Re: [regrep-semantic] Reg/Rep + Reasoning Use Cases

[Farrukh Najmi <Farrukh.Najmi@Sun.COM>]

Jeff Pollock wrote:

> Farrukh and SCM team-
>  
> I've attached a candidate use case document with three use cases:
>
>     * 11 - Reason on Reg/Rep Content Objects (via RIM Query Interfaces)
>     * 12 - Reason on Reg/Rep Schema Directly (Tight Binding with OWL)
>     * 13 - Reason on Reg/Rep Schema Directly (Loose Binding with OWL)
>
> These tend to be architectural use cases, contemplating different 
> mechanisms for accomplishing common business goals. All three 
> "include" a common use case "Infer Semantics on Content Objects." 
> Number 11 does no reasoning on the RIM directly.  Numbers 12 & 13 
> support reasoning on the RIM directly via OWL interfaces, but consider 
> different binding mechanisms.
>  
> I've attempted be as clear as possible, but the writing and thinking 
> raised more issues, some of which I only alluded to.  I look forward 
> to everyone's thoughts.
>  

Jeff,

I found the document very well thought out and a very good articulation 
of the alternative design approaches for supporting reasoning in the 
context of our SCM work.
A few small suggestions follow...

1. I think the document is a compilation of Design Alternatives rather 
than use cases. Call it what we may it is an important document for our 
work.
My suggestion is that we call it "Design Alternatives and Recommendation 
for Inference Support" (or something along those lines).

2. We still need some "Business" oriented use cases for inference 
support which would be the motivation for the above "Design 
Alternatives" document.
The use cases may be along the lines of the T-Box and A-Box Inference 
that you mentioned in today's meeting. For those that missed the informative
discussion led by Jeff T-Box inferencing is "Taxonomic" inferencing 
based upon inter-class relationships while A-Box inferencing is based 
upon "Associative"
inferencing based upon instance (aka individual) properties.

Would you be willing to contribute a few use cases involving T-Box and 
A-Box inferencing?

3. The current document could use a paragraph at the end that compares 
the pros and cons of each approach and makes a recommendation (I assume
it will be option 3 (loose coupling).

4. In option 3 (loose coupling) I like the idea of the Inference engine 
to be normatively speced as a pluggable web service along the same lines
as the design for Content Validation or Content Cataloging service. This 
would allow a registry to be configured with multiple inference engines,
one for each type of Ontology content type (e.g. OWL, KIF etc.).

5. After describing option 3 as the chosen option mention that a 
detailed specification for option 3 will be developed as a separate 
deliverable.
The specification would include a WSDL binding for the inference service 
and other details.
Lastly, I think we should include in our planned deliverables a concrete 
binding between OWL and RIM and leave other Ontology languages
to be defined elsewhere. I think even in the loosely coupled approach we 
need to make sure that OWL support is explicitly defined.

Thanks again for making this valuable contribution to the SC. Team, 
please comment on the document. Thanks.

-- 
Regards,
Farrukh