A theme of model-based analogy (Bhatta & Goel, 1997c) is creating ontologies of useful abstractions by making claims about what kinds of inferences are needed and what kinds of knowledge are required to draw the needed inferences. This functional, top-down approach contrasts with more bottom-up architectural approaches to knowledge representation. For example, a bottom-up theory might specify that knowledge is represented as chunks or as productions. In my work, I postulate specific kinds of knowledge that need to be encoded to enable particular kinds of inferences. For example, the KRITIK system (Goel 1991a; 1991b; Goel et al., 1997) represented knowledge of the functioning of physical devices in terms of structure, behavior and function models (Chandrasekaran et al., 1993; Prabhakar & Goel, 1996). The primitives of the SBF language enabled the inferences needed to retrieve and adapt previous design cases to solve new design problems. Similarly, the IDEAL system, (Bhatta & Goel, 1997a; 1997b) used a language of generic physical principles and generic teleological mechanisms, which are useful units of analogical transfer in creative device design. Generic teleological mechanisms provide a taxonomy of functional and causal transformations to physical devices. In contrast, the ToRQUE system (Griffith et al., 1996; 2000) used a taxonomy of generic structural transformations that could be applied to physical systems. These transformations were found to be useful in modeling a protocol of a human subject solving a problem dealing with spring systems.
How can the wall be mapped to the the weed-trimmer? At first blush the two ideas have little in common. The trimmer is moving and encounters a barrier; the wall is not moving and is a barrier itself. One systematic way to characterize and represent these objects is with the Structure-Behavior-Function language. To represent devices in this language, one describes the functions first, then the relevant behaviors and structures needed to enable those functions.
I will also compare the SBF to Covlan as languages. KRITIK (Goel et al. 1997) has a vocabulary of changes which I can compare to transformations:
The SBF language does not, at this point, possess the power to express the geometrical relations necessary to represent the problems in the Craig data. For example, in SBF there is no sense of sides, or how they connect. The closest SBF concept to such a thing would be the start and end point of a wire in a circuit design domain, or the connecting points of a component in the KRITIK system.
In this dissertation I will expand SBF so that it can represent these problems, and compare the resultant representation to the visual one.