Peter, et al., (01)
As I've said before, Thursdays are inconvenient for me to participate
in telecons. But I usually browse through the slides of talks, which
are generally of high quality and make many important points. (02)
While I was reading the slides from Thursday's talks, I noticed some
issues that are central to the way terminologies, taxonomies, and
ontologies are used now and will continue to be used. For the record,
following is the URL for Panel Session 08, which has all the URLs: (03)
http://ontolog.cim3.net/cgi-bin/wiki.pl?ConferenceCall_2013_03_07 (04)
I'll start with the slides by Patrick Lambrix, which cover a wide
range of basic issues: (05)
PL, Slide 11
> A taxonomy network is a set of taxonomies and sets of mappings
> between these taxonomies. (06)
PL, Slide 13
> We focus on taxonomies, named concepts, and is-a relations.
>
> We assume that all the existing mappings in the taxonomy networks
> are correct.
>
> The mappings represent equivalence and subsumption. (07)
PL, Slide 14
> Given a set of taxonomies networked by sets of correct mappings,
> how to detect and repair the missing and wrong is-a relations
> in these networked taxonomies? (08)
I agree with these points. But one missing point is the difference
between the detailed analysis in Slide 3 (about an inconsistency in
the DICE ontology) and the use of ontology for IR in Slide 5. (09)
The recall/precision issues have been the focus of IR since the 1960s.
There is some evidence and a lot of hope that ontology can help, but
the detailed axioms required for reasoning are rarely, if ever, used
to enhance precision and/or recall in IR. (010)
Question answering is an intermediate case. Some questions can be
answered by methods similar to IR -- those are questions that Google
and Bing can often answer with their first or second hits. But other
questions can require a huge amount of diverse kinds of analysis and
reasoning (e.g., Watson's answers for Jeopardy!). (011)
For something between those extremes, I spoke with Ron Kaplan, who
was the chief scientist of the PowerSet/Bing project. They had a
license to use the full ontology and software of Cyc, but Ron said
that the only part of Cyc they found useful for their purposes was
the Cyc hierarchy (taxonomy), not the detailed axioms. (012)
I'd also like to quote a point that Ed Barkmeyer made in a note
to Ontolog Forum (use Google to find Ed's complete note). (013)
EB
> Ontologies are made for a purpose. If the purpose is to enable
> interoperability of independent systems, that is different from the
> purpose of integrating operational systems and business practices
> of diverse business units. (014)
I agree. But I would emphasize that those integrating ontologies
are highly specialized for the purpose of designing compatible
software for a particular application domain. (015)
Even the engineers who design systems that conform to an integrating
ontology would also need a broad taxonomy for IR and Q/A. I'm sure
that they would ask questions about competitive systems and research
reports that might use the same terminology, but not the same axioms. (016)
I've mentioned Cyc many times, and I have been just as critical about
aspects of Cyc as I have about the Semantic Web or any other system.
But I repeat: Cyc is the largest formal ontology on the planet with
over 28 years of work by highly qualified researchers with PhDs in
computer science, AI, logic, linguistics, and philosophy. (017)
I certainly don't claim that Cyc has solved all the problems. But the
single most fundamental and successful Cyc design decision is based on
Guha's PhD thesis of 1991: Microtheories. He wrote that thesis under
the supervision of John McCarthy and Ed Feigenbaum at Stanford and with
collaboration with Doug Lenat and the other Cyc developers. (018)
Guha's basic idea wasn't completely new: an underspecified upper level
with an open-ended family of detailed microtheories. Guha cited other
people who said something similar -- including papers by John McCarthy
and my 1984 book -- but I gave most of the credit to logicians,
linguists, and philosophers from Aristotle to the present. (019)
As Matthew said, he successfully developed a monolithic ontology at
Shell as an integrating ontology. As Doug Foxvog replied and I agreed,
Matthew's ontology could be mapped to a microtheory in Cyc. (020)
Around 2000, Adam Pease was developing SUMO as a monolithic ontology,
which he discussed on the SUO email list. I recommended that he
reorganize SUMO along the lines of the Cyc microtheories, but Adam
resisted. But in the end, he made that reorganization. (021)
Bottom line: If you want to *evaluate* a network of taxonomies and
ontologies, check whether it has an underspecified upper level and
a hierarchy of microtheories. If not, give it an F. (022)
John (023)
_________________________________________________________________
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/OntologySummit2013/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2013
Community Portal: http://ontolog.cim3.net/wiki/ (024)
|