RE: SUO: RE: RE: Re: Missing Ingredients
Hmm, you seem to be arguing that natural language processing is
difficult. Obviously NL processing is heavily dependent on all sorts of
context, but I guess I missed the point where the topic became NL
processing.
As far as I'm concerned, the entire reason that activities such as KIF
and SUO exist is because smart people realize that NL processing is too
difficult for prime-time and that there are considerable benefits to be
reaped from taking a more low-tech approach. We simplify the task of KR
interop by designing systems that depend on shared terms and
vocabularies, and shared mappings between vocabularies. In these
systems, it is VERY important to respect the shared meanings.
It is completely possible to build knowledge-based systems which interop
based on shared sets of predefined terms, and which do so without
getting mired in semiotics debates. This is low-tech and has high admin
overhead, but is working in thousands of systems today. And this is
exactly the philosophy from which existing KR interop efforts are
building.
But I guess I am still missing your point. Are you arguing:
A) "it is impossible to build a KR interop system without paying
attention to all of these semiotics issues"
-or-
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
-J
> -----Original Message-----
> From: Murray Altheim [mailto:m.altheim@open.ac.uk]
> Sent: Wednesday, October 22, 2003 2:10 PM
> To: Joshua Allen
> Cc: Richard Cooper; Tom Johnston; Jon Awbrey; SUO
> Subject: Re: SUO: RE: RE: Re: Missing Ingredients
>
> Joshua Allen wrote:
> [...]
> > We should be very careful to avoid presenting too extreme a
position
> > with this sort of argument. Take this recent news article [0]
which
> > argues that "no means no":
> >
> > <quote>"The problem is not one of nomenclature. Words signify
concepts
> > and "no" still signals the concept of non-consent, for 100 percent
of
> > the world's English-speaking population, 100 percent of the
> > time."</quote>
>
> I think you're missing my point entirely. The above statement exactly
> expresses the relationship between words and concepts, namely one of
> signification. It's not a direct relation, and any given word does not
> have a meaning until you pick it up off the page and put it in your
> head. I never intended to state (if that's what you read) that words-
> in-interpretation have no meaning, only that until there is an
> interpretation, there is no meaning -- that's literally the meaning of
> the word "interpretation":
>
> 1. to give or provide the meaning of; explain, explicate,
eludicate.
>
> As for the word "no", no, it doesn't always signify or mean "no". It
> doesn't have an absolute meaning absent of any context. Ever read
> anything about date rape? The meaning of "no" is not so clear, even in
> legal matters. I find it interesting to note that the choice of proof
> here (i.e., the word "no") is very close to the logical concept of
> negation. Try to make that same statement with words and phrases like
> "right wing", "freedom fighter", "class", "priest", etc. Are these
> "edge cases", or just common phrases?
>
> > We can feel clever by pointing out edge cases and context-dependent
> > exceptions, and we can get truly sophomoric by pointing out that
"100
> > percent of the world's English-speaking population 100 percent of
the
> > time" is a "context". But clever arguments cannot dispense with
the
> > simple truth that "words mean things".
>
> I am not trying to "feel clever". I am pointing out a fundamental
> flaw in an approach to knowledge system design. I find it somewhat
> axiomatic that words do not mean thing, they mean thing_s_. That's
> the problem -- choosing which thing. If anything is by definition
> sophomoric, it's the idea that the world has "simple truths". The
> world, and specifically in this instance natural language, is
> enormously complex, and systems that are orders of magnitude simpler
> than the world they attempt to model are bound to fail. How else
> otherwise? Magic? Do you honestly believe you can create functional
> axioms to reason accurately upon absolute word meanings? (Pay no
> attention to the man behind the curtain...)
>
> I hardly think "dent" and "bird" are edge cases. They're typical and
> if anything, simple words. The English language is full of much more
> difficult to interpret but still simple words, such as "similar",
> "near", "is", "owns", etc. ("Do you own your home, or does the bank?"
> If you ask your customers, which answer will they provide? Is owner-
> ship legally defined, or socially? Or better, "do you own the software
> on your computer?" Define ownership there.)
>
> There are a vast number of simple words with complex meanings, words
> whose meaning varies within communities, words whose meaning is
> contested within communities ("life", "death", "Palestine",
"medicine",
> "food", "healthy", etc.). I see Israel is currently trying to redefine
> the boundaries of "Palestine", and the US Congress is actively in the
> process of redefining "life", and Monsanto is trying to tell us that
> GM foods are "foods" and "healthy", and that Orin Hatch has hatched
> a new attempt to amend the US constitution to redefine "citizenship"
> so that Arnold can run for president [1].
>
> There are words that are both complex and have complex meanings, words
> whose meanings are only known within specific communities ("nisi
prius",
> "mereological", "coprophagous", "ostension", "pragmatism",
"scuppernong",
> "nivation", etc.). Even within these communities, there are often
fights
> over the meaning of the vocabulary -- indeed, vocabulary definition is
> part and parcel of what defines a community, e.g., if you and I
disagree
> profoundly about the meaning of what constitutes "life", we'll likely
> not share a community. The schools and sects of philosophy and
religion
> are predicated on differences in interpretation, often of the same
> texts.
>
> > Granted, context plays an important role in semantics, but anyone
who
> > argues that situation is the *only* important factor should be
viewed
> > with deepest suspicion.
>
> I didn't say that it was the only consideration, I said that you can't
> interpret the meaning of a word absent situation, that words don't
have
> absolute meanings absent context. Can you tell me the absolute meaning
> of the word "bird"? What does "bird" mean? For any answer you give, I
> can show a context in which the definition you provide is wrong.
>
> > Dag Hammarskjold explained why this is such an
> > important matter of ethics:
> >
> > <quote>"Respect for the word is the first commandment in the
discipline
> > by which a man can be educated to maturity - intellectual,
emotional,
> > and moral. Respect for the word - to employ it with scrupulous
care
> and
> > an incorruptible heartfelt love of truth - is essential if there is
to
> > be any growth in a society or in the human race. To misuse the
word is
> > to show contempt for man. It undermines the bridges and poisons
the
> > wells. It causes man to regress down the long path of his
> > evolution."</quote>
> >
> > [0] http://slate.msn.com/id/2089687/
>
> I completely agree. I believe that the idea of developing a
> "knowledge representation system" designed to use words in ways
> that do violence to their meaning or in ways that cannot grasp
> their meaning to be deeply disrespectful for the power of
> words and the semiotic relationship between words and meaning.
> I'm specifically worried about the development of systems that
> undermine meaning and potentially, in their abuse, are damaging
> to the very society for which Hammarskjold speaks. The rather
> cavalier approach to natural language I've seen employed in many
> systems would have Hammarskjold rolling in his grave, there's
> so little "scrupulous care" for maintaining context or meaning.
>
> Put it simply, think about how someone uses a thesaurus, as I
> mentioned in my last post. If we can't develop a reasoning engine
> that could correctly choose from amongst the array of word choices
> in a thesaurus*, understanding the same contexts, inflections, and
> meanings of those choices, then how can we expect that same system
> to be able to reason correctly upon the relationships expressed in
> that thesaurus? Even if WordNet were to have ten times its current
> granularity, if systems developed using it don't employ context,
> the reasoning will be flawed, merely good guesses. Sort of like
> Ask Jeeves was 25% of the time.
>
> Murray
>
> * WordNet considers incision, scratch, prick, slit, and dent as
> synonyms, just within one word sense of the three available. I
> as a human being would choose very carefully amongst those
> choices, as they're not quite the same. Could a computer make
> the same choice as me, choosing "dent" because the material
> was metal and I *knew* what I was describing wasn't a scratch?
> Sounds like a challenge, like the Turing test.
> [1] http://blogcritics.org/archives/2003/09/28/112225.php
> ......................................................................
> 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
>
>