Minutes of SWSA Telecon of July 6, 2004
Mike Dean, Tim Finin, Mike Huhns, Stuart Williams

Mike Huhns
Two services need to interoperate, using different terminology. Mappings could exist pairwise.
If n services, dont want n**2 pairwise mappings. - try to use common interlingua. Mappings should
have its own terminology, relating to a global terminology.  

Tim: Like an interlingua, as in MCC project.

MH: Carnot at MCC used CYC as an interlingua, to relate database schemas

Tim Finin. That runs counter to SW dogma. No single interlingua. Hve to deal with at least handful
of common ontologies. 

MH: I agree, there will be lots of ontologies for subdomains. There will have to be mappings. 

MB: These have to be published on the SW. Not sure if this should be
treated as a SW issue or a SWS Issue.  

TF: Cant do it in OWL

MB: Mappings can be done using a rule language

MD: SWRL was intended for this, OWL equivalences as mappings.

TF: But Mappings will be lossy. Need more expressive lang.

MB: Yes, need priority based or default rules to express exceptions. 
What do we want to say about the architectural supports. We will need some kind of registry system
to index mappings between ontologies since the ontologies themselves wont point to the other langs/onts
they can be rewritten as. Also want to support client/server/middleagent translations

MH: Architecture should support all 3 possibilities. Mappings could exist wherever
community ontology is, or at each endpoint or at mapping service.  For
scalability, having it at the endpoints is most viable. If mappings get
broken, then service should maintain mappings as quickly as possible, as they have most self interest.

TF: if you send a query to a service, need mechanism to specify preferences
for results message ontology?  So service could appear to speak several ontologies. 

MB: Big services like SAP may demand clients maintain the ontology mappings on their own. Then clients 
have the self interest to stay connected. 

TF: Weather.com - lets you set prefs for presentation (centigrade vs farenheit)

TF: some ontologies/mappings not related to services. 
MB: Yes, need translations even for annotated text to do indexing retrieval. But SWS is a big
motivation for mappings. 

MD: May want to do concept-concept mappings rather than index by ontologies
which are just the containers.

MB: But all you are likely to know when looking for a mapping is what the
concept to be translated is and what the target ontology is.

TF: car size example: compact, midsize, full size 
MB: or the Starbucks sizes: tall, grande, venti
(Each example points out that one needs to translate the terms as well as the classes. )

MB: each agent will also use multiple ontologies, so one needs to find which ontology the
target agent has that can be used to translate each term.

MH: on topic of loss, most mappings applied in pairs, for 2 services to
interact, there will be mapping to common ontology and to other
service. Good 3rd party translator will want to use an ontology that is superset of ontologies
of others that it translates to/from.

MB: Same problem if translating directly, may need to extend target ontology to enable 
full translation. 

MB: Mike H - was CYCL good language for writing mapping rules?
MH: Not especially, hard thing was breaking up fields. 
MH: ex: field called address, vs city, state, zip.  substring processing not handled in CYCL or any logic repesentation. Still don't know how to 


TF: You may want to separate the logical parts from the procedural (string manip) parts. 

MB: and you need to do the mappings in each direction separately (string breakup and recombine) XSLT often used for this.

MB: Action for MB to write up outline of space of models we want to encompass, so we can push down on 
protocols and mapping issues for each. 

NOTE: Stuart W. to attend August F2F.