Re: SUO: RE: Re: Missing Ingredients
Richard Cooper wrote:
[...]
> This set of three questions is the most important triple we're
> dealing with in all SUO work. Getting clear answers to how
> meaning is represented, communicated, stored, compared and
> organized would be a successful result.
Rich,
Yes, I agree very strongly that until there is some clear
answers to this, until there is a foundation such as you
describe, everything else is resting on sand. This is in
part why I thought tackling the LBase document might be a
good start, since it was Pat's best shot at the problem.
John and Jon have both written extensively on this, but
rather than point at existing, lengthy texts, it'd be good
if we just talked I think. Informally, until the formalisms
and the assumptions behind them become clearer. Background
reading is good too, but conversation seems called for.
> We have predefined the answer to be an ontology. Then we refined
> that concept to include the lattice of ontologies, plus the IFF
> framework, but I still get the feeling there's a lot of stuff left
> out.
>
> So I agree with Tom that the focus should be refined further
> to incorporate real world database concepts, and I add one more
> suggestion; that we should be working with natural language
> words and sentences to impose the type structure, or class
> structure, and property lists, of common everyday concepts like
> address, customer, person, ..., fill in your favorite concepts.
>
> Finally, since we haven't been able to agree on more enhanced
> ontologies than WordNet, perhaps we should start the bottom-up
> process by extracting exactly the ontology that WordNet provides.
> This could be one of the bottom-level concept sets, along with
> others that may appear in the lattice as we continue.
Before we head down the WordNet road, there are a number of
questions that pop to mind. The first is that if there is to be
any value in using something like WordNet, we have to assume that
it has some kind of legitimate, ordered, and relatively accurate
structure to its ontology, and this would rely on language itself
having the same kinds of order (i.e., any structures meant to
describe natural language must assume that natural language has
structure), that WordNet is not simply arbitrarily ordered in some
way similar to the ordering of natural language. (Perhaps
"arbitrary" is not quite right -- "organic" or even recursively
associative, but I don't think of natural language as remotely
being formally ordered, not even within one person's head, much
less the world population. If there is an inherent structure in
natural language, it is so enormously more complex and intricate
than our current understanding that we cannot assume to find a
solution for many decades, which jibes with what I hear from the
computational linguists.)
My difficulty with this is that from my understanding of the
current state of computational linguistics, we're many, many
years from understanding language to the level, and indeed,
several comp. ling. experts I've heard (such as Geoff Nunberg,
who spoke at ICCS 2001) or read imply that it's likely that
language is so inextricably bound to individuals and individual
contexts and usages, linguistic families, cultures, communities,
etc., that no formal set of rules for natural languages are
likely to be devised. I tend to rule out WordNet as a tool to
support computer-based reasoning for this reason, and just
consider it a handy tool for human judgment and use, like a
book-form thesaurus. I note a number of KR projects that use
WordNet for reasoning, but that doesn't mean I think such use
is viable. As we all know, word order, a comma, pretty much
anything can alter meaning. Systems that use word frequency
to make inferences about a text are a great deal different
than those that attempt to actually reason upon natural
language sentences written informally.
John has talked at times about use of controlled vocabularies,
but even given the limitations imposed on communication by
them still leaves open the question of interpretation on the
part of author and reader. Given that many application areas
cannot rely on the availability or use of such control over
usage, this isn't a general solution IMO.
Without having this devolve into yet another discussion of Peirce,
one of the reasons I've been enthusiastic to learn more about
application of Peirce's Thirdness principle is that (from a
top-down POV now), an ontological system that at its core took
into account the given that use of natural language injects into
the mix the element of interpretation and relative meaning would
be able to at least in principle handle natural language. Perhaps
better said, a system that doesn't employ Thirdness couldn't
handle anything to do with interpretation, and natural language
is about interpretation (i.e., meaning). I realize that it's a
big step between FOL and McCarthy, but we gotta start somewhere.
So to my understanding, in order to start at bottom-level or
bottom-up, you first have to have a system that at its heart
(top-down) is structured to take into account interpretation.
So we really need to understand that first triple of questions
you mentioned above, structurally, semantically. So if a system
were to use say, WordNet, every natural language structure in
the ontology would be bound to a specific context, and this
context would follow it around -- no abstract, idealized notions
of a word, phrase, or concept could do.
[I'm not sure if this is clear, but it's the best my neurons
can do for now... they're kinda fried lately. I'm waiting for
a convenient time for a gin and tonic, after I don't need them,
when I can properly fry them.]
Murray
......................................................................
Murray Altheim http://kmi.open.ac.uk/people/murray/
Knowledge Media Institute
The Open University, Milton Keynes, Bucks, MK7 6AA, UK .
Monkeys use thoughts to control robotic arm
http://www.sfgate.com/cgi-bin/article.cgi?file=/c/a/2003/10/13/MN2018.DTL
Bush uses media expertly to push apocalyptic view
http://truthout.org/docs_03/091403J.shtml