[Corpora-List] Semantic primitives

[John F. Sowa]

On 1/18/2011 12:46 PM, Ken Litkowski wrote:
> I hope we don't take John's words to stifle the hunt for semantic
> primitives (look at the billions in the hunt for the Higgs boson).

I certainly do not want to stifle anybody's insights, and I also
believe that it's important to analyze complex terms into simpler
terms.  In my own work (and with colleagues at our VivoMind company)
we make extensive use of a convenient set of relations, which I
summarized in my book _Knowledge Representation_.

A summary of the "KR Ontology" is on the following web page:

    http://jfsowa.com/ontology/

Click "Thematic Roles" for the roles that were used for the
IBM-CSLI Verb Ontology:

    http://www-csli.stanford.edu/~arunm/

We have found those terms very useful, but I would hesitate
to call them 'primitive' in the sense that no further analysis
would be possible (and often desirable).

> I view this work as being very a posteriori. Robert Amsler recently
> expressed his enthusiasm for Google n-grams as being useful
> identifying component elements of noun compounds, likening this
> to analyzing chemical compounds into their component elements
> and binding properties.

I definitely approve of deriving the basic terms from corpora by
a posteriori methods, usually guided by some a priori hypotheses,
which will be revised and refined as the study progresses.

And I really like the analogy with chemistry.  My philosophical hero,
Charles Sanders Peirce, had a BS degree in chemistry, and he was
stimulated by the work in organic chemistry to define his existential
graphs as a way of representing the "atoms and molecules of logic".

But the analogy suggests two points:  atoms and molecules are just
as important today for doing chemistry, even though the physicists
have discovered far more detailed constituents at the subatomic
level.  Both approaches are valid for their range of applications.

John


_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora