Dear Matthew, Mike, Kathy, and Leo, (01)
MW
> My interpretation of [Ali's claim is that it's] contextual.
> It is certainly true that in some cases a small amount of machine
> readable semantics can go a long way. As noted on bullet (38G7),
> it really seems to depend on the target application and the underlying
> value proposition that drives the creation or application of
> computational ontology to the problem space. (02)
Absolutely! That is an issue that the AI community discussed and
worked on for years. Any given piece of knowledge can be used in
many different problems or applications. It's counterproductive
to require the domain expert or knowledge engineer to anticipate
every kind of context or problem for which it may be used. (03)
MP
> Inference rules that involve more than one support from the abox
> cannot be parallelized as easily as say, RDFS, which never involves
> more than one assertional component. (04)
I don't disagree with the technical point, but I strongly disagree
with the assumption that the person who is entering the knowledge
should think, know, or care about such issues. (05)
That is my major complaint about the SW toolset: they require the
knowledge engineer to anticipate how the knowledge is going to be
used and to choose the tools for processing it. See the link below. (06)
KL
> I got a 404 when I clicked on the link. (07)
Sorry. I was in a hurry, and I typed .com instead of .pdf: (08)
http://www.jfsowa.com/pubs/fflogic.pdf
Fads and fallacies about logic (09)
Leo
> 1) The level of expressiveness (representation) it takes to develop the
>ontology
> you need for your domain. This is development time expressiveness.
>
> 2) Transformation of the representation of (1) to (3), i.e., knowledge
>compilation.
>
> 3) The level of expressiveness (representation) it takes to efficiently
> reason over the ontology at runtime. This is run time expressiveness. (010)
That's better, but the major question is who or what does the work.
An even more important question is why? For what purpose? (011)
Re #1: There is a huge difference between the domain and the problem.
Expressiveness is *not* a property of a domain, but of a *problem* . (012)
As I said in the fflogic.pdf article, the same statement can be
undecidable for one problem (trying to prove that it's a theorem
-- i.e., true in every model) but very efficient for another problem:
asking whether it's true in just one model -- the current DB. (013)
Re #2: This is the single most important point, which I would highlight
more than anything else. If you have a suitable knowledge compiler,
all the rest of the discussion about expressive power is irrelevant
for the person who is entering the knowledge. See the discussion
in fflogic.pdf about Cyc and about Bill Andersen's article from
1998 (which gave them the confidence to found Ontology Works). (014)
Re #3: Again, this depends on the *problem*. Cyc has a wide
range of inference methods tailored for different kinds of problems.
For any given problem, it extracts what it needs for that problem
and uses an appropriate algorithm to handle it. That is why
Lenat said that decidability has *never* been a problem for them. (015)
The Andersen et al. article was written to describe a novel way
of doing that. They wrote it while they were using Cyc at DoD,
but Doug L. did not like their strategy: (016)
1. Use Cyc as a development tool. (017)
2. For any particular problem, extract the relevant knowledge
from the Cyc KB plus the newly entered axioms. (018)
3. Compile the knowledge into a form that can be used by
other systems without having Cyc involved. (019)
My comment was that Doug lost a valuable opportunity. Instead
of disapproving, he should have hired them to develop Cyc as a
general-purpose knowledge acquisition and development platform. (020)
Don't tell knowledge experts about expressive power. Just tell
them to enter whatever they know about a subject in controlled NLs
supplemented with diagrams, such as UML. Then let the knowledge
compiler translate the knowledge to formats that can be processed
efficiently for each type of problem. (021)
That is why I kept talking about Tools, Tools, Tools. You can't
achieve the goals stated for the Semantic Web or anything else
unless you have good knowledge analysis and compilation tools. (022)
And the people who most need to learn that lesson are the ones
who designed OWL. They are very intelligent people who are very
knowledgeable about one specialized topic: how to prove theorems
about decidability. But they are totally clueless about how to
design a system that people can actually use to solve problems. (023)
We have to tell them that there is a better way. (024)
John (025)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2012/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2012
Community Portal: http://ontolog.cim3.net/wiki/ (026)
|