Re: OWL DB, OWL UML, etc.

Jim,

At 05:44 AM 5/14/2004, Jim Hendler wrote:
>ok, I guess this could mean a lot of things -- we're looking at the fact 
>that some recent papers in the DB community have been discussing a notion 
>of ontology as a set of relations and showing how adding these relations 
>to the data schema allows improved recall (possibly at the cost of 
>precision) for queries against structured DBs.  Simple example is what we 
>might call inheritance -- if you say in a typical database that clyde is 
>of type "elephant" and query for "animal" you won't get Clyde -- with ISA 
>links added in an ontology, then you can show this is doable and (with 
>certain extensions) stay within polynomial time.  I know that my group 
>proved this in Parka almost a decade ago, but database folks don't read AI 
>papers, and the new work is actually stronger in that it is more 
>integrated in with the DB calculi.   I've been thinking that their results 
>could be extended further by using more of OWL's expressivity rather than 
>their purely mathematical definitions of what ontological relations 
>are.  Thus, OWL DB would be parts of OWL, added to database scheme, which 
>extend the recall of database systems by use of OWL modeling constructs 
>(and inheriting all the other benefits of OWL - standardization, URI 
>based, etc.)   >From the DB perspective you wouldn't lose anything - keys 
>and etc. would be defined as they currently are - but you would gain the 
>ability to use some new expressivity within the rules of the game as 
>played by DB, rather than AI, people (i.e. all algorithms polynomial, 
>reducible to relational calculus, etc.)  - I guess this means I would be 
>thinking of adding some OWL constructs to what is already there, thus 
>yielding a new langauge that is more expressive than the current 
>relational calculus, without sacrificing the expressiveness thereof.
>   -JH

It would be terrific if an OWL DB language as you are describing it would 
come into being.
However, I would imagine (and hope) that database constructs like the 
closed world assumption
would be preserved, and if they make the mapping from tuples to entities, 
then they
may want something roughly equivalent to the unique name assumption 
(something like, "if
two entities have keys which differ, then the entities are not the 
same").  My impression,
strengthened by discussion with Pat Hayes, is that this language would not 
be compatible with
OWLs Lite, DL, and Full (but I'll let Pat or someone else have the final 
say on this).  That means
that it would not fit neatly into the OWL pantheon.

I of course am all for that.  I think its a shame that OWL has forced 
things down a path that
is incompatible with deductive database logics, and would like to see an 
alternative growth path.

Cheers, Bob

Received on Saturday, 15 May 2004 10:47:35 UTC