W3C home > Mailing lists > Public > www-rdf-logic@w3.org > March 2003

Re: An exciting inference?

From: Leo Obrst <lobrst@mitre.org>
Date: Sat, 15 Mar 2003 20:17:35 -0500
Message-ID: <3E73D0AE.5196AEC0@mitre.org>
To: "Roger L. Costello" <costello@mitre.org>
Cc: www-rdf-logic@w3.org

In general, I think, you want 1) two different extensions (instances) to be
characterized by two separate intensions (descriptions) and 2) those two
intensions to be linked and 3) thereby those two extensions to be judged
referentially equivalent (under two different intensions). As far as I know,
that kind of inference is not possible unless there are additional "rules"
(axioms) which associate the two intensions somehow (which there could well
be), i.e.,
if there is a ticketing(PoliceOfficer, Speeder) then law_violator(Speeder)
if there is a robbery(PoliceOfficer, Robber) then law_violator(Robber)

Then you can at least query based on the law_violator predicate and see the
joint set of possible Speeders and Robbers -- which would help. However,
referential (extensional) identity is more tricky. Even if you are using the
same knowledge representation language. What if Speeder individual's name is
listed as "John Smith" and Robber individual's name is "Harry Foobar"? What
if John's and Harry's other attributed properties are completely different
(i.e., one's weight is 5'1'' and the other's is 6'11'', weights, addresses,
relatives, event participation, etc. are different)?

You can see that if you take property restriction into consideration, i.e.,
those who weigh less than 250 lbs, those who are less than 6'11'', those who
were arrested on the East Side near Main St. in the last two weeks, those
who have an alias of John or Harry, etc., that you can get arbitrarily close
to a subset of individuals (if all the information is totally accurate).

In simpler cases, the inference might be easier, i.e., to determine that
address(Address1, Tenant, Time1) NOT_EQUAL_TO address(Address1, Tenant,
Time2) where the capitalized arguments are variables. Maybe the Chinese
embassy in Yugoslavia would not have gotten bombed, for example. However,
you would still need to know that the first Tenant is possibly not the
second Tenant. Even if you didn't know who the tenant was, the time change
itself could flag a warning (and you could have a predicate that described
that).  Temporal and spatial reasoning, as someone suggested. Also the fact
that no person can be at two places at the same time would narrow things
(though of course organizations and many other things can).  Reference
resolution can be a hell of a problem.

Using a language like OWL, however, you can at least express theories about
Speeders and Robbers and other theories that relate them.


ps. This is what passes as "link analysis" sometimes.

"Roger L. Costello" wrote:

> Hi Folks,
> I am seeking a *simple* example that enables an *impressive* inference
> to be made.
> Below is the outline of a scenario that is universally understandable
> and would be very compelling if the inference shown below could be made.
> The Robber and the Speeder
> Let's suppose that yesterday two incidents occurred in Boston - a
> robbery and separately (at a later time) a person was ticketed for
> speeding.  The police officer at the robbery wrote up his report and
> submited it into the police database.  Likewise, the police officer that
> ticketed the person for speeding wrote up a report and entered it into
> the police database.
> At the end of the day an OWL tool automatically pairwise checks all the
> reports entered into the police database, looking for any
> "relationships".
> Let's suppose that the Robber and the Speeder are one and the same.  If
> the OWL tool could infer that the Robber and the Speeder are the same
> person then that would be most impressive (and would provide compelling
> evidence of the usefulness of OWL).
> Okay, that's the scenario outline.  Can someone fill in the details?
> Remember, it must be *simple* and it must lead to this inference:
>    The Robber and the Speeder are the same person.
> /Roger

Dr. Leo Obrst  The MITRE Corporation
mailto:lobrst@mitre.org Intelligent Information Management/Exploitation
Voice: 703-883-6770 7515 Colshire Drive, M/S H305
Fax: 703-883-1379       McLean, VA 22102-7508, USA
Received on Saturday, 15 March 2003 20:18:36 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 2 March 2016 11:10:39 UTC