Charter and vision of SC (Was Re: [regrep-semantic] Blurb about SCMWork (For Interview))

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

I have taken the liberty to fork/rename this thread to reflect the new 
directions in the conversation.

Jeff, you raise some thought provoking issues. Here are some initails 
thoughts inline below....

Jeff Pollock wrote:

>Farrukh, et al.
>
>This has been a fascinating thread to follow. Over the course of the
>past couple weeks I have been consciously disengaging to this group
>because of what I perceived as a soft commitment to OWL and, more
>importantly, a soft commitment to actual 'semantic capabilities' -
>beyond simply storing ontology and defining mappings to RIM concepts.
>  
>
We are in our formative stages. We started with the direction you prefer 
(that was my
bias) to have a tight binding to OWL and built in inference 
capabilities. In fact
I was thinking that we should recast RIM to be defined as an OWL Ontology.

As we got into the use cases phase there were some in the SC that did 
not want
a tight binding to OWL. We therefor began discussing a possible 
alternative where
we define a minimal canonical Ontology represention in RIM, allow multiple
Ontology syntaxes to bind to it and allow an external inference engine to be
plugged into the registry.

No decision has been made about the OWL or generic approach *YET*.

>However, if this group's statements (albeit, some have been removed)
>during this thread about a commitment towards (1) data harmonization,
>(2) semantic web server of the future, (3) precise content discovery,
>and (4) semantic web services are indeed the intention for the SCM group
>- then I have misunderstood the direction thus far.
>  
>
Our vision and goals have not changed sinced we started. But the 
discussion has
meandered (sometime all over the place). Carl and I have tried to keep 
the discussion
focused using use cases and requirements to light the path.

>It seems to me that the charter is loose in regard to the visionary
>statements made on this thread and that internal fractures to SCM have
>misdirected the effort away from the vision I am hearing folks describe
>today.
>  
>
The charter is definitely quite flexible and that is by design. Please 
do not make any judgements
yet about the decisions we make in the SC. It is much too early.

>Recently, with this group's statements, and with the compelling work
>presented by professor Dogac on Tuesday, I am inclined to offer a few
>thoughts about Network Inference's position with respect to this RegRep
>SCM work.
>
>First, I want to state the obvious ... Network Inference is a technology
>vendor that is highly committed to the widespread use and adoption of
>all OWL, and in particular OWL-DL.  With regard to our participation
>here, we do not wish to be perceived as too vendor-centric with our
>thoughts, instead we want to be supportive and share ideas about our
>vision and implementation of OWL-based semantic tools.
>  
>
And please do stay actively involved in the SC and influence
as you are with this email. There are no pre-set agendas. All we have is 
a high level vision, some proposed use cases
an,d even fewer, some proposed requirements.

If you would like to discuss your thoughts on wher eyou think the SC 
should be heading
we can assign that to our next meeting's agenda.

>Second, our position is that we strongly favor a tight binding between
>OWL and the RegRep RIM interfaces.  This position is clearly influenced
>by my first point, but most importantly, we favor this tight binding
>because we feel that it will enable a new generation of tools to
>directly interact with a regrep implementation to provide valuable
>semantic capabilities (inference, mediation, dynamic workflow, etc.)
>without requiring kludgy/lossy adaptors to move between ontological
>representations.  This pluggability of semantic tools is crucial, in our
>opinion, to making the regrep relevant for next-gen Semantic Web
>Services capabilities.
>
>Third, our expertise and early customer implementations with OWL should
>be valuable, to this SCM group, but in a 'tightly-bound OWL/regrep'
>scenario.  We have built Xquery and WSDL/SOAP interfaces to an OWL/RDF
>inference engine that allows users to merge, inference across, and query
>instances or classifications within inferred and told OWL/RDF models.
>Without getting into implementation specifics (Abox,Tbox,performance
>metrics etc.) about the inference engine - it should be apparent that
>our interfaces (Xquery and OWL manipulation) could have some relevance
>to this SCM effort. As we drill into certain use cases this could become
>useful, but only when/if the binding is tighter.
>  
>
I am curious to do a what if scenario... What if Network Inference could 
shape SCM SC
just the way you want... What would that look like. What use cases would 
that be
trying to address. Maybe this could be part of your presentation to the SC.

>Finally, I want to share with you some technical use cases for how we
>are seeing people use ontologies and web services together.  With due
>respect to professor Dogac's assertion that there are two primary roles
>for ontologies in web services (service orchestration & ontological
>reasoning), we disagree and believe that the possible roles for
>ontologies is limited only by the number of architectural layers in a
>web services architecture. 
>
I think Asuman was just picking a couple of immediate applications of 
ontology and web services.
I did not take it as being exhaustive. I agree with you that there are 
many more applications limited
only by our imagination.

>A further distinction is that we believe that
>the _only significant_ advantage for using ontologies is that you can
>reason or reason across them - thus, all of our use cases leverage the
>value of the precision that OWL & RDF provide in a data representation
>language.  With that said, our emerging technical use cases include:
>  
>
Precisely! That one advantage is the common theme to all other higher 
level perceived advantages.
That is why reasoning engine capability is so essential to the vision.

>Service Orchestration - this involves the use of a reasoning engine to
>enable dynamic discovery, composition and monitoring of services without
>apriori codified knowledge of those services. We view OWL-S -- as it has
>been most recently embodied in OWL-DL -- as the crucial enabler of these
>capabilities.
>  
>
I do not think we should have any first class treatment for OWL-S. I 
think our
approach should be generic and allow semantic metadata to be attached to 
arbitrary content
(not just Services). If we can use any (OWL) ontology(s) to semantically 
describe content
then we are in good shape.

>Business Inferencing - this involves the use of a reasoning engine to
>load multiple DL-based ontology, merge them into a common inferred model
>and then query that inferred model for answers about the domain(s) that
>accurately reflect the rules that have been asserted in the model (via
>DL axioms).
>  
>
>Query Mediation - this involves the use of a reasoning engine, plus
>other EAI type adapters, to mediate queries across federated sources. By
>issuing queries against conceptual models that logically overlay
>different physical systems (or objects inside the RIM for that matter)
>vendors can build tools that reachout in native query languages and get
>subsets of data from a centrally issued query.
>
>Transformation Mediation - this involves the use of a reasoning engine
>that can read ontology-based maps, in OWL-DL for example, that unify
>schema/instances in a point-to-point or hub-and-spoke manner and then
>resolve the physical data's incongruities as an output of the
>newly-aligned schema/instance semantics found in the reasoner.
>  
>
Above use cases seem quite powerful but will require some time and 
discussion for me to grok it.
I look forward to learning more in next meeting or email.

>The "semantic" aspect of these use cases is two-fold (1) the metadata
>precision allowed by OWL/RDF to treat relationships as first-class
>objects and (2) that their context is late-bound to their execution
>state.  In other words, since these capabilities (mediation, inferencing
>and orchestration) are loaded "just-in-time" from models, they are
>dynamic and unique at any given time.  Add in the feature that models
>can change based upon the output of these capabilities, the feedback
>loop is closed and you can begin to see meaningful emergent and adaptive
>behavior.
>  
>
+1
Very well put.

>To wrap up this winded email, I just want to ask for clarification
>around the RegRep SCM scope and charter:
>
>Are we in fact moving towards a position that supports the idea that
>(vendor) tools *should* use a RegRep implementation _directly_ to
>support semantic capabilities --- or are we going to take an
>intermediary (perhaps wiser?) step to simply ensure RIM support for
>ontology storage in multiple formats? (thus allowing tools to store
>ontology in RegRep, but not reason directly to it)
>  
>
We have not made any decision yet on this point. We are leaning towards 
keeping
our options open for multiple ontology syntax in deference to input from 
some
team memebrs but I have concerns about the extra complexity and 
constraints this
adds.

Incidently the main proponent of supporting multiple syntax (Zach)
seems to have dropped of the list and has missed past several meetings.

>This will help me to determine the level of involvement from our team.
>
>Thoughts? Opinions?
>  
>
In my opinion you should actively engage and artculate your position on 
why a tight binding
to OWL and full inference support in registry is the right approach. If 
the team decides to go
a different direction after a reasonable debate then it would be the 
right time for you to tune out.

>Many kind regards and no offense intended to anyone,
>
>  
>
None taken. In fact I am very glad for you having taken the time to 
write down this well articuolated and
thought provoking post.

-- 
Regards,
Farrukh