RE: SUO: RE: RE: Re: Missing Ingredients
> Not to head back into the semiotic mire, but it's also been shown that
> it's very difficult to have shared meanings, even when there are
> "formal" vocabularies available, as people will interpret definitions
Yes, in practice shared meaning is a lot more difficult that people
think, even when vocabularies are strictly shared and exhaustively
defined. Doing this with formal vocabularies is still vastly simpler
than NLP, but by no means is it a cakewalk.
> I believe with the SUO we are discussing going beyond such systems,
> into an area where inferencing amongst concepts and relations based
> on shared ontologies is one of the primary goals. I would posit,
Right; mapping between formal vocabularies is the next obvious
incremental step beyond simple interop via shared vocabularies.
> language. And microtheories will only go so far in bridging the gap
> between mind and machine. But that's for another day...
Well we probably agree here. It's hard enough to get systems to interop
with shared schemas, let alone attempts to share mappings between
vocabularies. In theory, this approach seems to be more pragmatic than
NLP, but we really won't know how suitable for widespread use this
approach is until we have more deployment and experience. I also think
the only hope is to keep things as simple as possible and opt for the
minimal subset of inferencing rules, etc. -- the problems will be ironed
out by humans and bureaucratic processes rather than by increasingly
complex systems of logical processing rules. I have a feeling that the
current set of rules embodied in OWL is getting very close to the limit
of what "normal" human organizations will be able to understand and
manage.
> > B) Using WordNet terms in a system is bound to fail because the very
act
> > of using English words as terms somehow forces you into an
NL-processing
> > rathole
>
> The rathole, choice B.
My opinion was that there is really no difference between
http://purl.org/wn/nouns#house and
http://tempuri.org/terms#0289237498327 as far as a KR system is
concerned, as long as both terms are well-defined and understood by both
parties. But I suppose that using neutral identifiers reduces chances
that one or the other party will assume unintended meaning (of course
the risks apply to artificial identifiers as well; but admittedly
perhaps less).