[Corpora-List] WordNet vs Ontology

[John F Sowa]

I'll make one last posting on this thread to respond to a few points.

Merhnoush Shamsfard
> if we consider texts as the source of knowledge extraction for
> an ontology learning (OL)system then we cannot avoid word senses.

Yes, but please read Adam K's article.  He also cites Alan Cruse,
who coined the term 'microsense' for the fine-grained variations
that occur even in different documents on the same subject.  Just
look at all the variations in the senses of 'ontology' in each of
the notes of this thread.  (Sometimes even in the same note.)

Basic principle:  The ability to learn, discover, or invent new
word senses is key to language understanding -- even for adults
and even for experts in a field they know very well.  No fixed,
predefined set of word senses can be adequate.

Sebastian
> @John: I hope you are noticing, that I am trying to be keep
> all  of it as underspecified as possible

Fine.  But the primary requirement is flexibility in being able
to adapt to new variations.  That implies an underspecified core,
a collection of specializations that have proved to be useful,
and the ability to learn or invent new variations as needed.

Sebastian
> If we are focusing on engineering of information machines,
> then things are much clearer.

Yes, but different engineering projects have different goals
with different requirements.  See the next point.

Yannick
> [Sebastian's] post seems to boil down to a claim  that only
> RDF-encoded data should count as ontology. This  seems to be
> a bit near-sighted to me, as LemonRDF’s encoding  of WordNet
> is just that, an encoding which is very convenient but which
> adds nothing to the existing semantics.

I'll avoid pros and cons about notations.  Note that Cyc
supports mappings to and from RDF, SQL, and almost any
version of logic anyone has invented.

Some projects require very detailed and expressive logics,
and others don't.  But no predefined set -- not even the
600,000+ concepts of Cyc or the 900,000+ concepts derived
from Wikipedia -- can be adequate for everything.

A three-year-old child can understand language and learn new
concepts much faster and more creatively than anything based
on any knowledge representation system running today.

Eduard Barbu
> there is no system capable of learning an ontology
> from text. The best systems are supervised ones.

That depends on what you mean by supervision.  Children
(and adults) get positive and negative feedback from the
responses that other people make to their statements.

It's also possible to get positive and negative feedback by
the cycle of scientific research:  (1) start with some data;
(2) generate hypotheses; (3) make predictions; (4) test the
predictions against more data; (5) refine, revise, or replace
the hypotheses; and repeat from step (3).

And the supervision can be incorporated in semi-automated
learning:  adopt the scientific cycle, and add the option
of an "oracle" (some human tutoring) to supplement the
testing (step 4) and the revision (step 5).

For research on learning by reading, the Aristo project at
the Allen Institute for AI is doing some interesting work:

    http://allenai.org/TemplateGeneric.aspx?contentId=12

I would also cite some of our work at VivoMind Research:

    http://www.jfsowa.com/talks/goal7.pdf

John

_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora