- From: Drew McDermott <drew.mcdermott@yale.edu>
- Date: Sun, 20 Jul 2003 08:52:19 -0400 (EDT)
- To: sandro@w3.org
- CC: drew.mcdermott@yale.edu, abcharl@keyworld.net, www-rdf-logic@w3.org
[me]
> Why is SW reasoning monotonic and open-worldish? Who will enforce
> these constraints?
[Sandro]
I think people are using these terms almost as buzzwords to remind
folks that some common kinds of reasoning (as in Prolog: closed world,
unique names, non-mon) do not allow ad hoc composibility. I'd say a
basic goal for the SemWeb is that agents be able to merge knowledge
from multiple sources with relative ease, and as far as I can tell,
that means using monotonic reasoning with no CWA or UNA.
Monotonic reasoning is essentially deductive reasoning, and most
interesting inference problems are not deductive. Suppose an agent is
doing scheduling of a fleet of trucks, using information from various
sources (retail stores, highway-condition information, weather
predictions, ...). It will generate a schedule from a given set of
facts, and given more facts it will usually generate a different
schedule. One could argue that scheduling is not an inference
problem, but no matter what we call it, the same issues about
monotonicity, UNAs, and CWAs will arise in connection with it.
Another point: It's in merging information from multiple sources that
the UNA becomes most useful, in the following form: if you can't prove
that two objects are equal, then assume they're distinct. For
example, if I have a list of employees obtained from company A, and a
similar list obtained from company B, then it's reasonable to assume
that if v works for A and w works for B, then v is not equal to w.
--
-- Drew McDermott
Received on Sunday, 20 July 2003 08:52:25 UTC