SUO: RE: CG: RE: Question about CLCE
John F. Sowa wrote
> Rich,
>
> Every knowledge representation language, every
> programming language, and every artificial language
> for design, specification, or whatever requires
> monosemy -- one and only one meaning for each
> word. Since CLCE is designed to map to those
> languages, there is no way that it can do so
> in a simple, convenient way without requiring
> every word to have a unique meaning.
As the parse continues, ultimately one meaning must
be singled out. But there are languages that support
some degree of polysemy. For example, CELT is a
polysemic language for disambiguating an utterance
in a controlled language that is not as wide as
full English, but is wider than monosemic logic
statements. Using WordNet word senses in a slightly
restricted way, it transforms input sentences into
FOL statements in KIF.
The reason I like this approach is based on experience
in trying to discuss domains with domain experts.
These experts have vocabularies which they use with
other experts based on their mutual familiarity with
the domain terms. It the constant tuning of this
knowledge in groups of people over time that makes
a domain deeper than everyday experience. So the
words they use are ingrained into their thinking,
and its very difficult to get them to change to
a vocabulary that fits the FOL language instead of
their own.
As a knowledge engineer on some projects, I have
repeatedly observed this linguistic grounding in
the experts' thought processes. I have trouble
learning their words, but they have even more
trouble learning my words. And they have little or
no interest in the FOL representations, which they
find pretty meaningless. Its their rootedness in
a domain that they have lived in for many years that
makes them more expert than me in that domain. So
by giving them the names, verbs, phrases, and other
linguistic concepts they find familiar, the project's
productivity and clarity is greatly improved.
With ROSIE, I had to first find out what their
words meant, and then translate that into statements
that defined these words in ROSIE terms. But when
I went back to the experts with just a few words
defined each day, they felt the progress was too
slow, and often lost interest. An unmotivated
expert, even ordered by his boss to cooperate, isn't
very helpful. He just wants to get on with his
own problems and solutions.
So a better way is to record the experts' statements
directly in the language and then to use tools to
translate those statements into parsable FOL. Its
the reverse of the restricted FOL concept.
For many domains, there are large numbers of documents
with lots of material that can be dragged through
sieves to get better languages. I think that
approach, with a FOL target language for the tools,
is better than having the experts and the engineers
forced into using a vocabulary and phrase structure
that is unnatural to the domain. I didn't find
the ROSIE rules very readable myself; these experts
really do have deep roots into their language.
JMHO,
Rich
> > While I like the basic idea of a controlled language,
> > requiring monosemy is too strict a constraint for most
> > people. I write from the familiarity of using a form
> > of CLCE called ROSIE back in the late eighties.
>
> If you want a word, such as "bank" to have more
> than one word sense, you can do so by using
> distinct spellings, such as "bank1" and "bank2".
> Or better, you can use hyphenated terms, such
> as "river-bank" and "money-bank".
>
> John
>
>
>
> ==============================================================
> ==========
> To post a message, send mail to cg@cs.uah.edu.
> To unsubscribe, send mail to majordomo@cs.uah.edu with the command
> 'unsubscribe cg' in the message body.
> See http://www.virtual-earth.de/CG/cg-list/ for the mailing
> list archive
> See http://www.cs.uah.edu/~delugach/CG for the Conceptual
> Graph Home Page
> For help or administrative assistance, mail to owner-cg@cs.uah.edu
> e
>