Re: CR Exit Criteria

On 27 May 2009, at 17:53, Michael Schneider wrote:
[snip]
> Just to be clear: I don't claim that the criterion is too hard to  
> meet *per se*, which would really be an, erm, "exit criterion" for  
> OWL 2 Full. I don't see anything obviously "un-implementable" in  
> OWL 2 Full,

Really? Can I ask what implementation experience that grounds this  
judgment? I'm not being snarky, but I think you are underestimating  
the implementation challenges. Even if you were building off a FOL  
reasoner, it doesn't seem trivial to me. There's also no literature  
to drawn on. Inventing reasoning algorithms is much harder, in my  
experience, than implementing one from a book.

I'm not saying it's impossible, just that the uncertainy is pretty  
high. (And depends on what you mean by "obviously un-implementable".)

[snip]
>  But if one is about creating a serious implementation of OWL 2  
> Full, then this will almost certainly take much too long for this  
> working group to build one. The same would probably be true for an  
> OWL 2 /DL/ reasoner, if one were about implementing it from scratch  
> at the beginning of the CR phase.

This last bit seems pretty wrong to me. Pellet is sort of a proof of  
concept that we could build such a thing (done, part time, for WebOnt  
CR). And the literature and general experience in building DL  
reasoners was much less then. And Evren and I were doing it from books.

Now, I'm sure you could put together a test suite that would be, er,  
challenging to beat in this time frame. And I'm not saying it would  
be a super duper production system. But I think it's not so very  
likely, given a test suite that focused on language coverage (but  
also was "reasonable"). Heck, Boris builds reasoners to relax over  
bank holiday weekends :)

For building an OWL 2 Full reasoner...well...it seems like research.  
It is closer to what building a reasonable SHOIN reasoner was back at  
webont (Ian and Uli had to do *research* before we could do it).

(None of this should be construed as opposition to fairly easy exit  
criteria for OWL Full. But lets be clear on what is and is not likely  
possible here.)

Cheers,
Bijan.

Received on Wednesday, 27 May 2009 17:20:01 UTC