Re: Proposed issue: What does using an URI require of me and my software?

At 05:05 24/09/03 -0400, Bijan Parsia wrote:
>On Tuesday, September 23, 2003, at 04:07  PM, Graham Klyne wrote:
>
>>At 14:49 23/09/03 -0400, Bijan Parsia wrote:
>>>---- Original message ----
>>> >At 11:00 23/09/03 -0400, Bijan Parsia wrote:
>>> >>The main line is "that use of a URI in RDF implies a commitment to
>>>its
>>> >>ontology".
>>
>>[...]
>>
>>> >I originally read that statement loosely as meaning something like
>>>"Using
>>> >the URI in RDF is to make statements about what the owner of the URI
>>>claims
>>> >it to identify".
>>>
>>>See Pat's claims that this kind of separation doesn't work for
>>>concepts and
>>>relations.
>>
>>If you're referring to Pat's earlier posting to this list, I didn't
>>observe any such claim.   (Quite possibly my oversight.)
>
>From
>http://lists.w3.org/Archives/Public/public-sw-meaning/2003Sep/0047.html:

[...]

Ah, OK, I didn't read it the same way.  But rather than dwell on that, I'll 
cut to the immediate debate...

>>My perspective is that there are, quite clearly in my mind, (at least)
>>two kinds of "meaning" associated with RDF, or indeed any descriptive
>>data:
>>
>>(1) that which is defined in formal terms, and which can be subjected
>>to formal analysis. ("formal meaning")
>>
>>(2) that which exists by virtue of consensus between people. ("social
>>meaning")
>
>Uh...And you do think that 1 is a subset of 2, yes? Or what has RDF
>Core been doing?

To be a subset suggests to me that they are in some sense the same kind of 
thing.  My view is that they're quite different.

What I would say is that the formal meaning attempts to capture aspects of 
the social meaning to the extent that systems that behave solely according 
to formal rules do so in a way that is consistent with peoples expectations.

>Perhaps worse as i ordinarily think of meanings as getting
>*formalized*. Thus, all meaning is subject to (more or less successful)
>formal analysis.

That would appear to be a point of divergence between us.

>>If one takes a purely formalistic stance, then the second kind is no
>>meaning at all.
>
>I think this is on a wrong track.
>
>>  But surely there's more to life, the web and everything than
>>formalism?  In particular...
>>
>>>Plus, your reading seems pointless. Operationally, what does it
>>>affect?
>>
>>... it affects the ways that people might write software to process
>>RDF containing said URI.
>
>Yes, that's what I'm after. But I want to know what those ways are. I
>wouldn't mind formalizing them, if at all possible. owl:imports (at
>least according to pat) isn't formalized (or even semantically clear).
>But it specifies SOMETHING. And it makes a difference when writing owl
>reasoners...believe me.

OK, so here we have some common ground.

When a programmer writes a program based (only) on an English (or other 
human language) description, I'd say that the *programmer* is creating an 
automaton that complies with some aspects of the social meaning.  But, at 
this level, the human translation is crucial.

I think I'm also agreeing with you when I say that as out formalisms get 
better/more complete, we capture more of the social meaning (or can create 
formally specified automata that are compliant with a broader range of 
behaviour that a human might consent to be consistent with the social meaning).

But until we have software that understands human languages as well as the 
people that speak them (and I'm not holding my breath), I assert there will 
always be some aspects of the social meaning that are opaque to formal 
reasoning.  As far as our Semantic Web software goes, I see that as meaning 
that we'll continue to depend on programmers writing special-purpose 
software to deal with the social meaning of particular vocabularies in a 
specific contexts.

>>  (Does writing a program to process some URI in RDF is a particular
>>way give some added formal meaning to the URI?  Whatever the answer, I
>>don't see that it helps us to formally reason more thoroughly about
>>the thing referenced.)
>
>I'm really not sure what it *is* to "reason formally".
>
>But it still seems red herringland. Right now, I want to know if I have
>to reason over the transitive closure of the set of documents
>retrievable by the set of URIs in some RDF document, rather than over
>the statements in the RDF document alone, or over some specifiable
>subset of that transitive closure. And then, there's the freshness of
>the closure.

I suppose that's an option, somewhat orthogonal to the preceding 
discussion.  As far as formal treatment of (just) RDF is concerned, I'd say 
the answer is "no", there's no further documents you have to 
retrieve.  I'vr seen some debate about the role of owl:imports, but don't 
personally have an opinion about that.  I'm sure we could (and maybe will?) 
end up with some kind of specification that describes how additional web 
retrievals can be performed to obtain more information to refine the formal 
meaning of some RDF+(OWL/whatever) data.  But while these may increase the 
reach and utility of SWeb information, I don't see this as addressing every 
situation.  At some stage, I think we'll still end up with aspects of 
meaning that are described by imformal language, and maybe processed using 
ad hoc software developments that embody the programmers understanding of 
that language.

>It seems to me that this is the kind of thing you can put in a spec.
>"Yes, grab that closure." Yes, it's a good idea, but you don't have
>to." "No, are you nuts?" "No,not for mere use, but if you have an
>owl:imports, then yes"
>
>Well, ok, they'll go into an unhappy spec since there's a ton more to
>figure out.

Yup (about the "ton more ...")


>>I perceive this debate arose because there is more meaning in RDF
>>expressions than is captured by the formal semantics, but we couldn't
>>agree how to describe the nature of that meaning.
>
>Well, I thought it arose because Tim, specifically, though that there
>was *less* such meaning :) The fear of meaninglessness!

Er, I don't follow that.

>>One might take a view that if can't be formalized we've no business at
>>all trying to describe this "social meaning".  I don't subscribe to
>>that view.  If we don't try to describe the meaning that can't be
>>formalized, then we have no basis to judge whether a program that
>>operates on certain data in a certain way is behaving as we humans
>>believe it should do.
>
>Describing is a kind of formalizing. Specifying is a kind of
>formalizing.

I don't see it quite that way.  To my view, it would require that the 
describing/specifying is itself done in some formal language.  By which, I 
mean a language with well-founded inference rules, etc., like logic, RDF, 
Owl, or whatever.

>>Also, an informal description may often, I believe, be a precursor to
>>formal definitions that capture more (but never all) of the intended
>>"social meaning".
>
>Since I've not provided a formal analysis of owl:imports, consider my
>presentation to be that precursor :)

Sure!

>>> > By this, using a URI *formally* means nothing.
>>>
>>>My understanding, both from looking at the text (see my earlier
>>>reading)
>>>and from conversation, is that this is not the case.
>>>
>>>I don't see how to read "committment to an ontology" in this context
>>>that
>>>*doesn't* have an effect on the formal meaning.
>>
>>Well, maybe.  I don't want to get hung up on that text.  I don't think
>>we're here to create new aspects of formal meaning.  We've got two
>>working groups doing that already.
>
>Well, that, itself, is at issue. I don't see how to read tim's issue
>statement otherwise. *I'm* not the one to be convinced! But if we don't
>beat up on the concrete proposals, especially the ones with status, we
>risk doing nothing at all. It seems to me Tim cares about his view.

I'm not going to try and tell folks here what Tim meant.  I'm describing my 
response to it, which I grant may not be supported by an in-depth analysis 
of the words used.  (But I know I find it infernally difficult to find the 
right words to pin down a difficult concept in ways that others can't 
misinterpret, so I tend to allow that others do too.)

>>>I mean, really, what does this "identify what the URI claims it
>>>identifies"
>>>*mean*? Is it even true?
>>
>>Er, I don't recognize that statement.  What I said was "Using the URI
>>in RDF is to make statements about what the owner of the URI claims it
>>to identify", which explicitly appeals to a human as opposed to a
>>formal judgement of meaning.  Maybe you refer to something else?
>
>Sorry, I meant yours.
>
>But I still deny your distinction, between appeals to human vs. formal
>judgement of meaning. I don't know what the latter is.

By the latter, I mean new assertions that may be derived from available 
data by purely formal (mechanical) reasoning processes;  deductions that 
can be supported by some proof.

>I want to know the *practical* consequences of this notion of meaning.
>Specifically, that URIs have a meaning and that meaning is "under the
>control of the URI owner" and that I (document author or RDF software
>writer) should, in some sense *respect* that meaning.

I'm not sure which notion of meaning you're questioning here.  When it 
somes to running software, I come back to the difference between a 
programmer interpreting an informal specification and a deduction that can 
be supported by proof.

>What must I do to exhibit that respect. What does web/semantic web
>architecture require of me? (Nothing, imho, but others disagree.) What
>do we intend (as a community) to *specify* one must do to so respect
>that meaning.

For social meaning, I don't think there's a lot we can specify.  But we can 
still note that there are a range of social mechanisms and procedures that 
will permit consensus to be formed around deductions that are not supported 
by any formal proof.

And insisting on conclusions that are not supported by such consensus might 
be regarded as disrespect for the social meaning.

>Tim claims that we're engages in prescription rather than description.
>I want to know what I'm going to be told to *do*. If we're just
>theorizing about how the web or semantic web actually works, well,
>that's a different task. And, I say this as a philosopher, who cares?
>Why is the W3C spending money on this, or our member organizations that
>aren't universities? The W3C probably isn't THAT sort of research
>entity (nor does it seem especially well equipped to do this sort of
>research).

<Nods/>

I suspect there's only a very limited amount we can do about social 
meaning, other than to acknowledge it's existence and maybe put together a 
characterization that helps people to understand that not all useful 
information can be proven.  I absolutely don't want to spend the next 6+ 
months arguing this ... at the end of the day, I (and others, I suspect) 
will go write software and make stuff happen.  If and when there's 
observable harm then maybe we'll review what we're doing.  Meanwhile, maybe 
we can spot some of the pitfalls before we fall in;  if this discussion 
alerts me to any then that would be a useful outcome.  And maybe we can 
alert some other folks too.

#g


------------
Graham Klyne
GK@NineByNine.org

Received on Wednesday, 24 September 2003 16:14:02 UTC