Re: draft response for LC comment 26 (a and b)

On 16 Feb 2009, at 09:26, Ivan Herman wrote:
[snip]
>  That said, I
> think part of the issue is that there is no clear understanding  
> when QL
> could be or should be used as opposed to, say, RL (or EL or the DL
> altogether for that matter). Neither the profile document nor any  
> other

Check!
	http://www.w3.org/2007/OWL/wiki/Primer#OWL_2_EL
	http://www.w3.org/2007/OWL/wiki/Primer#OWL_2_QL
	http://www.w3.org/2007/OWL/wiki/Primer#OWL_2_RL

> gives any help for that and my understanding is that this is Lilly's
> main concern...

Yeah, but let's not go overboard. I think it's easy to add too *much*  
advice. If we believe there will be implementors and users for the  
various profiles (and I'm convinced), and that they are sensibly and  
wisely designed (which I'm pretty on board with), *and* we think they  
are differentiatable (which I also agree with), we should let the  
market take care of guiding people. After all, the largest  
determinant, I'd bet, is what tools you have and where the ontology  
you receive falls.

> Profile checkers are very good things to refer to. But those are ad
> posteriori tools,

Not necessarily. They can be incorporated into an editor.

> what we might need to have is some clear, a priori
> guidelines for users (not implementers).

I very much disagree. We can't possibly do an adequate job for  
something that is, in general, very situation specific. We run a  
heavy risk of closing off useful perceptions (viz the complaints that  
the use cases/examples are all biohealth...that's silly; none of  
these are *domain specific* and it's strange to conflate  
applicability for one task as lack of any rationale for any other task).

Also, the above is a false comparison. You check the profile of  
something to know what it is and what tools you can use it with. You  
*commit* to a profile for tactical or strategic reasons. E.g., I  
imagine SNOMED-CT will stick with OWL EL because 1) it's expressive  
enough, and 2) there are a growing number of reasoners which are  
robustly scalable with respect to it. But it's actually pretty hard  
to see how we can translate that into useful general advice: "Pick a  
profile that works for your domain and for which the available  
tooling is sufficient unto your needs."

Cheers,
Bijan.

Received on Monday, 16 February 2009 11:16:21 UTC