- University of Southern California,
Information Sciences Institute
- PRINCIPAL INVESTIGATORS:
- Robert MacGregor
- TITLE OF EFFORT:
- Tools for Assembling and Managing Scalable Knowledge
- ISI is developing a new system, called Knowledge Builder,
to construct, manage, and reason with very large, high-performance
knowledge bases. Knowledge Builder will be a Web-based collaboration
system (derived from ISI's
Ontosaurus Tool) for assembling,
validating, maintaining, and exporting large ontologies and knowledge
bases. Underlying Knowledge Builder is a scalable KR system (called
able to represent and reason effectively with a library of
knowledge bases too large to fit within the virtual memory of a single
- Knowledge Builder focusses the development of ISI's
and PowerLoom systems
toward the task of managing large
ontologies. It will expand existing
Ontosaurus capabilities with a
full-suite of operations for knowledge base assemby and manipulation.
We will will extend PowerLoom
with view, abstraction, and mapping
components needed to support KB construction, and we will make it
scalable by developing a "paging module" that couples it to commercial
To insure the successful use of Knowledge Builder both within the
research community and in fielded DARPA applications, we place unusual
stress on interoperability and portability. Knowledge Builder will be
able to communicate using a variety of KR languages
GFP), and it
will support direct, efficient interfaces to multiple native object
systems (C++, Java, IDL, and CLOS). ISI's unique STELLA translator
technology will enable us to release C++, Common Lisp, and Java
versions of Knowledge Builder.
- Our work during this period has focused primarily on
readying our PowerLoom knowledge representation system for its first
release, scheduled for September, 1997. PowerLoom will provide the
foundation for the scalable reasong system used by Knowledge Builder.
We have also developed a new paradigm for coupling classifier-based
systems like Loom and PowerLoom to conventional software. Our
"knowledge amplifier" inputs data stored in an object system, applies
knowledge-based inferencing to it, and then exports the enhanced
result back to the object system. This technology will help programs
like JFACC to overcome the representational limitations they face
using current object-based software tools.
- FY97 ACCOMPLISHMENTS:
- Released C++ and Common-Lisp versions of the PowerLoom knowledge
representation system. The C++-based PowerLoom will facilitate the
transition of AI technology to non-Lisp programming environments.
Enhanced Ontosaurus with a "highlight" feature that makes it easy to
differentiate knowledge acquired from different sources. Demonstrated
the highlighting on a knowledge base containing a
Cyc ontology merged
with a JFACC ontology.
Demonstrated a prototype "knowledge amplifier" that caches the results
of knowledge-based inferencing as augmentations to an object base or
database. The amplifier introduces a new paradigm for supporting
interactions between conventional software and AI software.
- FY98 PLANS:
- Develop a "knowledge paging module" that interfaces PowerLoom to a
commercial data storage system. This will create a platform for
conducting experiments on how to efficiently perform complex reasoning
over very large knowledge bases.
Develop a schema mapping tool that will facilitate the exchange of
knowledge between a knowledge base and other database(s) or knowledge
base(s). Use the tool to semi-automate the process of coupling an
object-oriented application to a knowledge representation system using
our "knowledge amplifier" approach.
Upgrade Ontosaurus' capabilities for assisting the process of
aquiring, merging, and maintaining large ontologies. Use the enhanced
tool to support knowledge acquisition (KA) acitivities within the HPKB
and JFACC programs.
Extend PowerLoom's translator repertoire with a bi-directional
Ontolingua translator and a GFP interface, thereby enabling its use as
a knowledge server within DARPA research and ISO-services programs.
- TECHNOLOGY TRANSITION:
- PowerLoom's precedessor, ISI's Loom system, has
been distributed to more than 100 universities and corporations, and
it is being used by numerous ARPA contractors in ARPA's Intelligent
Integration of Information (I*3), planning (ARPI), Air Campaign
Planning (ACP), machine translation, and image understanding programs.
We have similar goals for PowerLoom and Ontosaurus. ISI has made an
unusually strong commitment to developing AI technology that will
transition easily. We now develop all of our software in the STELLA
language; a STELLA program is automatically translable into both C++
and Common Lisp (a STELLA-to-Java translator is planned as a future
HPKB milestone). Our Fall '97 release of PowerLoom will represent the
first occurrence of a knowledge representation system written in C++.
- DATE PREPARED: