Re: Comment: Defining N-ary relations, use of cardinality

> A small remark: for the diagnosis relation example I would expect to 
> have a cardinality restriction on the diagnosis_value and the 
> diagnosis_probability set to one (there must be a diagnosis, after 
> all...). I do not think that the the fact of defining this as a 
> functional property covers that (functional means that there must be 
> at most one value, right?) You use the cardinality restriction in the 
> 'buyer' relation later. Regardless on how one interprets this very 
> example, I think the usage of cardinality is very important in n-ary 
> relations in general, so it might be more 'didactic' to use it imho.

Yes, you are right. We tried to be reasonably careful about this, but 
apparently not completely. Indeed, making something a functional 
property sets the max cardinality to 1, but doesn't require a value. 
Setting cardinality for diagnosis_value to 1 would make a lot of sense. 
I am not sure  though if diagnosis_probability should require a value. 
What if we don't know the probability?

> Also, in this example, I would expect the diagnosis probability 
> relation to be a datatype property rather than an object property, 
> with an xml schema datatype of an interval between 0 and 1. After all, 
> this is what probability is... (at least as an alternative example to 
> the 'literal' type object property you seem to use right now)

I think in this example probability really is a qualitative measure, 
rather than a quantitative  one. I think we talk about someone with 
high probability of breast cancer, rather than about someone with 
probability of breast cancer at 0.85 Besides, OWL doesn't have a good 
way of defining property ranges, and we'll leave that to a different TF 
note to discuss. It's probably not very relevant for this particular 
note (We struggled quite a bit to isolate all the related thorny 
issues)

> A slightly different question: I am not sure what a DL reasoner would 
> do if, say, Diagnosis_relation was also defined to be of a type 
> rdfs:Statement. It may not add anything to any inference results, but 
> might have a 'descriptive' value nevertheless. But I am a bit out of 
> my league to predict the behaviour of a DL reasoner on this...

It's out of my area of expertise as well, but I don't think making it 
rdf:statement would change much.

Natasha

Received on Tuesday, 10 August 2004 19:53:08 UTC