The description identification problem

Abstract

In this note we introduce a notion of description identification which generalises both concept learning (as conceptualised by Winston, Young et al. and Mitchell) and also incremental description refinement (as described by Bobrow and Webber). Assuming certain properties of the description space involved, there is an algorithm for solving the more general description identification problem, which extends the version space strategy of Mitchell, and we present this. The work described here can be regarded as a further formalisation and development of the work of Mitchell on version space representation and the work of Young et al. and Plotkin on description spaces. Plotkin's unpublished work presented a similar, though slightly more restricted, approach to the concept learning part of our subject and gave examples of types of description spaces that connect the work with the earlier work of Winston.