Key questions on Common Upper Ontology
4/20/2004 9:47
AM
Note from Larry Reeker,
I want to call additional attention to Mike’s succinct but important specifics.
If you look at #2, you will see something that is very important to the role of ontologies in AI.
Machine learning (from clustering to supervised learning to reinforcement learning) has developed in the past several years to a place where it provides solutions in a number of areas, but it still has some way to go in the area of transfer of learning.
The most promising approach in that area is case based reasoning (case based reasoning is sometimes called case-based learning, when used for transfer), and the breadth of ontology is controlling for case based reasoning.
If computer programs are ever to be deemed "creative", they are likely to have some form of case based reasoning and a broad ontology that allows metaphors and adaptations to novel situations. Mike's Specific #2 points out why an upper level is important to an ontology used in this way.
The breadth allows for expansion of learned expertise by finding similarities over a variety of subsumed categories inheriting similar characteristics, and the upper level handles the breadth of subsumption. Humans make use of analogies and metaphors constantly in their speech and in their thinking.
I would commend to the readers some important work being done at the Naval Research Lab in the area by that ties case based reasoning into ontologies based on text (both language and personal experience being carriers of ontologies).
Kalyan Moy Gupta, David W. Aha and some co-authors at NRL have written a series of papers, for using sub-language ontologies in limited ways.
Sub-language ontologies is the way to start, but will lead to broader pieces of language with meanings over broader ontologies.
The papers, one of which is to be given at FLAIRS in May, should be on the Web.
http://home.earthlink.net/~dwaha
David (< aha@aic.nrl.navy.mil >) will send electronic copies if you email him.
I agree with all of Mike's well-expressed specifics and am glad to see them articulated. Ontologies are important for practical reasons. They have theoretical and philosophical implications, too; but I don't see how anyone can ignore the practical implications that they carry.
Larry