ABSTRACT
The Subset Principle (SP) is invoked to prevent overgeneralization. The transitions of the automata are defined in terms of the states of the daughters of nodes in the tree, the label of the mother node, and the state assigned to the mother node as a result of the transition. This automaton walks over the tree bottom-up and finally ends at the root node of the tree. If the parameter values are already innately ordered for the learner in some way, then, of course, the SP can be instantiated very easily. Complexity of computation is obviously crucial in determining the appropriateness of learning strategies. The SP has an intuitive appeal, but there are problems in instantiating it. These considerations have led to two research strategies—one trying to adjust the definitions and strategies of instantiation of SP, a conservative strategy, and the other a more general and radical one that just ignores the SP.
