Although domain-oriented knowledge-acquisition tools provide effective support for the development of knowledge-based systems, the task of implementing such domain-oriented tools is laborious [3][7][15]. Previous approaches to automatic generation of knowledge-acquisition tools, such as PROTÉGÉ-I, have relied mainly on method-oriented views for the specification, which, unfortunately, make the metatools specific to the problem-solving methods of the target knowledge-based systems. The abstract-architecture view and DOTS contribute to knowledge acquisition by enabling developers to construct domain-oriented knowledge-acquisition tools.
The DOTS project demonstrated that the abstract-architecture view can be used for method-independent specification of knowledge-acquisition tools, and that the tools generated automatically from such specifications are comparable to hand-crafted tools. For instance, we have used DOTS to generate a basic knowledge-acquisition tool for the Sisyphus VT task. Given a domain ontology, the development time for this knowledge-acquisition tool was about 9 hours. Because the problem definition is based on the VT system and the associated domain, the knowledge-acquisition tool generated performs a knowledge-acquisition task similar to that of SALT [10], a knowledge-acquisition tool developed for the original VT system [11].
Another general conclusion is that the abstract-architecture view is sufficient for specification of graphical knowledge-acquisition tools that provide knowledge-editing support for domain experts. In our approach, a generic knowledge-base generator, which is parameterized by transformation rules, provides flexibility in terms of output knowledge bases from the knowledge-acquisition tools.
Although we can remove most domain and method restrictions by using the abstract-architecture view, this approach uncovers other obstacles. Although DOTS is method independent, it is restricted in terms of the types of knowledge-acquisition tools that it can generate. In DOTS, the design space of the target knowledge-acquisition tools is restricted to knowledge-acquisition tools with a common basic architecture. DOTS cannot generate easily other knowledge-acquisition tools, such as repertory-grid based tools and machine-learning tools. However, this restriction has not been a severe hindrance in practical development, because we are mainly interested in generating domain-oriented knowledge-acquisition tools that support the experts' domain models.
We have applied DOTS to several domain tasks. For instance, we have used
DOTS to develop a knowledge-acquisition tool for a knowledge-based system
that troubleshoots DNA sequencing machines
[5]. Also, we are currently
formulating metaviews that bridge the PROTÉGÉ-I and DOTS approaches by
providing multiple perspectives. We are developing a metatool, DASH, that
supports knowledge-acquisition analysis, and that helps developers to
design target knowledge-acquisition tools from domain ontologies (i.e.,
class definitions).