Re: P.S. Re: Model Theory

At 9:41 am -0500 7/1/02, tim finin wrote:
>"Peter F. Patel-Schneider" wrote:
>>  ... mechanism.  In the absence of a workable specification for a construct,
>>  e.g, reification, and, moreover, one that has good connection to the
>>  intended meaning of the construct, I am unwilling to include the construct
>>  in a representation language. ...
>>  PS:  Actually defaults do have several decent specifications.  However, the
>>  specifications have very bad computational properties, which make them
>>  problematic in a *useful* representation language.
>Many applications require some way of representing and reasoning with some
>constructs for which we don't yet have good to formalize or which can
>have bad computational properties.  If we are too inflexible, we will run up
>against the "worse is better" phenomenon and create a language that while
>arguably the best web KR language from many perspectives is also one which
>is just not used by practitioners.  Unfortunately, I don't have a concrete
>proposal for how to resolve this tension, but we should try to find one.
>Perhaps we can design a pure core that has good facilities and hooks
>to allow people to easily create and experiment with plug-ins and extensions.

Yes! And the "(we) don't yet have good (way) to formalize" is for me 
the crucial point.  We do not want bad representations, but to 
formalize and prove the properties of some of the trickier constructs 
take a very long time (it took over 10 years to get the first decent 
treatments of default reasoning).  So, again for pragmatic reasons, 
we should be prepared to consider the use of simpler specification 
techniques, when appropriate for some of the trickier constructs. 
Otherwise, as Tim says, we may end up with  too conservative a 


Received on Monday, 7 January 2002 10:22:38 UTC