Abstract
In some previous papers I have made efforts at drawing attention to some crucial representational problems standing in the way of learning and at clarifying the technical underpinnings of these. These problems occur both in learning from examples and in learning from explanations. In the introduction to this paper these problems are reiterated in summary. It is also pointed out that they have direct implications in the way one attaches statistical significance to concepts learned from examples.
Later sections to this paper describes an effort being made at present to develop a learning algorithm which uses a flexible language of representation. The algorithm has a limited amount of ability to modify the representation language. The limitations of the method are also discussed.
Get full access to this article
View all access options for this article.