Re: [OEP,ALL] Potential topics for OEP notes

Chris, All

Can I add something like -

a) options for using Ontologies in applications - whether in OWL or RDF - this is the thing the SWBP  really has to crack. Picking up where Classes as Values left off.  I am not sure where pointers to specific tricks with current technology fit in, but in the "deployment" part of SWBP&D I think many people would welcome a list of tool combinations that were known to work, however time limited that list will inevitably be.  I am certainly not in a position to produce such a list; I don't think the list per se is really part of OEP, but we need someplace where we coordinate the principles the notes with practice..

b) When to use a reasoner and normalisation.

c) I am not sure whether it is a note, but it is worth pointing people at 'common pitfalls'. One contribution towards that is


Other comments below...



Christopher Welty wrote:

>
> ...

> The partOf relation.  There really isn't that much that can be "said" in OWL (and therefore less in RDF) regarding the typical axiomatizations of partOf, but knowing the different kinds of partOf relations and what they are supposed to mean would be useful.  I'm hoping that some subset of Nicola, Alan, and I can take the lead on this one, but I also see the need for a couple of notes here, so I think this needs further discussion.  For example Deborah expressed interest in a simpler note (less ambitious but quicker turnaround) on geographical containment.

I am happy to take a major role in this, but not until after 6 Dec (ISWC plus a major week-long teaching stint - some of which will include material on partonomy.)  I shall also be writing a paper on the topic or probably two.

On the other hand, partonomy is a big topic.   The trick is to keep it simple for the simple cases. I suggest that we need at least two kinds of notes or note fragments.

1)    Notes laying out the classic distinctions, pointing to the the literature on merology, and pointing out things like that most users of partonomy probably want something that is time specific - X is a part of Y at some implied time T (the type is a part of the car now, but it may not be after the tyre has been changed) - or normative (Xs are considered parts of Ys).  You need one or the other to avoid getting into issues about amputated fingers, cat's tails, etc.  Also I think we have to say that this area is far from settled so we are giving guidance on workable principles plus caveats for controversies.

2)    Implementation mechanisms. Transitive properties for simple things. SEP triples are related trickse..

    RegionOfFrance = France or restriction(is_geographical_region_of someValuesFrom(France))]

property hierarchies showing different relations between containment, location and partonomy.

3) Also warnings that with current classifiers (possibly excepting FaCT++ but we aren't sure yet) large ontologies containing extensive networks linked by both has_part and is_part_of (or any other transitive relation and its inverse) are potentially combinatorially explosive.  If anybody does try to use a classifier it is disconcerting to see what seemed to work for a toy run indefinitely for something real.



>
>
> Units and measures.  There has been some work on this, including in Cyc, Tom Gruber's ontology in Ontolingua, and Helena Sofia-Pinto did a nice one for the old SUO effort.  Evan was interested in this and it certainly makes sense to have someone at NIST do it.
>
> Subjects.  The notion of what a subject "is" and what the "subjectOf" relation means can be quite confusing.  I have done a lot of work on this and am willing to take this one on, however I will want to do one at a time.
>
> Time.  Jerry Hobbs has done a very thorough job putting together a consensus ontology of time based on a lot of existing time ontologies, most of which draw from the Allen calculus.  The ontology is expressed in FOL (KIF), but there are (necessarily simplified) DAML+OIL and OWL ("OWL-Time") versions  available.  Jerry has expressed interest in seeing this as a W3C note.

I'd be very interested in seeing this.

>
>
> Fluents.  Closely tied to the notion of time is being able to say that a binary property "holds" for a time. e.g. one may want to say that "Chris is a member of the W3C from Sept, 2004 - Sept 2005".  A property like memberOf is a fluent because it can be said to hold at a time (this is not strictly a correct definition, but it will suffice).  While OWL-Time let's you represent a time interval like "Sept, 2004-Sept, 2005", it remains neutral wrt what happens at or during such a time interval.  The typical move in FOL is to use a function or add an argument to the predicate, e.g. memberOf(Chris, W3C, time-interval-1), however clearly we can't do that in OWL or RDF, since we are limited to binary predicates.  One solution is to go for full reification of fluents, as in the exsiting not on n-ary relations, however there are some other choices.  I'm hoping I can get Pat Hayes and Richard Fikes to work with me on this one.

An important distinction for the medical community is whether it is the fluent that is reified or the observation of the fluent, i.e. whether we have

(X as observed by O at time T) is_member_of Group or

  X is_member_of (Group at time T)

The first is the way to describe a log book of observations such as the medical record; the second is probably more important for models of language phenomena such as the classic "The King of France".

The other alternative which fits closely with some indexed notions of parthood such as barry Smith's, is

   X (is_member_of at_time t) Group.

but that involves reifying 'is_member_of'

>
>
> On the side of "ontology engineering":
>
> Ontology 101 tutorial specifically for OWL/RDF.

I'd hope that some of the pizza tutorial material could get into this, or perhaps beside it.
There is an interesting comparison - I am not sure it is for a best practice note - that the pizzas and wines require different paradigms, or at least different emphasis.  Pizzas are primarily about construction - pizzas have someValuesFrom topping.  Wines are primarily about exclusion - CarbernetSavignon is made only (allValuesFrom) CarbernetSauvignonGrapes.

>
>
> I think a note to help orient people on the role OWL and RDF in semantic integration is critical, I get pinged on that regularly.  I lot of people think OWL is the silver bullet for semantic integration (I suggested at ISWC last year that semantic integration is a mountain, not a werewolf, and OWL is, at best, a small silver chisel). There was just a Dagstuhl symposium on this subject in general (i.e. not specific to OWL), and special issues of AI Magazine and Sigmod record coming out as well.  I hope Natasha and/or MikeU will take the lead on such a note.

>
>
> People who know what "ontology" and "semantics" actually mean (in the much larger world outside of computer science), often ask why the two have become nearly synonymous on the semantic web.  Personally, I think its a fair question and a short note on why we're so confused would be worthwhile.  Maybe this goes in another task force (wasn't there a clean up the mess we've made task force?)
>
> We're open to other suggestions.
>
> -Chris (OEP co-co)
>
> Dr. Christopher A. Welty, Knowledge Structures Group
> IBM Watson Research Center, 19 Skyline Dr., Hawthorne, NY  10532     USA
> Voice: +1 914.784.7055,  IBM T/L: 863.7055, Fax: +1 914.784.7455
> Email: welty@watson.ibm.com, Web: http://www.research.ibm.com/people/w/welty/

--
Alan L Rector
Professor of Medical Informatics
Department of Computer Science
University of Manchester
Manchester M13 9PL, UK
TEL: +44-161-275-6188/6149/7183
FAX: +44-161-275-6236/6204
Room: 2.88a, Kilburn Building
email: rector@cs.man.ac.uk
web: www.cs.man.ac.uk/mig
        www.opengalen.org
        www.clinical-escience.org
        www.co-ode.org