Re: defaults

From: Enrico Motta <e.motta@open.ac.uk>
Subject: Re: defaults
Date: Tue, 22 Jan 2002 13:38:04 +0000

> Aside from all the technical problems, the problem with ``input 
> completion'' is that even when it works, it does not really implement 
> the idea of a default value.  Defaults are essentially a mechanism 
> for stating typical cases, which can be overridden in those 
> statistically rare examples in which teh default information does not 
> apply.   ``input completion'' is really a completion mechanism, which 
> is a different thing. The point about defaults is that they provide a 
> mechanism for dealing with incomplete knowledge.  For instance, "I am 
> told there is a new type of bird to be entered in teh KB, I have 
> never seen it, but I will assume that it can fly".  One can use 
> ``input completion'' in this scenario to ask the user: "Do you want 
> also to state that this new bird flies".  In many domains the end 
> result will be the same as using defaults. In other domains it won't 
> work, because it may not be a good idea to enforce completeness, 
> where completeness does not exist (i.e., the user may not have a clue 
> about whether the new type of bird flies, therefore he/she will be 
> unwilling to state the information explicitly in the KB, although 
> he/she may be happy to use such information as a default, in the 
> absence of contrary information).

Input completion is definitely a poor relation of general defaults, at
least in some aspects.  However, I claim that it is often adequate for the
job. 

> I agree with Guus that defaults are a good thing to have, given that 
> they are ubiquitous in applications, but of course only if some 
> 'clean' mechanism can be designed for OWL.

Aye, and there's the rub.  There are many views of what defaults are.
There are several clean technical views of general defaults, but each of
them has serious computational drawbacks, as well as the general
philosophical issues having to do with non-monotonicity.  There are many
unclean views of general defaults (consider, for example, overridable
inheritance with mulitiple, ordered parents) which, in my view, are totally
unsuitable for use in a representation formalism.

> Enrico

peter

Received on Tuesday, 22 January 2002 09:07:55 UTC