From: Russell Power (rjpower_at_u.washington.edu)
Date: Wed Oct 22 2003 - 00:13:53 PDT
Authors: Jon Doyle and Ramesh Patil
Title: Two Theses of Knowledge Representation
The authors present arguments that the current vogue in knowledge
representation - restricting languages for to ensure efficient operation -
has fundamental problems that restrict its use in a variety of contexts.
The paper describes the set of KR systems developed from the principles
espoused by Levesque and Brachman, who argue in previous work that KR
languages should be restricted to provide 'efficient' responses to queries,
where efficiency is measured by the worst-case performance of the system.
The authors provide a set of arguments against this philosophy, most hinging
on the impracticality of such systems in practice, in terms of both input
limitations and internal functionality. The authors also provide a short
list of suggestions for modifications to the restrictive paradigm that would
allow for a more expressive environment.
Unfortunately the conclusions drawn in the paper are for the most part
obvious. It seems quite clear that if we restrict the way in which a KR
system works, it will put limitations on what the system can represent. The
authors go into exhaustive detail examining the list of cases in which
restricted systems have problems, but this is akin to drawing up a list of
all the flaws we find in cars - despite the fact that problems exist, we
still find them useful.
The paper presents a large series of troubles that the authors have with the
Levesque/Brachman method of KR, but they only provide a small sampling of
possible solutions. Thus, the paper comes across as being a set of nitpicks
targeted at a specific category of KR systems, rather then an appeal for a
philisophical change as to how KR systems should behave.
On a side note, I found it interesting that the designers of the OWL
language appear to have thought over the potential effiency/completeness
concerns mentioned in this paper as well, as the breakdown of OWL into
sublanguages attests to: http://www.w3.org/TR/owl-ref/#Sublanguages.
One potential research topic that the paper brings up is an investigation
into the feasibility of developing a system that allowed a greater variety
of input, but which would be able to categorize queries according to the
efficiency of their lookup and inform the user. Alternatively, one could
imagine research into providing an efficient and general framework by
loosening completeness and/or soundness constraints, and providing output as
to confidence in a proposed response.
This archive was generated by hypermail 2.1.6 : Wed Oct 22 2003 - 00:14:08 PDT