From: Raphael Hoffmann (raphaelh_at_cs.washington.edu)
Date: Wed Oct 22 2003 - 11:28:31 PDT
"Two Theses of Knowledge Representation" by Jon Doyle and Ramesh S. Patil
Summary:
The paper criticizes the Levesque and Brachman approach of restricting
languages of general purpose knowledge representation systems to gain
efficiency, and shows several examples of real-world problems that require a
fully expressive language.
Main ideas:
The assumptions made by Lesesque and Brachman do not reflect the
requirements of many real-world general purpose knowledge representation
systems and therefore, their restricted language approach leads to problems
when modeling several common applications.
The authors suggest using fully expressive languages, applying more relaxed
measures in evaluating the performance and tolerating lower worst-case
performance.
Flaws:
The authors focus merely on the Levesque and Brachman approach and hardly
address other research work that has been done in the field. Their
comparison is obviously biased and strengths of the Levesque and Brachman
approach are not objectively presented.
Furthermore, they discuss strengths and weaknesses of general purpose
knowledge representation systems without directly considering the various
applications and its various and varying demands. Certainly, Levesque and
Brachman had particular applications in their mind when they developed their
theories. The paper does not show the backgrounds of the different
approaches.
Future Research:
The authors talk about performance issues regarding knowledge representation
systems without ever mentioning the exact computational complexity and
especially without showing why and when this is an issue. Performance
certainly depends on the siye of the knowledge base. But perhaps certain
low-performance constructs can be tolerated up to a specific KB size.
When implementing a general purpose knowledge base, one has to make several
design decisions, for instance regarding the expressiveness of the language.
However, the requirements definetly depend on the type of application and
general recommendations like in the paper are probably not ideal for every
application. Further research could assess more application-specific design
criteria, taking into account the individual requirements.
This archive was generated by hypermail 2.1.6 : Wed Oct 22 2003 - 11:28:11 PDT