We have found that the distinction between instances and classes is frequently blurred, and this gives rise to undesirable ambiguities and misinterpretations of represented knowledge. A more disciplined approach to the use and interpretation of instance and class is required to eliminate these problems.
We begin by formally stating the restriction that instances and classes must be kept distinct. We discuss the importance of this restriction and show that problems do arise when it is violated. We further discuss a common inference used when modeling with instances and taxonomic relationships, and show that it is important to modeling domains. We then consider objects whose intuitive conceptualization requires that they be represented as a class and an instance. In conventional modeling systems, this requirement seems to call for the dismissal of either our important restriction or our important inference. We show how these objects can be represented without causing such problems by introducing a construct called the spanning object. We conclude with previous intuitions of spanning objects and a simple example illustrating their use in an existing domain modeling effort.
Generated with CERN WebMaker