Fragments - specific proposal

In light of the conversations and presentation at the f2f as best I  
could follow - I propose the following which I believe is actually  
within the spirit of consensus

Summary (details below)  I propose we redefine OWL Lite to be  
somewhat less than is currently in the language, that we name a new  
subset with the intent of adding just enough OWL to RDFS to match  
current "low end" needs, and I propose that we create an easily  
findable Working Note and Wiki page with named subsets of OWL that  
are provably polynomial based on Bernardo's current document.



1 - OWL Lite
I propose we redefine OWL Lite to be closer to the current language  
coverage that is in Oracle's OWL Prime.
  For the DL version of this, we will have a model theory and the  
restrictions needed to prove whatever it is we wish to prove about it  
(completeness, decidability, etc)
  There will be a Full version which is defined as the subset of the  
OWL Full Semantics (however we end up defining those) corresponding  
to the same language constructs (note: this may well require removing  
some language features that Oracle and others currently support from  
this named subset)

2 - RDFS 3.0
I propose we name a subset called RDFS 3.0 which is less than OWL  
Lite -  aimed primarily at universals - i.e. named classes and  
properties, no restriction statements involved.
  There should be a version of this which is provably polynomial  
within certain restrictions (at least no redefinition of the language  
features, possibly those DL restrictions required to keep this true -  
ie only inversefunctional datatypes used as keys per Uli's  
presentation, or whatever).  This version could be called something  
like RDFS 3.0 DB (if we get it close to datalog) or DL (if that's the  
best we can do)
  The language would likely be widely used (or more correctly, is  
already widely used) - I also note that something very similar to the  
fragment I proposed has been used by a couple of companies using map/ 
reduce algorithms, meaning RDFS 3.0 would be effective for use in  
large scale web applications since it appears that it can be deployed  
on large server farms (note that these results are based on hearsay,  
as most of the companies exploring RDF parallelization require NDAs  
to discuss their work).
  RDFS 3.0 DB would also entice DB researchers to explore issues such  
as query optimization in SPARQL and the like.

3 - other fragments
The many other polynomial fragments that Bernardo has documented are  
valuable.  I would suggest we create a WG note based on the current  
fragments documents, and a Wiki page to make it easier for users to  
link to implementations and etc.

I propose the first two go rec track.  The documentation of these  
fragments would clearly follow the practices for the remainder of the  
language - i.e. model theory, functional spec or whatever could be  
developed (since these would mainly be subsetting of larger sets,  
this should be doable).  Compliance with both of these would be  
specified in some way that the group agrees to (consistent with  
whatever practice we develop for the remainder of the language).

I believe this would meet the following needs:
  1 - OWL Lite already exists, but is not widely used in its current  
form (most things in OWL Lite appear to be less expressive than the  
whole of the language).  This would allow us to create a more easily  
implemented OWL Lite for DL tools, and for companies like HP, Oracle  
and the others mentioned at the f2f, they could claim to support OWL  
Lite (Full) and comply to it fairly easily from where they are now.
  2 - RDFS 3.0 would entice the Web 3.0 market segment to look at  
including some OWL - and my hope is in for a penny, in for a pound,  
they'd eventual move to greater use - and if not, there'd still be  
wide value.  The ease of moving from the full to the DB or DL would  
also help bring users not yet thinking about reasoning per se (i.e.  
doing simple inferences via procedural code) into exploring what  
tools like Pellet and Racer can do.





"If we knew what we were doing, it wouldn't be called research, would  
it?." - Albert Einstein

Prof James Hendler				http://www.cs.rpi.edu/~hendler
Tetherless World Constellation Chair
Computer Science Dept
Rensselaer Polytechnic Institute, Troy NY 12180