Hey Pat, Pat, Jim & All: (01)
Great discussion, thanks for the opportunity to offer my perspectives on
knowledge representation, standards and scalability. (02)
First scalabity: When I think about the scalability of knowledge, I
don't think about compute cycles. I think about the ability of our
models to approximate the world. When our models fail to approximate the
world, our knowledge doesn't scale. Of course any one approximation only
serves to reveal what has been eliminated from the model, to bring
clarity to what the model hopes to explain. (03)
Or, in the words of Alfred North Whitehead: "Systems, scientific and
philosophic come and go. Each method of limited understanding is
exhausted. In its prime each system is a triumphant success; in its
decay an obstructive nuisance." (04)
Around these issues, you'll find me in the camp of complex adaptive
systems or what Christopher Alexander calls Living Structure. John Sowa
provides a useful diagram that I reflect on quite often. You can find it
here in figure 12 (05)
http://www.jfsowa.com/ontology/causal.htm (06)
I'd suggest the debate regarding whether OWL or other languages as KR
standards is reframed in a broader context. I'm sure you didn't want to
hear that, but I think its important to reframe standardization in a
context that supports living structure. In addition to the models and
theories defined in John's diagram, I suggest we add languages and
logics to a framework around which discuss knowledge standardization. (07)
I'm not saying think this framework will give us a triumphant success,
but it will take us one more step beyond an obstructive nuisance. ;-) (08)
Again, many thanks for considering my thoughts on this important paper. (09)
--
Best wishes, (010)
Rick (011)
email: rick@xxxxxxxxxxxxxx
web: http://www.rickmurphy.org
cell: 703-201-9129 (012)
Pat Hayes wrote:
>>The reference mechanism has some merit, but very limited. It's useful
>>if the referenced concepts are very well and unambiguously defined, and
>>those who use it read and understand the documentation properly.
>
>
> True, you do have to actually read the ontologies
> you are referring to. But the process most
> emphatically does not require unambiguous
> definitions of concepts. On the contrary, in
> fact: it seems to work best with comparatively
> weakly defined ones. First, we should be clear:
> there is no such thing as a *definition* in OWL
> (or CL or IKL) in the strict sense. There are
> only assertions. Some of the most widely used
> 'concepts' are those about which the base
> ontology says virtually nothing, such as
> dc:author from the Dublin Core. This has been
> used so widely now that its meaning is in fact
> defined more by its uses than anything any
> ontology says about it. The DC authors weren't
> thinking about authorship of emails, but there
> are now tens of thousands of websites which use
> it in this sense. So, does it mean that? I would
> say, yes.
>
> As for the rest of your issues, they are all
> possible, of course. But to me they have the air
> of 'the sky might be falling'. Of course
> misunderstandings and mis-uses will occur: but
> the normal socio-economic forces of the Web
> (especially the commercial Web) all act so as to
> quickly locate, correct, minimize or isolate such
> mis-uses: and that is what will in fact occur.
> The SWeb will never be perfect, just as the real
> Web isn't. But it will be quickly
> self-correcting, and it will be always useful.
>
>
>>But:
>> (1) Two different ontologists may mean exactly the same thing and
>> point to two different URI's - which may or may not be
>> intended to mean the same thing -- but the relation of the
>> two concept will be indeterminate.
>
>
> Why? If someone points to them both and says they
> are the same, and if this assertion is widely
> reiterated or widely depended on, then they are
> the same concept, whether or not the original
> authors thought they were. More usefully, if one
> is used by one subcommunity and the other by
> another, and these subcommunities come to use
> them with slightly different meanings (say,
> 'person' in the legal and biological senses) then
> indeed, the two meanings will evolve apart,
> usefully for each subcommunity; but also in their
> history recording their common lineage.
>
>
>> (2) two different ontologists may interpret a URI differently and
>>point to it
>> though they mean different things -- this can easily happen with
>>the
>> very poor documentation typical of on-line knowledge
>>classifications
>
>
> Yes, it can, but again there are only pressures
> to correct such mutual divergences of opinion (if
> they matter, ie if they give rise to unwanted
> contradictions; and if they do not, then this
> situation is harmless, in spite of the intentions
> of the ontologists)
>
>
>> (3) in an ontology, there may be a pointer to a URI in some online
>>ontology for
>> concept A, and a pointer to a URI in another ontology for concept
>>B,
>> and in the referring ontology, concept B may be defined as a
>>subclass of
>> concept A, though in reality the original B and A may have little
>>in common.
>
>
> What is 'reality' here? Are you saying that the
> referring ontology (call it C) is simply wrong?
> Well, yes, ontologies may have errors in them, I
> see no way to legislate against that possibility.
> But if you mean only that C disagrees with what
> the A and B authors had in mind, then I think
> here we simply have divergences of opinion about
> the facts: again, nothing can be done to prevent
> such disagreements. Indeed, they can often be a
> valuable source of information about the sources
> of the ontologies themselves.
>
>
>> The act of referring to some URI will only be as useful as the
>>definition
>>attached to that URI is unambiguous.
>
>
> I PROFOUNDLY disagree. Quite the opposite. The
> very idea of unambiguous definitions is anathema
> to good ontology engineering, IMO.
>
>
>> Ambiguity abounds, and
>>indeterminacy of
>>the relations of concepts in different ontologies is very high. When
>>we think that concepts are clear, it may just be because we are
>>disambiguating in context in a way that no machine can do with
>>existing programs.
>
>
> Quite. So we should revel in ambiguity (which
> would be better described as under-specification)
> rather than try to prevent or avoid it.
>
>
>> A common foundation ontology ('upper ontology' is a misleading
>>metaphor,
>>causing much misconception and grief) is the easiest way to achieve the
>>interoperability
>>we all think is desirable.
>
>
> I hope not, as we will never have such a common
> foundation on which all will agree. Fortunately,
> this impossible dream is also completely
> unnecessary.
>
>
>> But . . .
>>
>> Winston Churchill once remarked that "You can always count on
>>Americans to do the
>>right thing -- after they have tried all the alternatives."
>>
>> Alas, it seems that ontologists are proving him right once again.
>
>
> Well, I'm still not an American.
>
> Pat Hayes
>
>
>>Pat
>>
>>-----Original Message-----
>>From: cuo-wg-bounces@xxxxxxxxxxxxxx
>>[mailto:cuo-wg-bounces@xxxxxxxxxxxxxx] On Behalf Of Schoening, James R
>>C-E LCMC CIO/G6
>>Sent: Monday, February 05, 2007 7:23 AM
>>To: cuo-wg@xxxxxxxxxxxxxx
>>Subject: [cuo-wg] The next key question
>>
>>CDSI WG,
>>
>> Given Pat Hayes' description (below) of what Jim Hendler refers
>>to as "URI-based reference mechanism
>> coupled with the standard for KR and other aspects is
>>aimed exactly at scalability.":
>>
>> The key question now is: Could the above referenced technology
>>(when it matures) be used to achieve semantic interoperability across
>>large numbers of domains (with independently developed ontologies)?
>>Any takers?
>>
>>Jim Schoening
>>
>>
>>
>>
>>
>>
>>Pat Hayes wrote:
>>
>>> OK, here's my take on that.
>>>
>>> First, "standard for KR". I think all that Jim means is, the SWeb is
>>> intended to use whatever is the best available KR mechanism that can
>>> be adopted as a 'standard', ie which a wide enough spectrum of users
>>> can be persuaded to agree to use. No such choice will be free from
>>> controversy. OWL wasn't and isn't free from controversy, and nobody
>>> even knows if a large enough community can ever be brought to
>>> consensus on an acceptable Rule language. But assuming that some WG
>>> can get its job completed, and produces a useful notation, then that
>>> can be used on the SWeb. There are plenty of potential candidates
>>> which are way more expressive than OWL readily available. So it would
>>
>>> be a mistake to identify the SWeb vision with OWL or DL technology in
>>
>>> particular. OWL-DL is just the first in what one hopes will be an
>>> evolving series of KR standards which will provide the infrastructure
>>
>>> of the SWeb. Perhaps the next one will be more like
>>> Python+Prolog on steroids. Or it may be a
>>> breakthrough in CL reasoners using the guarded fragment, who knows?
>>> The decision is as much political as technical, or even subject to
>>> whims of intellectual fashion.
>>>
>>> "URI-based reference mechanism" is more interesting. This is one of
>>> the few things that really is new and different about the SWeb: it is
>>
>>> part of the Web, and subject to Web conventions and protocols. It
>>> isn't *just* applied ontology engineering. So, every SWeb ontology is
>>
>>> required to use names drawn from a (literally) global set of names.
>>> The scope of these names is the entire Web. There are no 'locally
>>> scoped' or 'private'
>>> names on the Sweb. So if your ontology uses a name for a concept, my
>>> ontology can use it too.
>>> Anyone can 'say' anything about everyone else's concepts, on the
>>
>>Sweb.
>>
>>> This is a whole new game, which nobody has played before. A can
>>> introduce a concept called A:thingie1 and B can introduce B:thingie2,
>>
>>> and C can then, entirely independently and without asking for A or
>>
>>B's
>>
>>> permission, assert that (say) A:thingie1 is the same as B:thingie2. A
>>
>>> and B may disagree: tough tittie, they can't stop C from making the
>>> assertion. C can say things about A's ontology, in fact, such as
>>> assert that it is all BS. The globality of the namespace has a whole
>>> range of consequences which we are only beginning to explore. And
>>
>> > being URIs (actually IRIs these
>>
>>> days) , these names can also be used as identifiers which *access*
>>> things on the Web.
>>> Whether these accessed things should be the referents of the names is
>>
>>> currently controversial (I think not, in general), but that they
>>> access
>>> *something* is not even remotely at issue. So SWeb concept names have
>>
>>> a whole new dimensionality to them, which is (or at any rate can be)
>>> orthogonal to their use as referring names. In particular, it allows
>>> ontologies to "address" other ontologies (a pale version of which is
>>> the OWL:imports primitive, but one can do a lot more than this),
>>
>>which
>>
>>> obviously has many potential applications relevant to scaling.
>>>
>>> Hope this helps.
>>>
>>> BTW, I entirely agree with Jim's optimism. I think people are way too
>>
>>> scared of inconsistencies. Lets wait and see what problems actually
>>> crop up before trying to solve or avoid ones that we only worry about
>>
>>> rather than actually find.
>>>
>>> Pat
>>>
>>
>>_________________________________________________________________
>>Message Archives: http://colab.cim3.net/forum/cuo-wg/
>>Subscribe/Unsubscribe/Config:
>>http://colab.cim3.net/mailman/listinfo/cuo-wg/
>>To Post: mailto:cuo-wg@xxxxxxxxxxxxxx
>>Community Portal: http://colab.cim3.net/
>>Shared Files: http://colab.cim3.net/file/work/SICoP/cuo-wg/
>>Community Wiki:
>>http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/CommonUpperOntologyWG
>>
>>
>>_________________________________________________________________
>>Message Archives: http://colab.cim3.net/forum/cuo-wg/
>>Subscribe/Unsubscribe/Config: http://colab.cim3.net/mailman/listinfo/cuo-wg/
>>To Post: mailto:cuo-wg@xxxxxxxxxxxxxx
>>Community Portal: http://colab.cim3.net/
>>Shared Files: http://colab.cim3.net/file/work/SICoP/cuo-wg/
>>Community Wiki:
>>http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/CommonUpperOntologyWG
>
>
> (013)
_________________________________________________________________
Message Archives: http://colab.cim3.net/forum/cuo-wg/
Subscribe/Unsubscribe/Config: http://colab.cim3.net/mailman/listinfo/cuo-wg/
To Post: mailto:cuo-wg@xxxxxxxxxxxxxx
Community Portal: http://colab.cim3.net/
Shared Files: http://colab.cim3.net/file/work/SICoP/cuo-wg/
Community Wiki:
http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/CommonUpperOntologyWG (014)
|