[topicmapmail] Inference rules

Tue, 25 May 2004 12:07:33 +0200

On Tue, May 18, 2004 at 06:14:46PM +0200, Stefan Henke wrote:
>                                Murray, you are right that it is possible to
> express these inference rules in topic maps as topics and associations. But
> I think it is not a good idea due to reasons of constistence. Imagine the
> example of Eric using the cousin rule. If you model this rule as an
> association, each time a new person is added as topic, this association has
> to be added by the user that added the person topic. So, if it is not added,
> the map would not be in a consistent state. By defining it as a rule, it
> would be no problem as this rule is valid for all topics.

Stefan,

Consistency might be an issue, but I have a more serious concern
regarding inference rules represented within a topic map.

To make that work, as Murray points out correctly, the constraint (the
logical formula constraining the map or providing the inference rule)
has to represented using PSI'ed topics and using a canonical
serialization into a 'TM' format.  "Unreadable" would be a rather
euphemistic term for the outcome. This is unusable.

Now some people might say "this is all a matter of tools" and the
users (= humans) are never supposed to see that mess. I would argue
that a forcing users into toolkits is maybe good for vendors, not for
users.

Murray's argument, that inferencing should not be built in into TMs
is good and bad. The good part is that everyone simply can do what suits
him/her. The bad part is that there will be no interoperability.

> Maybe I?m getting something wrong or I?m in a totally wrong direction, but I
> think inferencing is a very powerfull tool to "create" new knowledge.
> Eric, you mentioned TMCL in the context of inference rules. I?m not familiar
> with TMCL, do you think it is possible to define such rules with it?

If we accept the framework that

  - maps to hold factual knowledge (topics, associations about our subjects)

  - a constraint document constrains the form and content of maps and thus
    defines what the 'application is all about' ( = ontology)

  - and that queries extract information and regenerate content

then TMCL as the constraint language might be the natural place for
that. I would not over-interpret the fact that some vendors have put
inference rules within the query process. It makes sense to have
ad-hoc ontologies there, I guess.

\rho

> -----Urspr?ngliche Nachricht-----
> Von: topicmapmail-admin@infoloom.com
> [mailto:topicmapmail-admin@infoloom.com]Im Auftrag von Freese, Eric D.
> (LNG-DAY)
> Gesendet: Dienstag, 18. Mai 2004 14:54
> An: 'Stefan Henke'; topicmapmail@infoloom.com
> Betreff: RE: [topicmapmail] Inference rules
> 
> 
> I believe inference rules can play an extremely important role in topic
> maps.  I envision them mainly being used to build new associations between
> topics based on certain conditions.  In a genealogy topic map, for example,
> you could define a rule that would infer that Bill and Sally are cousins if
> one of each of their parents are siblings.
> 
> I also envision them being used to allow a system to interpret the meaning
> of any association with more than 2 members in 2 roles.  For example imagine
> a meeting association association where Tom, Dick and Harry play the role of
> attendee, San Francisco plays the role of location, "inference rules" was
> the meeting subject, and 18 May 2004 plays the role of event date.  While a
> human could make some inferences about this meeting and other facts, a
> machine cannot.  Inference rules could be written to say that "[attendee]
> was in [location] on [eventDate]", "[attendee] discussed [meetingSubject]
> with [attendee]", and so on.  Once the TMCL standard is done, it will
> hopefully be possible to write inference rules based on the constraints
> placed the roles and members in an association.
> 
> A couple of years ago I developed a system called SemanText
> (www.semantext.com) that included a rule-based inference engine that could
> do both of these things.  When processing the inference rules, it added new
> associations to the topic map in a special scope so that it was possible to
> differentiate between supplied and inferred knowledge.  It is discussed in
> chapter 10 of the XML Topic Maps book along with illustrations of the
> interface to building the rules.  I am sad to say, however, that the
> software is somewhat outdated.  It was built on tmproc which was never (as
> far as I know) updated to handle XTM.  I also haven't updated it to the 2.x
> versions of Python.  I am looking at developing the next generation of the
> software, though, to be built in Java on an XML database.  Time is my main
> limiting factor now.
> 
> > -----Original Message-----
> > From: topicmapmail-admin@infoloom.com
> > [mailto:topicmapmail-admin@infoloom.com]On Behalf Of Stefan Henke
> > Sent: Tuesday, May 18, 2004 5:58 AM
> > To: topicmapmail@infoloom.com
> > Subject: [topicmapmail] Inference rules
> >
> >
> > Hi,
> > I have a very basic question about topic maps. Which role do
> > inference rules
> > play in the world of topic maps? I think in the field of
> > ontologies, they
> > play a quite important role. In the last time, I had a look
> > at the tolog
> > query language. It allows to define inference rules which are
> > used for the
> > query. But the rules are not getting part of the topic map
> > itself. So, they
> > are not used in other queries. Are there any plans to build
> > in some concepts
> > for inference rules in topic maps?
> >
> > Regards,
> > Stefan
> >
> > _______________________________________________
> > topicmapmail mailing list
> > topicmapmail@infoloom.com
> > http://www.infoloom.com/mailman/listinfo/topicmapmail
> >
> _______________________________________________
> topicmapmail mailing list
> topicmapmail@infoloom.com
> http://www.infoloom.com/mailman/listinfo/topicmapmail
> 
> _______________________________________________
> topicmapmail mailing list
> topicmapmail@infoloom.com
> http://www.infoloom.com/mailman/listinfo/topicmapmail
> 