Re: Please help on description-logic-xxx test cases.

Hi Minsu

I am copying Sean on this message, he was the author of these particular tests 
(as you can see from the dc:creator in the Manifest files),  

The tests themselves come from DL'98.

The following link gives the introduction:
http://sunsite.informatik.rwth-aachen.de/Publications/CEUR-WS/Vol-11/Intro.ps

Page 2 of that document is probably the best description of these tests.

The original test data is still available from Horrocks:
http://www.cs.man.ac.uk/~horrocks/FaCT/dl98-test.tar.gz


(I have just noticed that the link in the CR document was broken, I believe 
the above links are OK). 
See [DL 98] in the references in the OWL Test Cases

> I have added to my OWL inference rulebase a bunch of inference
> rules for owl:intersectionOf and owl:complementOf, and it made
> my Bossam engine successfully pass five description-logic-2xx tests,
> which are 201,202,204,205 and 207. :-)
> But I got two failures on 203 and 206. :-(

> What are the purposes of these tests? The descriptions on
> the tests just say something cryptic like k_branch, k_d4,
> k_dum, k_grz, k_lin, k_path, and k_ph. I cannot see any
> differences between the tests by reading premise documents.
> They just look very similar to each other.

From my point of view your message reveals the purpose of these tests - to 
break your system !! (And other peoples). A failing test is an opportunity to 
improve your code.

We chose to include tests from previous work by the Description Logic 
community. We hoped to gain from their experience of some things that are 
difficult to implement.
The tests in the test suite are intended to have a range of difficulty, so 
that even the best systems struggle to pass all of them. We have tried to 
avoid really impossible tests (except perhaps in the extra credit section).

In the acknowledgements section there is the list of test authors. You will 
see that it is fairly long, and because of that the test themselves show a 
variety of flavours - those authored by myself and Sean tend to have rather 
cryptic abstract concept names - and we do not appear to be thinking about a 
real world problem. I personally tend to think about OWL in a fairly abstract 
way, and my tests are merely symbolic manipulation. Those from Dan Connolly 
or Jos De Roo tend in general to refer to real world problems, and hence tend 
to be easier to understand.

I hope this message helps - 
I take it that your comment was not a request to change the document in any 
way, merely one implementor talking to another ...

If you actually want it to be taken as a formal comment, maybe a request for 
additional clarifying text to be included in the document, please reply and I 
will take such a request to the working group. (Personally I would not be too 
happy - because it looks like a lot of work to do that for every test).

Good luck, I hope you get the rules right soon.

Jeremy

Received on Friday, 28 November 2003 18:14:08 UTC