Advertisement

Knowledge Revision as a Concept Formation Context

  • Stefan Wrobel
Chapter

Abstract

After discussing the technical aspects of what concepts are and how they are represented in the preceding chapter, let us now return to the issue of how concepts are formed. In chapter 2, we had concluded that one of the most important sources of constraints on concept formation is its embedding in a particular context. Based on those psychological arguments, our choice was to examine a demand-driven concept formation approach, i.e., an approach that forms concepts only when there is a specific need for a new concept arising out of the problem solving activities of the system.

Keywords

Revision Operation Choice Function Inference Engine Revision Operator Base Contraction 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Here, we define the revision operations with respect to single fact revisions; in section 4.5.3, we generalize them to perform multiple revisions at the same time.Google Scholar
  2. 2.
    is to be read as “exclusive or”. This third operation is necessary because MOBAL permits additions which are inconsistent at first and are resolved later on, see below.Google Scholar
  3. 3.
    This is a simplified version of the Levi identity known from theory change work ([Levi, 1977] foll. [Nebel, 1989]).Google Scholar
  4. 4.
    Two less important supplementary postulates that deal with the properties of conjunction under removal are also sometimes used.Google Scholar
  5. 6.
    The term besserwisser is due to [Gärdenfors, 1988].Google Scholar
  6. 14.
    Clearly, this measure is not without problems, as the number of applications of a rule directly depends on what other rules are in the theory. Thus a rule that appears very strong before revision may end up very weak if another rule that produced the former rule’s antecedents is removed.Google Scholar
  7. 15.
    Thanks to J.U. Kietz for pointing this out.Google Scholar
  8. 16.
    This also implies that ø will always keep the side of a contradiction that is supported by an input fact, which corresponds well with our intentions.Google Scholar
  9. 17.
    Without assuming maximal predicate arities.Google Scholar
  10. 18.
    This table shows only the reformulation operators that use existing concepts. In chapter 5, we define additional operators that use CLT to introduce new concepts for reformulation.Google Scholar
  11. 19.
    In the following, we omit the variable list of the support set.Google Scholar
  12. 20.
    As pointed out in [Emde, 1991], using supports sets instead of premises also means that rules which have been revised are easy to find in the knowledge base if this should become necessary.Google Scholar
  13. 21.
    In chapter 5, this table will be extended to include the concept formation operators.Google Scholar
  14. 22.
    If desired, such rules can also be learned by letting RDT or some other learning algorithm analyze the manually assigned access rights.Google Scholar
  15. 23.
    It can be shown that different programming languages change the size of this quantity only by an additive factor.Google Scholar
  16. 24.
    [Srinivasan et al., 1994] point out that equivalently this evaluation could be incorporated into the CWS specialization directly.Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 1994

Authors and Affiliations

  • Stefan Wrobel
    • 1
  1. 1.GMDSankt AugustinGermany

Personalised recommendations