[OWL] inverse datatype properties

[Ian Horrocks]

On 27 Feb 2006, at 21:46, Ian Horrocks wrote:

> On 27 Feb 2006, at 18:44, Evren Sirin wrote:
>
>> Matthew Pocock wrote:
>>
>>> Hi,
>>>
>>> When representing actual data in OWL (rather than  
>>> individual-graphs), I keep
>>> needing datatype properties to be more expressive. Our application is
>>> data-mining and data-visualisation, where much of the actual  
>>> information is
>>> encoded in concrete data types. In particular, it would be helpful  
>>> to have:
>>>
>>>  inverseFunctional: the value of the datatype uniquely identifies the
>>> individual
>>>
>>>
>> I agree this is very helpful. There are already algorithms proposed to
>> handle inverse functional datatype properties, a.k.a. keys, in DLs.  
>> For
>> example, see [1] that describes a tableau algorithm for SHOQK(D).
>
> Just to explicate what is implicit in Evren's reply: it is easy to  
> show (see [1]) that, in the general case, allowing inverse functional  
> datatype properties would lead to undecidability (of OWL).

Oops - I just realised that the above statement is not correct. Of  
course the undecidability results in [1] depend on the use of concrete  
domains, which are more expressive than the very simple datatypes  
provided by OWL. In fact OWL's datatypes can be simulated using  
nominals, so allowing inverse functional datatype properties would be  
harmless from a theoretical perspective. The argument for not  
supporting inverse functional datatype properties in OWL was a  
pragmatic one based on the difficulty of providing "practical"  
reasoning services (see  
http://www.w3.org/2001/sw/WebOnt/webont-issues.html#I5.1-Uniform- 
treatment-of-literal-data-values).

Sorry for the confusion.

Ian


>
> Ian
>
>
>> Actually [1] considers much more expressive ways of describing keys  
>> and
>> IFDP is just a special case. Howeer, even for this special case, there
>> are some problems that need to be solved in order to have an efficient
>> implementation. I have implemented a preliminary version of IFDP  
>> support
>> in Pellet but this is still work in progress.
>>
>>>  inverseDatatypeProperty: a property from datatype to individual,  
>>> allowing us
>>> to say things like: give me all the values of the hasAge property
>>>
>>>
>> I'm not sure if inverse datatype properties are needed for this  
>> purpose.
>> I would argue it is enough just to have a feature in the query  
>> language.
>> If you want to get the values of hasAge property you can write a query
>> in your (or your reasoner's) favorite query language (SPARQL, DIG or
>> nRQL). Each of these languages have features to ask for the values of
>> datatype properties. For example, in SPARQL you would simply write
>> {?person hasAge ?age}.
>>
>>>  datatypeProperty( ... hasInverse inverseDatatypePropertyID)
>>>  inverseDatatypeProperty( ... hasInverse datypePropertyID )
>>>
>>>
>>> Assuming that you could only invert datatypeProperties that refer to  
>>> concrete
>>> datatypes with an equivalence relation,
>>>
>> I'm probably missing the use case here. Is there something that the
>> query language would not let you to do?
>>
>>> would this make the reasoners choke?
>>>
>>>
>>>
>> I can talk about my experience about implementing IFDP support. For  
>> most
>> of the cases, e.g. FOAF data that heavily uses IFDP, reasoning is not
>> affected much because the ontology itself  is not very expressive. But
>> it is easy to come up with use cases that will use IFDP in complex  
>> ways
>> and that can kill the performance. But I believe it is possible (and  
>> we
>> are working on) to devise more sophisticated optimizations for such  
>> cases
>>
>> Regards,
>> Evren
>>
>> [1] Carsten Lutz, Carlos Areces, Ian Horrocks, and Ulrike Sattler.  
>> Keys,
>> nominals, and concrete domains. /J. of Artificial Intelligence
>> Research/, 23:667-726, 2004.
>> (http://www.cs.man.ac.uk/~horrocks/Publications/#LAHS04a)
>>
>>> Matthew
>>> _______________________________________________
>>> OWL mailing list
>>> OWL at lists.mindswap.org
>>> http://lists.mindswap.org/mailman/listinfo/owl
>>>
>>>
>>>
>>>
>>
>>
>> -- 
>> Evren Sirin evren at cs.umd.edu
>> Graduate Research Assistant
>> Computer Science Department
>> Univ of Maryland, College Park, MD 20742
>> Phone: (301) 405-7027, Fax: (301) 405-6707
>>
>>
>> _______________________________________________
>> OWL mailing list
>> OWL at lists.mindswap.org
>> http://lists.mindswap.org/mailman/listinfo/owl
>