In looking at WordNet, I've noticed that it's almost a single-inheritance
tree, in which almost all synsets each have exactly one hypernym. There
are a few exceptions, like the noun piano having both stringed
instrument and percussion instrument as hypernyms, but this is rare. They
had a half-hearted dedication to "treedom". The choice of such a tree-like
structure was a mistake, in my view. I think it should have a
non-tree-like poset structure (partially ordered set, DAG) with much more
multiple inheritance from multiple hypernyms. Putting concepts into a tree
forces you to make some silly decisions as to which of several salient
superclasses should be designated as "the" hypernym.
The higher you go in WordNet, the more you get into controversial or
dubious linkings; the lower levels are more obviously reliable and WordNet
is quite useful in its lower reaches. And it's big.
Cyc has established over 6,000 links to WordNet (Cyc "constants" linked to
WordNet "synsets") with many more to go, and I believe these cover all
3000+ of the Cyc-based, publicly released "reference ontology". (In case
you dont't know, those are on Cycorp's web page at http://www.cyc.com ---
see the section on the Upper Cyc Ontology.)
Yours truly, Fritz Lehmann
Fritz Lehmann, Cycorp, 3721 Executive Center Dr., Austin, TX 78731 USA
email: fritz@cyc.com telephone: (512) 342-4013 fax: (512) 342-4040
======================================================================