In this paper, we propose a probabilistic framework for memory-based r
easoning (MBR). The framework allows us to clarify the technical merit
s and limitations of several recently published MBR methods and to des
ign new variants. The proposed computational framework consists of thr
ee components: a specification language to define an adaptive notion o
f relevant context for a query; mechanisms for retrieving this context
; and local learning procedures that are used to induce the desired ac
tion from this context. We primarily focus on actions in the form of a
classification. Based on the framework we derive several analytical a
nd empirical results that shed light on MBR algorithms. We introduce t
he notion of an MBR transform, and discuss its utility for learning al
gorithms. We also provide several perspectives on memory-based reasoni
ng from a multi-disciplinary point of view. (C) 1998 Published by Else
vier Science B.V.