[NLP2RDF] NIF: mandatory properties and types

[Sebastian Hellmann]

Hi Martin,
you raise some interesting questions.


On 06/14/2012 06:49 AM, Martín Rezk wrote:
> Dear all,
>              I am new in the project, so I have been asking Sebastian
> several question about the NIF standard, in particular, about  the
> mandatory properties in NIF and their domain and range.
> He asked me to switch to the mailing-list, so here I am, nice to meet you all.
> In the last email, Sebastian, you told me that the only mandatory
> property (including the SSO?) is referenceContext. Question: Can an
> object have as referenceContext itself?
> In page 6, in your EKAW paper, you have a 7 lines example. In such
> example the first triple defines the whole document (offset_0_26546)
> and it is not defined for the  (functional?) object property
> referenceContext.
> Instead, it is defined for the data property str:occursIn. In such
> scenario, what does NIF require?  should offset_0_2654 have itself as
> referenceContext? Should referenceContext not be mandatory in that
> case?
Indeed, the current version is underspecified regarding this respect. We 
could say, that each instance of str:Context has a reflexive connection 
via referenceContext to itself. This would make sense probably.
Actually, this would solve the problem between the distinction between 
Context and StringInContext. As we could model the Context as a 
subClassOf String, i.e. the subclass, that has a reflexive reference to 
itself ..
Do you know how to express this in OWL? Maybe 
http://www.w3.org/TR/2009/WD-owl2-new-features-20090611/#F6:__Reflexive.2C_Irreflexive.2C_and_Asymmetric_Object_Properties

This will make detecting any cycles harder probably.

>
> Regarding the answer from the Pellet forum,  you told me that you will
> not add axioms to prevent the wrong usage of OWL.
> However, I think it would be interesting to give an optional simple
> set of axioms that can be used to check the correctness of the
> ontology.
> Some of them could be
>
> 1) referenceContext is functional and mandatory.
There is already an owl:hasKey axiom. This  provide something similar to 
functionality and is more light-weight :
HasKey StringInContext ( referenceContext ) ( beginIndex , endIndex  ))
> 2) There can not be cycles in the property referenceContext. That is,
> it should disallow things like: A has as reference context B, and B
> has as reference context A.
The best way to really solve this is to make some examples and then give 
different T-Boxes  and see the consequences. I am almost sure that there 
will not be *one* string ontology, but rather different sets of axioms 
for certain use cases. If you have a large text collection, you just 
might want to have RDFS inference over the hierarchy. Consistency might 
be (1) expensive to calculate (2) annoying. When I implemented it the 
first time, I had one tool, that made a tiny encoding mistake, but  
there were too many functional properties involved. It was really 
annoying, because sometimes you just might want to continue and ignore 
it. Subclassof inference would have worked out fine regardless of this 
functional/encoding mistake.

> 3)  If the class str:Nothing is meant to be bottom (from the logic
> point of view), then the class str:Nothing should be empty.
There is no class called str:Nothing. The OWL Doc documentation at 
http://nlp2rdf.lod2.eu/schema/doc/string/index.htmlis misleading. We 
will try to replace it soon. The class you see is owl:Nothing and it 
should not be in there.
Thanks for your feedback,
Sebastian

>
> Best wishes,
>
> Martin Rezk.-
> _______________________________________________
> NLP2RDF mailing list
> NLP2RDF at lists.informatik.uni-leipzig.de
> http://lists.informatik.uni-leipzig.de/mailman/listinfo/nlp2rdf
>


-- 
Dipl. Inf. Sebastian Hellmann
Department of Computer Science, University of Leipzig
Projects: http://nlp2rdf.org , http://dbpedia.org
Homepage: http://bis.informatik.uni-leipzig.de/SebastianHellmann
Research Group: http://aksw.org