Re: W3C Web Ontology WG: Content Interoperability Use Casediscussion

Pat,

One general comment: the first section (point 1) was simply extracted
from mail messages mostly from Jim Hendler (and I think Jeff Heflin)
which tried to summarize the use case responses from the membership
homework up to that point. I just passed this on as a starting point and
it does not represent my thinking. 

Some additional comments below. 

Leo

Pat Hayes wrote:
> 
> >All,
> >
> >Welcome! I have condensed what I saw mentioned concerning content
> >interoperability so far and list it below. As you can see, these are
> >very sketchy working points so far.
> >
> >Let's send mail among our group to try to elaborate the use case, define
> >our terms, e.g., is "content interoperability" the same as "semantic
> >interoperability". I noticed in Jim Hendler's more recent message, that
> >he weakened/loosened the use case to be simply "Interoperability",
> >something I think we need to address immediately. I personally prefer
> >"content interoperability" as our focus, and largely equate this with
> >"semantic interoperability". But let me know what you all think.
> 
> I would like to see some short (paragraph length) statements of what
> these terms are supposed to mean. I feel like we are just tossing
> two-word phrases around.
> 
> >1) USE CASE: Content Interoperability (a/k/a/ agent markup) (compiled
> >from WOWG messages)
> >
> >- RDF has advantage over XML
> 
> Bad way to say this. After all, XML is purely syntactic and RDF is
> expressible in XML syntax; and the point is not to contrast RDF in
> particular with XML, but rather to say that one needs some kind of
> agreed semantics in order to even address interoperability, right?

I agree. RDF is the first rung for semantics, if you will, and in
principle enables XML and XML Schema schemas to begin to have a
semantics (beyond simple datatyping and local constraints). So as far as
content interoperability is concerned (as opposed to format, structural,
read: syntactic, interoperability -- which XML and XML Schema enable),
RDF/S and DAML+OIL, and the WebOnt language are the only web candidates
which can represent semantics. Outside the web, other languages include
KIF, conceptual graphs, various description logics, CYCL, UML.

> 
> >in allowing easy merging of content found
> >on different sites/resources,
> 
> Does RDF in fact do that? Seems to me that RDF just ducks the issue.
> If two different websites use incompatible terminology, RDF has no
> way to fix, or even to detect, the problem. DAML+OIL hasn't either ,
> of course.

Yes, I don't agree that RDF does this. No language currently does this.
Some folks use XSLT (style sheets) for creating mappings, but there is
really no intent to preserve/map semantics in any formal sense. A
question: should there be a language (or part of the WebOnt language)
for expressing semantic equivalence statements between 2 different
ontologies. I would say, yes. 

But automated detecting and mapping would seem to need an application.
And I think only semi-automating the process is possible, since it
generally requires a human being to interpret the semantics of both
sides and make the associations. Again, we would need to define content
interoperability and semantic interoperability (and assert they are
equivalent or not, for our purposes). To me, semantic interoperability
(between 2 ontologies) is looser than semantic equivalence. The latter
formally is something like sharing the same set/s (subsets) of formal
models. Perhaps our WebOnt language can aspire to expressing this
(assuming it has a formal model theory).  

> 
> >and the use of the combined sources.
> >DAML+OIL may offer additional advantage. Use cases include linking of
> >databases (DB schemas), coupling data to  pages, linking instance data
> >to ontologies.  Also allows linking of  ontology to ontology for mapping
> >of vocabulary, etc.
> 
> That is what these languages conspicuously lack, seems to me. How
> would one state a vocabulary mapping in DAML or RDF ? Eg suppose that
> one ontology treats 'parent' as a property, but another treats it as
> a class. There is no way to state the relationship in DAML+OIL (is
> there?).

Same as my previous comment, i.e., I don't think the languages can. I
think the languages enable you to formally specify the semantics you
want (i.e., the ontology concepts, relations, axioms, etc.) but you have
to map the DB schemas to the ontologies. When mapping from ontology to
ontology (property vs. class 'parent') there is no language (as far as I
know) that enables you to state the relationship. The IFF candidate of
the IEEE SUO aspires to this, i.e., enabling a statement of the
relationships between ontologies, but I won't yet suggest this (though I
like the IFF). 

> 
> >In general, the issue of semantic mapping.
> >
> >- Adapation of content to user/device. The content exists in some form
> >(ontology), and needs to be translated to another form (ontology) for
> >use by a different user or device.
> >
> >- Finally, the notion of condradiction/inconsistencies: when we
> >integrate heterogeneous content, then we need a means for detecting and
> >resolving inconsistencies.
> 
> Not clear that we should be talking about 'methods' here; and this
> seems to come with the semantics pretty much automatically, right?
> 
> Different point, though: there may be other kinds of ontological
> incompatibility than outright inconsistency. Eg if you use two
> different names in DAML+OIL for the same the same entity, that is
> consistent but may suggest the need for a vocabulary mapping.

Agree.

> 
> >
> >2) This is a message I sent out recently to the X318-news@nist.gov
> >group, who are interested in terminology standards, among other things,
> >suggesting one definition of "semantic interoperability". Specifically
> >the requester was working on the draft ISO standard on Requirements for
> >Electronic Health Records Architecture and referred to the 11179/3
> >standard which defined semantic interoperability as (paraphrased):
> >"ensuring that the receiver understands the data as intended by the
> >sender" -- which I do not like at all.
> 
> I agree, that is completely useless for ontologies, unless we are
> allowed to redefine what 'understands' means.
> 
> >One obvious issue is whether the
> >semantics will be machine intepretable or just a human agreement (as
> >most standards up to now have been).
> >
> >My reply:
> >We did a study of semantic interoperability in 1999 and defined it (from
> >the perspective of interoperating object-based systems) in our paper:
> >·       Obrst, L., G. Whittaker, A. Meng. Semantic Interoperability via
> >Context Interpretation, submitted to Context-99, Trento, Italy, April,
> >1999, invited poster session.
> >A related and shorter paper is:
> >·       Obrst, L., G. Whittaker, A. Meng. Semantic Context for Object
> >Exchange, AAAI Workshop on Context in AAI Applications, Orlando, FL,
> >July 19, 1999.
> >
> >>From the 1st paper:
> >Semantic interoperability is defined as the enablement of software
> >systems ... to interoperate at a level in which the exchange of
> >information is at the enterprise [or community] level. This means each
> >system (or object of a system) can map from its own conceptual model to
> >the conceptual model of other systems, thereby ensuring that the meaning
> >of their information is transmitted, accepted, understood, and used
> >across the enterprise [or community]. We argue that the primary way by
> >which semantic interoperability can be realized is by defining a notion
> >of context which includes the object to be exchanged and its internal
> >state, its interpretation with respect to both the source and the target
> >system object models, and the particular use of and intent for the
> >object in both the source and target systems.
> 
> Not to knock your work, Leo (BTW, are those papers online?), but I
> would be very unhappy at any kind of ontoweb standard that referred
> to a notion of 'context' without being EXTREMELY precise about what
> it means. That word has almost more baggage than 'model' or 'system',
> and it is used by people to refer to almost anything and everything.
> There are enough  social thinkers waiting to pounce on the ontoweb
> that it would be dangerous to use this term without having it very
> tightly battened down first.

No, these papers are still behind the firewall, but will be available
soon. I could also send them to you, if you're interested.

Yes, I know that "context" itself would need unpacking, since there are
so many meanings for this. In the papers we try to begin sketching a
notion of context, along the lines of (this was wrt objects in
object-oriented systems):

(1) Def. Context := <O1, O2, [[O1]], [[O2]], map([[O1]], [[O2]]>
    A. Object O1, a compiled runtime closure
    B. Object O2, a variable or virtual object, the kind of object that
the receiver system can               currently handle
    C. Interpretation of object O1 in sender system object model 
    D. Interpretation of object O2 in receiver system object model 
    E. Mapping between the interpretations of objects O1 and O2

This obviously is only a beginning and does not include "intent"
(pragmatics). And in the paper try to differentiate context from frame
of reference.

A more formal definition (and one we talk about in our paper) is given
in the Local Models Semantics papers of Giunchiglia and Ghidini, and
Bouquet. 

- Giunchiglia, Fausto; Bouquet, Paolo.  1998. A Context-Based Framework
for Mental Representation. Istituto per la Ricerca Scientifica e
Tecnologica (IRST), Trento, Italy, Technical report 9807-02, July, 1998.
- Giunchiglia, Fausto; Ghidini, Chiara.  1998. Local Models Semantics,
or Contextual Reasoning = Locality + Compatibility. Principles of
Knowledge Representation and Reasoning (KR'98), Proceedings of the Sixth
International Conference, Trento, Italy, June 2-5, 1998, Anthony Cohn,
Lenhart Schubert, Stuart Shapiro, eds., pp. 282-289.

But, in the end, I agree that "context" has its own baggage and we
should be picky about our definition.

> 
> Pat
> 
> --
> ---------------------------------------------------------------------
> IHMC                                    (850)434 8903   home
> 40 South Alcaniz St.                    (850)202 4416   office
> Pensacola,  FL 32501                    (850)202 4440   fax
> phayes@ai.uwf.edu
> http://www.coginst.uwf.edu/~phayes

-- 
_____________________________________________
Dr. Leo Obrst		The MITRE Corporation
mailto:lobrst@mitre.org Intelligent Information Management/Exploitation
Voice: 703-883-6770	7515 Colshire Drive, M/S W640
Fax: 703-883-1379       McLean, VA 22102-7508, USA

Received on Tuesday, 11 December 2001 17:53:04 UTC