
> 
> Well I disagree.

Fair enough. I still wait to see even one convincing use case that
shows any benefit whatsoever from ambiguity in the denotation of URIs.

Until then, I guess we can just agree to disagree and not expend
any more energy on this debate.

> > I consider any version of a Semantic Web which supports ambiguity
> > of denotation and does not consider it an error to be ill defined
> > and far less useful for the reliable communication of knowledge
> > between arbitrary systems on a global basis.
> > 
> > > I would then support you in any effort to state that the 
> > > denotation of URI
> > > references is fixed and universal given this assumption.  
> > 
> > Not that is *is*, but that it is *presumed* and that if in fact
> > it is determined that there is ambiguity and disagreement about
> > the denotation of a given URI, that that is a problem, a but, an
> > error, and is bad and should be resolved -- by whatever means
> > (social, legal, technical, whatever) might be useful to do so.
> 
> Why?  People get along quite fine without this.

If you are speaking about natural language, interpreted by humans,
then I see no point in responding to that comment, as it is completely
out of scope and irrelevant to the SW architecture, which is not
based on natural language nor intended to be interpreted only by
humans.

> > I've never said (and have several times reiterated this) that
> > one can guaruntee that all URIs have unambiguous global denotation
> > (yet you seem to continue to harp on this point).
> > 
> > It is one thing to say that the SW *presumes* unambiguous global
> > denotation of URIs yet another thing entirely to say that the SW
> > *guaruntees* unambiguous global denotation of URIs. I assert the
> > former. I have never asserted the latter. Please take note of this.
> 
> I don't see the difference here.  If (the formal machinery of) the SW
> presumes that that URIs have unambiguous global denotation 
> then tools are
> going to have this presumption built into them.  They will 
> break if the
> presumption is violated.  

Not necessarily break. No more than an HTML browser "breaks" if it
encounters invalid HTML.

SW agents should be able to presume that URIs have unambigous denotation,
and yes, if they don't detect any ambiguity there may very well be 
unreliable or unexpected inferences made, but if ambiguity is detected,
SW agents may have facilities for at least notifying about and possible
accommodating (in various ways) that ambiguity.

But they can presume unambiguous denotation. They must.

> > > Feel free to work
> > > under this assumption provided that you don't eliminate the 
> > > possibility of
> > > other visions of the Semantic Web that lift this assumption, 
> > > which is what
> > > you appear to have been attempting to do.
> > 
> > Absolutely. I consider it harmful to the SW architecture to
> > posit that it is in any way *acceptable* to have ambiguity in
> > denotation. It is IMO *never* acceptable to have such ambiguity,
> > and in any case where there is ambiguity of denotation, it will
> > be harmful to the SW.
> 
> Well, this is your opinion.  I disagree.

Again, I wait for a single example to prove my viewpoint wrong.

The example, derived from your own example

> ... if the URI being used to denote who the first lady
> is, is being used ambiguously, then there is no way to actually
> test for any disagreement. I.e., I may assert that _:a denotes
> Jane and she lives at address X, but that _:b denotes Betty and
> she lives at address Y. If you use _:a to denote Betty and assert
> that Betty lives at address X, I will see no contradiction, because
> according to my understanding of _:a you are saying that Jane lives
> at address X!

IMO proves my viewpoint to be right.

> > > > The same is true for SW agents. Ambiguity in denotation 
> > > should be clearly
> > > > and strongly identified as bad, wrong, detrimental, 
> etc. yet the SW 
> > > > architecture can still have machinery to help identify 
> any occurring
> > > > ambiguity and deal with it productively.
> > > 
> > > Drawing a line around the part of the Semantic Web that needs this
> > > assumption is fine.  When working in this part of the 
> > > Semantic Web it is
> > > also fine to try to identify parts of the Semantic Web that 
> > > don't quite
> > > satisfy the assumption and try to modify them as necessary to 
> > > satisfy the
> > > assumption. 
> > 
> > I just don't see how it is possible to reconcile the two views
> > that (a) ambiguity of denotation is bad and should be avoided,
> > eliminated, fixed and (b) ambiguity of denotation is fine and
> > acceptable.
> 
> Well I don't need to reconcile the two views because I don't 
> believe that
> the first view is correct.

Fair enough. But I do, so I don't see these as reconcilable.

> > I understand that some folks may wish to be able to explore
> > various means to deal with ambiguity, and that's great, since
> > I'm quite sure that we will face ambiguity of denotation of the
> > SW and it will be wonderful to have effective means of dealing
> > with it.
> 
> Great.  Then let me get on with it.

Great, I welcome my viewpoint being disproved by running code and
applications with clear utility.

> > > > Being able to deal with an error does not validate the 
> > > error and make it OK.
> > > 
> > > Well, sure, if you view it as an error.  However, others may 
> > > disagree, and
> > > view it as a source of power.
> > 
> > Well, I'd be quite surprised if a majority of folks would not
> > consider it an error.
> 
> Sure, a majority of folks may want to work under this 
> assumption.  I don't
> mind this at all.  I don't even mind this being a blessed 
> assumption for
> parts of the Semantic Web.  Just don't make it be part of the formal
> machinery of all of the Semantic Web.

Well, I honestly am very much in favor of minimal constraints
and saying less than more in specs, and happy to see everyone
accomodated, but as I don't see these two views as reconcilable,
the only way to accomodate both seems to have two semantic webs,
which I don't think anyone wants.

> > And I've yet to see any use case, from you or elsewhere, which
> > demonstrates even marginal benefit from the existence of
> > ambiguity in URI denotation.
> 
> What about my ``first lady'' example?

What about my counter example? Repeated above?

If there is no agreement about denotation of the atomic primitives
of the shared, global system (which the SW is) then there is no
basis for concluding anything about agreement or disagreement. The
RDF statements become nothing but decorative and lose any utility
for communication.

> > I've provided several examples showing how it is detrimental
> > to the SW and could easily provide many more.
> 
> Sure, denotation ambiguity is difficult to deal with in many 
> case.  I don't
> deny that.  If tools are built based on this assumption, then 
> of course
> violating it will be detrimental to their functioning.  
> However, limited
> ambiguity can be handled, as people do it all the time.

*People* do it all the time, using natural language and very large
brains capable of complex reasoning based on a broad base of world
knowledge and experience.

But SW agents will *not* be able to deal with ambiguity the way
that humans do.

Please stop comparing humans + natural language to SW agents + RDF.
It's apples and oranges. No, it's apples and orange *groves*.

Again, even if some *particular* SW agents are able to deal with
some degree of ambiguity (a) it will likely be far less than humans
are able to deal with (b) it will still make results suspect and
(c) most SW agents, for a long time to come, will not be able to
deal with ambiguity and should not be forced to.

If researchers want to grapple with the problem of ambiguity of
denotation in knowledge bases, fine, but ambiguity should not be
considered an acceptable characteristic of the SW architecture and
should be guarded against, flagged, and fixed always.

> > Anyone who wishes to argue that the SW architecture should "bless"
> > ambiguity in URI denotation is going to have to back it up with
> > clear and motivating evidence that it is beneficial.
> 
> Well, what is the vision of the Semantic Web?  One vision, 
> that used to be
> in the RDF documents, is that the Semantic Web, even in the 
> form of RDF, is
> supposed to be able to ``say anything about anything''.  If 
> it is not even
> possible to have divergence of denotation, then it sure won't 
> be possible
> to talk about it.

Not so. I've given examples of how you can talk about disagreement
of denotation, by reifying the URIs. Though of course, in order to
discuss the denotation of certain URIs you need *other* URIs which
denote the resources in question which are *not* ambiguous, otherwise
your assertions about what the first URIs should or should not denote
will be meaningless.

Ambiguity of denotation on the SW will *always* be detrimental.

> > I wait to see that evidence...
> > 
> > > > Ambiguity in denotation will always be a bad thing. Always. 
> > > Even if SW agents
> > > > can deal with it trivially (which I doubt they will be able 
> > > to do short of
> > > > cognitive abilities).
> > > 
> > > Ambiguity in denotation is a fact of communication.  Always. 
> > 
> > Have I ever said anything in any of my posts that indicated I
> > think otherwise? No.
> 
> No, but you sure seem to be trying to eliminate it from 
> communication in
> the Semantic Web.

Not at all. I am just opposed to "blessing" it as OK, which is what
you seem to be wanting to do.

Saying that something is bad or wrong or an error does not mean it
won't occur. But admitting that it does or can occur does not mean
that one is accepting it as OK.

Ambiguity of URI denotation on the SW will happen, but will always be bad.

> > When I say that "ambiguity in URI denotation is bad" you seem to
> > be interpreting that as meaning "ambiguity in URI denotation is
> > impossible". Please feel free to interpret what I say literally.
> 
> I'm not saying that you said that "ambiguity in URI denotation is
> impossible".  I am, instead, saying that ambiguity is a fact of
> communication, and that therefore the Semantic Web will have 
> to deal with
> it.

But there are different ways of dealing with it. One can presume
unambiguous denotation and when one gets undesirable/unreliable
results from a SW reasoner, one can identify the source of the
the problem and either correct it or exclude that source from 
ones reasoning due to being unreliable/untrusted.

The SW agent itself need not have any other presumption but that
URIs have unambigous denotation.

> > Again, being able to deal with *problems* does not validate
> > those problems and make them OK.
> >  
> > > SW agents
> > > will eventually have to deal with it (and should be able to 
> > > deal with some
> > > ambiguity in denotation without full cognitive abilitities).   
> > > 
> > > It may be that some cognitive-like abilities are needed to 
> > > deal with any
> > > sort of ambiguity in denotation, but so what?  Aren't SW 
> > > agents supposed to
> > > have cognitive-like abilities?
> > 
> > They might, but need not. And I think it will be some time before
> > they will be able to handle ambiguity of URI denotation effectively,
> > which is why it must be a fundamental assumption of the SW 
> architecture
> > that SW agents may *presume* that URIs have unambiguous denotation.
> 
> Then the Semantic Web will have to be replaced by something 
> else in the
> not-too-distant future.  I would be much happier if this 
> wasn't necessary,
> and that ambiguity of denotation would not be ruled out of all of the
> Semantic Web.

Well, I think that a presumption of unambigous denotation of URIs will
be a key to the success of the SW, not a hinderance to its adoption and
evolution.

I guess time will tell which position is correct.

> > Any ability of SW agents for dealing with ambiguity will ultimately
> > serve the purpose of identifying and correcting such ambiguity, IMO,
> > not encouraging further ambiguity on the SW. It will be a long, long
> > time (decades, I think) before ambiguity in URI denotation would
> > become innocuous on the SW.
> 
> I don't think that ambiguity in denotation will ever be 
> innocuous.  I think
> that it will always be difficult to deal with.  That is, 
> however, no reason
> to rule it completely out.

That statement simply scares me. If something is always bad, and we *can*
rule it out, then we should.

> > So I'll assert again (for the umpteenth time) just because 
> a SW agent
> > can deal with noise does not mean that noise is good or beneficial
> > or correct or acceptable.
> 
> Sure, but that doesn' t mean that ambiguity in denotation should be
> forbidden.

Yes, it does.

> > Perhaps I'm misunderstanding you, but you seem opposed to the idea
> > of saying that ambiguity in URI denotation is a bad thing. You seem
> > to be asserting that it's OK. Feel free to correct me if I've not
> > understood your position corectly.
> 
> A am opposed to saying that ambiguity in URI denotation is a 
> bad thing, at
> least for the Semantic Web in general.  I'm not opposed to 
> you building
> applications that view ambiguity in URi denotation is a  bad 
> thing, just
> don't foist that view on my applications.

Then I think that you and I will not be using the same semantic web.

Patrick

--
Patrick Stickler, Nokia/Finland, (+358 40) 801 9690, patrick.stickler@nokia.com

Received on Friday, 11 April 2003 02:24:04 UTC