SUO: RE: Re: Enhancing Data Interoperability with Ontologies...
John Sowa wrote:
> Just two comments:
> RC> The goal of constructing an English-like language
> > should be to orient it toward the way people actually
> > use English in written form.
> First, there is no unique way of using any natural
That's correct, but the next step in AI requires
that we better understand human uses of language.
> As Wittgenstein pointed out, there is an
> open=ended number of "language games" that can be played
> with any NL.
Yes, speech acts, dialogue models, plot layers,
metaphor and other human language acts have to be
studied and theorized about. That's why I would
like a CLCE to be more humanly oriented.
> Controlled English is just one game that
> can be played with English syntax and vocabulary, and
> it is a subset of the game that mathematicians routinely
> play with English (and other NLs).
> Second, I asked for specific examples where polysemy
> would be useful for knowledge representation. I did not
> mean a general discussion, but actual sentences -- i.e.,
> specific examples of English sentences in which polysemy
> was desirable for kn. rep.
> You said that in your use of Rosie, you found cases
> where polysemy would have been useful. What were
> they? Show the actual sentences you would have
> liked to write, but couldn't.
Since I couldn't write them John, I don't have examples
in mind from what happened back in the 80s. I just
remember being repelled by the shallow verbosity of
ROSIE. It simply didn't work any better than a
programming language, yet didn't provide anything
I would call English interaction either.
For examples of sentences I would like to process, see
which lists 38 e-books written by one author (Louis
L'Amour) over his lifetime. E-books can be used to
study the texts and organize theories about the patterns
that L'Amour used. From that, a set of dialogue frames
could possibly be developed that would model realistic
Cynthia Whissel publishes papers on how she uses her
"emotion word dictionary" to investigate psychological
factors in authors of many kinds. She built this
dictionary over many years and was able to correlate
emotional word rankings over hundreds of volunteers.
From that, she posited how words can be modeled for
It would be nice to use her methods on L'Amour's
dialogs to try to measure emotional drives within
L'Amour's dialoge patterns and characterizations of
Yes, a controlled language has to ultimately be
disambiguated into a single meaning to be properly
performed, but the process of disambiguating is just
too important a factor of study to be eliminated
by design fiat.
The methods I mentioned in my previous post are
all reasonable first ideas that could be applied
to a CLCE. If they were, the writings of L'Amour
could be analyzed, and those patterns that are
completely unable to be disambiguated could be
studied as a set of patterns that defy the signature
method of disambiguation. Then further progress
could be made on how much world knowledge really
can be required for a believable character, and
how that world knowledge could be provided through
dialogues with human operators.
Think of T'ilce on Stargate, or that logical
character (I can't remember his name) on Star Trek.
Highely logical people have been used in fiction
to contrast the more human characters they
That's my own goal for a CLCE; to be a first
kernel for a true natural language character,
and to support extensions by adding packages
that provide those capabilities that we know
how to implement presently.