Intent
of Work Statement for 2003
Yale-BBN
Project: Automated Tools for Mapping Among Ontologies
Award
number: F30602-00-2-0600
PI:
Drew McDermott (plus Mark Burstein with a subcontract at BBN)
The focus of our work in the last six months
has been to create an ontology translation server, populate it with as many
ontologies as possible, and offer it for use to other DARPA/DAML researchers
(or interested parties outside the project). In the next year, we intend to
continue this work and integrate it more tightly with the DAML Experiment.
We had hoped by this point to have picked
up several existing ontologies and created a library of links among them.
(These links are themselves ontologies, called merged ontologies, hence the
name of our system, OntoMerge.) Unfortunately, with one exception, we have not
encountered many ontologies that are rich enough to make the exercise
worthwhile.
The one exception is the DAML-Time
ontology, which is the subject of current electronic discussion and evolution.
This ontology is orthogonal to most other ontologies, and yet obviously an
important candidate for merging. The goal is to be able to take a static,
timeless theory and make it timeful by merging it with the DAML-Time standard.
We are currently in the middle of this project, and expect it to come to
fruition early in the next year.
In parallel, we are now attempting to take
advantage of the fact that the Web is full of “semi-formalized” ontologies in
the form of metadata for XML languages: DTDs, schemas, and documentation. These
languages are often more complex and more detailed than formalized RDF
vocabularies. The reason is because that’s where communities of users have been
putting their energy in order to produce useful tools for information exchange
in real-world applications. What we are now embarking on is an effort to
extract formalized ontologies, to the extent possible, from DTDs and XML
schemas. It turns out that it is fairly to easy to produce a “surface ontology”
automatically, that combines some purely syntactic aspects of the given XML
languages, and some actual ontological material. A human then connects this
surface ontology to “deep ontology,” either one that already exists, or a
cleaned-up version of the surface ontology. We anticipate producing a lot of
new ontologies in areas such as personnel files, industrial inventory
maintenance, scheduling, and web services, by this “cleanup” process. The
beauty of the scheme is that we can transform them into new DAML ontologies by running
our existing tools. Furthermore, given any document in the given XML language,
we can automatically translate it to RDF with respect to this new ontology. We
can also shift it to different ontologies using OntoMerge.
Metrics for measuring our progress remain the same: the number of
ontologies in our library, and the number of people who use them.
We are eager to integrate
as deeply as possible into the DAML Experiment. So far we have done a bit of
translation in connection with the SONAT map database. It might be appropriate
for us to try to get involved in building a tool that requires multiple
ontologies, such as the COA Effects Agent described in Version 0.8 of the DAML
Experiment Plan. That way, instead of waiting for customers for our services to
show up, we could actively look for how the multiple-ontology problem would
arise in the context of a concrete application.