From: Jessica Kristan Miller (jessica_at_cs.washington.edu)
Date: Wed Oct 22 2003 - 12:31:46 PDT
Paper Reviewed: Two Theses of Knowledge Representation
Paper Author: Jon Doyle and Ramesh S. Patil
Reviewed By: Jessica Miller
Summary:
In this paper the authors argue against restricting or omitting language
constructs which require non-polynomial worst-case response times in favor
of developing a fully expressive general purpose language used for
knowledge representation.
Two Important Most Ideas:
(1) The authors argue that if languages are restricted by omitting
language constructs where the worst-case response time is poor then you
limit the expressibility of the language significantly. The authors give
many examples of types concepts that cannot be expressed by language
restrictions in order to show the extent of this limitation. In fact, the
authors argue that Levesque and Brachman's ideas about restriction of
constructs defeat the purpose of having general purpose representation at
all.
(2) The author's also argue that worst-case time efficiency, logical
soundness and completeness are poor metrics to evaluate the utility of
general purpose representation services. I found this particularly
interesting because these seem to be such well accepted metrics and in
some way it was novel for the authors to argue against using them in this
context. However, I would like to hear more about what metrics they would
use (yes, they talk about value and benefit as being metrics, but how do
you use these quantitatively?).
Largest Flaw in the Paper:
Although I am sure the authors are very intelligent people, I feel they
were somewhat unscientific about the way they went about their argument.
I would have liked to see some kind of statistics or measurement for their
arguments about expressibility in addition to their passionate argument --
anything to show their way was better than Levesque and Brachman's.
Instead, the authors seem to "argue away" the other thesis.
Two Open Research Questions:
(1) The authors themselves suggest that perhaps instead of totally
forbidding "dangerous constructs" maybe there is a way to limit them
instead. It might be interesting to take a look at which constructs would
benefit from this in terms of expressibility and efficiency.
(2) Another possible subject to investigate is how and when to forbid
constructs completely versus allowing them in order to augment
expressibility. In other words, is there any reasonable way to marry the
two theses to get the most benefit from the two stances.
This archive was generated by hypermail 2.1.6 : Wed Oct 22 2003 - 12:31:46 PDT