Building Concept Representations from Reusable Components
Bruce Porter
Computer Science Department
University of Texas
Austin, Texas 78712
Porter@cs.texas.edu
Peter Clark
Research Division
Boeing Company
Seattle, WA 98124
Clarkp@redwood.rt.cs.boeing.com
Abstract
Our goal is the construction of knowledge-based systems capable of answering a wide range of questions, including questions unanticipated when the knowledge base was constructed. Our approach to achieving this goal is to develop ways of building knowledge bases from reusable components, and to develop a computational mechanism for ``plugging together'' representational components. One of our target domains, which we use for illustration here, is bioremediation: the removal of toxic waste using micro-organisms that convert pollutant into harmless bi-products.
1 Background
Our goal is the construction of knowledge-based systems capable of answering a wide range of questions, including questions unanticipated when the knowledge base was constructed. Our earlier research on one large-scale project -- the Botany knowledge-base project -- shows that if detailed, declarative representations of concepts are available, then sophisticated question-answering performance can be achieved. However, manually constructing such representations is laborious, and this proved to be a major bottleneck in the project; moreover, it is simply not possible to anticipate all the concept representations that may be needed for answering questions.
1 Approach
Our approach to achieving this goal is two-fold. First, we are developing ways of building knowledge bases from reusable components. Each component encodes the objects and relations that describe a generic concept, such as move, produce or contain. Second, we are developing a computational mechanism for ``plugging together'' representational components as they are needed to answer questions and perform tasks.
One of our target domains, which we use for illustration here, is bioremediation: the removal of toxic waste using micro-organisms that convert pollutant into harmless bi-products. For a system to answer a variety of questions about bioremediation (requiring tasks such as description, prediction, and explanation), it needs comprehensive knowledge about the process. Although bioremediation is rather specialized, it's representation can be built from numerous generic concepts, including conversion (in which pollutant is converted into a fertilizer-like compound), treatment (in which microbes are applied to the pollutant), and digestion (in which microbes digest the pollutant). See Figure 1.
Building representations by combining generic components simplifies both knowledge engineering and knowledge-base maintenance. Rather than encode detailed information for each domain concept, knowledge engineers need only specify the components, and their inter-relationships, that comprise each concept. The resulting knowledge base is relatively easy to maintain because encoded knowledge is localized in the components. By changing the conversion component, for example, the knowledge engineer changes the representation of every domain concept that includes that component.
Motivated by earlier work by Batory, Goguen, and others in software engineering, we define a component as a triple <P,A,R> where:
Finally, we add a well-defined interface to each component, which enables reference to the objects that participate in the component's axioms. During the workshop, we hope to further explore connections between the application of component technologies to software engineering and knowledge-base construction.
Figure 1:
The concept bioremediation is a composition of multiple concepts, including specializations of conversion, treatment, and digestion.