OntologDiscussion/Protege And KIF

Technical Discussion on TBD Schedule    (9MO)

• Topic proposed by: NicolasRouquette / 2005.02.03    (9MP)
• Discussion topic: Protege-KIF: ontologies & reasoning    (9MQ)

Background    (9MR)

PatrickCassidy has made a substantial effort to port the SUMO formal ontology to Protege. SUMO's formalization is written in KIF. However, KIF is more expressive than PAL, the constraint language of Protege. Switching to OWL, a port of SUMO to Protege-OWL will still run into issues of limited expressiveness that prevent a translation of some KIF formulas into their OWL equivalents.    (9MS)

Pat commented that if we are willing to perform "string" encoding of the part of SUMO that is not expressible or representable in PAL (or OWL), then we can have a "port" of SUMO that does not loose any information, i.e. it is "isagelous".    (9MT)

Problems & Questions    (9MU)

There is great utility in having full reasoning support for ontology development. e.g., see the "OWL and Description Logics" session of the 2004 Protege conference: http://protege.stanford.edu/conference/2004/schedule2.html#thursday.    (9MV)

However, the "string" encoding approach that Pat described will clearly result in the loss of the class of logical inferences that depend on the logical formulas that have been hidden from the reasoner due to their "string" encoding. This echoes one of the conduding remarks that Christine Golbreich and Atsutoshi Imai made in their presentation at that conference about combining OWL and SWRL reasoning. More generally, it reinforces the conventional wisdom of matching to tool one uses (a language for representing logic formulas in this case) to the complexity of the task at hand (an application-specific task that requires logic reasoning).    (9MW)

For the specific case of SUMO (which is in KIF), it's clear that there is a value of a mixed-language approach:    (9MX)

• some parts of SUMO formalized in OWL-DL to provide a practical & useful degree of reasoning capability,    (9MY)

e.g., in the spirit of Rector's presentation on the role of classifiers at the 2004 Protege conference.    (9MZ)

• some parts of SUMO formalized in KIF or SCL to provide more reasoning flexibility, e.g., in application    (9N0)

contexts where there task-specific knowledge can provide useful guidance for selecting appropriate reasoning tools and for guiding the reasoning process.    (9N1)

In this approach, both formalizations can be viewed as distinct annotations that formalize the ontology according to different viewpoints. In this case, the viewpoints are defined by the logic languages, e.g., OWL-DL and KID/SCL. As long as the formalizations are logically similar (i.e., the logic formulas as identical up-to the renaming of variables) this should not be a problem. This might be a problem if the formalizations are logically different (ie., some logic formulas are different even if we allow variable renaming).    (9N2)

•    (9N3)
  • Questions for Pat about Protege-KIF and SUMO in particular:    (9N4)

•    (9N5)
  •    (9N6)
    • Have you noticed any patterns in terms of the SUMO KIF formulas that had to be string-encoded due to limited expressiveness?    (9N7)

•    (9N8)
  •    (9N9)
    • Is there any significance to the set of string-encoded formulas from SUMO KIF?    (9NA)

•    (9NB)
  •    (9NC)
    • If you were to do it all over again, i.e., from SUMO-KIF to Protege-OWL (e.g., in the OWL-DL subset), what kind of additional meta-data annotation would we need to include in order to decide what is the relevance and usefulness of the SUMO-KIF formulas that have been string-encoded because they are not expressible in OWL-DL ?    (9ND)