Re: Schism in the Semantic Web community.

On 27 Jan 2009, at 10:16, Danny Ayers wrote:

> Hi Bijan, quick question - what shape would the practical synergy  
> take between OWL2 and linked data?

Several different ones, but consider one:

	OWL-QL (the profile) is such that its conjunctive query answering  
can be done straightforwardly and efficiently against a relational  
database without changing the underlying database's schema or  
implementation. Thus, OWL-QL can support a non-ETL based way of  
exposing arbitrary databases on the web. Furthermore, it is known how  
to take several OWL-QL ontologies for different databases, map them  
to a common schema, then (automagically) distribute the query to the  
several database even if those databases overlap in kind (i.e., both  
have "address" data, but in different sorts of table). Thus, one  
could publish an OWL QL "sparql api" and have multiple sources lift  
their databases to that api without disrupting their applications or  
having to synch their database with a store.

(A piece missing is SPARQL/OWL, but I am working on it and hope that  
the new group will give that work a home.)

> Speaking personally, I find it very reassuring the way the group  
> has taken into account Aboxy RDF, but still have trouble seeing how  
> everything ties together as an extension of the existing (document)  
> Web, i.e. maximally exploiting HTTP, surely a major motivation  
> behind semweb efforts.

Well, it's hard to say for formats (instead of protocols). After all,  
RDF says almost nothing about using HTTP. It would help if you  
explained what you would hope for.

OWL imports is arguably better than before.

OWL/XML is easier to compose, transform, style in normal XML ways.  
Obviously, you could use RDFa or variants to include OWL in HTML pages.

The profiles are "robustly" scalable. It is significantly easier with  
any of the profiles to produce an implementation that scales with  
arbitrary (within the profile) input (both TBox and ABox). While  
OWL2/1 as a whole are far more scalable than people often think (see  
HermiT and SHER for examples), it is definitely a much greater  
challenge (indeed, theoretically impossible) to make them scale  
robustly on arbitrary input. (Of course, this is technically and  
sometimes practically true for RDBMSs, but their range and  
personpower is much greater.)

I have no idea if this is answering your question since I don't  
understand your question, really :) It's a bit vague, eh?

Cheers,
Bijan.

Received on Tuesday, 27 January 2009 10:37:13 UTC