Re: SUO: Multi-Source Ontology (MSO) Draft Ballot Question
Adam, Murray, and Bill,
The distinction of logic-based vs. language-based
is meaningless because first-order logic is a
subset of every natural language. Aristotle
and the medieval scholastics developed logic in
subsets of NLs (Greek and Latin). When Leibniz,
Boole, and Peirce used algebra to express logic,
they viewed their work as a streamlined extension
and continuation of the Aristotelian tradition
-- definitely not as a break with it.
Furthermore, George Miller's primary aim was *not*
to build "just" a thesaurus. Look at the research
in psycholinguistics he published over the past 50
years, in which he did a great many studies of
how people do reasoning. George is a very strong
advocate of the Aristotelian tradition of strict
logical inheritance in the hierarchies.
Before George started the WordNet project,
every machine-readable dictionary available
was burdened with copyright restrictions that
created intolerable problems for anybody who
wanted to do any kind of research. Therefore,
he decided to develop a freely available resource
for computational linguistics.
However, George recognizes that there are problems
with the hypernyms, and he and his group have been
doing their best to correct them. Whenever strict
inheritance is violated, they correct the problem.
Unfortunately, WN is so big that fixing a problem
at one point can create new problems somewhere else.
The violations of strict inheritance, such as
the one Adam mentioned, are just as serious for
linguistics as they are for logic:
AP> As an example, take the inference path that is possible
> from the SUMO term Motion to the WordNet synset "motion,
> movement, move" then up the hypernym links to "change",
> "action" and then "act, human action, human activity".
> Through that faulty inference chain, one could conclude
> that any Motion is an intentional human action, which
> of course is false.
As I said earlier, this is a good example of an obvious
*error* in WN that is very serious for linguistics.
The linguistic constraints on word selection and other
semantic issues depend critically on having a valid
inheritance path for properties such as human vs.
nonhuman, intentional vs. accidental, etc.
So Adam's conclusion is absolutely false:
AP> Hypernym links are linguistically
> valid, but not logically valid.
For both linguitics and logic, it is essential that
the hypernym links preserve strict inheritance.
When they don't, they are wrong for both.
Philippe has been doing some very good work in checking
the WN links and sending his corrections to the WN people
at Princeton. So far, he has found over 300 corrections,
which he contributed to Princeton. Unfortunately, the
Princeton group is very much in need of funding, and
they don't always have the resources to fix all the
entries that need to be fixed.
In any case, the MSO is *not* permanently wedded to WN.
Philippe uses it because it is available, and he has
been doing his best to correct every deficiency that
he finds. If somebody magically provided a newer and
better hierarchy, he could unplug WN from MSO and
plug in the new hierarchy.
But unless and until somebody provides a hierarchy
that is more reliable, better debugged, and with as
broad or broader coverage than WN, Philippe is using
WN. Instead of letting Adam and other groups go off
on their own, I think it very important for all of
us to work together in building such a well-debugged
and well-maintained hierarchy -- and to use *all*
the resources that are available, not just one.
All the arguments that Adam has presented are very
strong reasons for us to *collaborate* on building
a very good, solid, broad coverage system rather
than a bunch of half-vast competing projects.
John