ABSTRACTION

Abstraction is characterized by a generalization of a representation and usually involves some loss of information. Unlike a derivation, an exact recreation of the source object cannot be obtained. The transformation is non-monotonic. The transformations in this paper were largely of abstraction. Simple features were abstracted from FEV graphs, complex features were abstracted from simple feature graphs. Rules were abstracted from expert experience.

An abstraction can be considered as a symbol, where the underlying object is related only by a declaration that the object is what it represents. It is not an icon because the meaning of the original object cannot be obtained from the abstraction. And it is not an index because nothing is pointed to.

On an applied level, abstractions are more readily obtained than derivations because only one transformation has to be provided for the system, whereas the derivation system requires the generative and abstractive transformations.

Abstractions are preferable to mapping because they do not require the overhead of maintaining links between representations. Abstractions may involve the addition of information from outside sources and may lose information from the original representation. As such, the transformation system is non-monotonic.

Information required in the destination representation has to be explicitly gleaned from the source representation. The explication of information is a requirement of abstraction and thereby requires some foresight in the use of the abstraction.