Specification Model

The abstract-architecture view comprises four component types, each of which constitutes a stage in the acquisition and generation of knowledge bases in knowledge-acquisition tools (see Figure 1). In this model, knowledge editors handle the user dialog; for instance, there are form-based and graph-oriented knowledge editors. Also, there are knowledge modules that represent the knowledge acquired by these knowledge editors. The relationships among the knowledge editors and the knowledge modules are defined by update rules. Finally, there is a description language of transformation rules that allows the developer to specify the knowledge-base generator of the target knowledge-acquisition tool. Specifically, the four component types in the abstract-architecture view are defined as follows:

1. Knowledge editors: The knowledge editors allow experts to enter and edit domain knowledge according to the experts' conceptual domain model. Examples of knowledge editors are domain-specific forms and graph editors. The knowledge editors are part of the user interface of the target knowledge-acquisition tool; they operate on the internal knowledge representation in the knowledge-acquisition tool, and provide views of the representation.

   In DOTS, developers can custom tailor the user interface of target knowledge-acquisition tools by specifying and refining knowledge editors, such as menu layouts, window-system behavior, and editor properties (see Figure 2). DOTS provides predefined types of knowledge editors for various types of knowledge, and also can allow the introduction of user-defined knowledge editors. Figure 3 shows a form-layout editor provided by DOTS. Developers use this tool to design layouts for form-based knowledge editors. The resulting form-based knowledge editor is shown in Figure 4. Note that this example is consistent with the fixes in the Sisyphus VT task.

   

   Figure 2. The specification of knowledge editors in DOTS. A sample user-defined knowledge editor (fix-ke) is highlighted in the knowledge-editor hierarchy. The developer uses an editor comprising subeditors for slots, menu layout, window properties, and attached menu to define the details of the knowledge editor (lower right).

   

   Figure 3. The editing of the layout for a form-based knowledge editor. DOTS provides this graphical tool, which developers can use to custom tailor forms for the domain-in this case, elevator configuration [11].

   

   Figure 4. The resulting form-based knowledge editor in the target knowledge-acquisition tool. In this example, the experts use this form to specify fixes for constraint violations.

2. Knowledge modules: The knowledge modules provide encapsulation and manipulation of the internal knowledge representation of the target knowledge-acquisition tools. The information that the expert enters in the knowledge editors is stored in knowledge modules. Furthermore, the knowledge modules serve as an intermediate representation in the transformation of the knowledge acquired into the target knowledge base (see Figure 1).

3. Update rules: Update rules preserve consistency among the knowledge editors and the knowledge modules in the target knowledge-acquisition tools. The specification of the update rules defines the mapping between the knowledge presented and edited in the knowledge editors and the knowledge modules.

4. Transformation rules: The knowledge-base generator of the target knowledge-acquisition tool generates the knowledge bases. In the DOTS approach, the knowledge-base generator is based on transformation rules that map knowledge represented by the knowledge modules into the appropriate knowledge-base structures (e.g., classes, instances, and rules). Generally, the rule preconditions of the transformation rules refer to the contents of knowledge modules, whereas the rule conclusions refer to structures in target knowledge bases. In other words, the transformation rules define the denotational semantics for the knowledge that the target knowledge-acquisition tools acquire, because, currently, the knowledge-acquisition tools do not interpret the knowledge modules.