RE: [SKOS] ISSUE-26: "Minimal Label Relation" Proposal

Dear Antoine and Alistar,

Thank you for this work which we think is very useful! Capturing
relationships between labels is what we are try to represent since some time.

Regarding the possible modelling approaches you mention in [4] here my
comments:

- I like First solution: Term-as-class. This will allow to create also any
other type of properties between terms. It also allow to attache terms to the
corresponding concept.

- Second solution: I personally do not like this too much. I think this will
duplicate the info (labels are written twice)

- Third solution: keeping standard SKOS and Term-as-class solutions
co-existing... not yet clear

Let me know if this become valid, then I will recreate agrovoc in skos using
this approach.

Recently I also discovered that in SKOS I can represent 3 different BTs: the
real BT (is-a), the instance-of and the part-of. This is also very good. We
are now trying to represent this differences in AGROVOC so that we can than
export data in this new version of SKOS.

Thanks
Margherita Sini

[4] http://www.w3.org/2006/07/SWD/wiki/SkosDesign/RelationshipsBetweenLabels



-----Original Message-----
From: public-esw-thes-request@w3.org [mailto:public-esw-thes-request@w3.org]
On Behalf Of Antoine Isaac
Sent: 15 June 2007 16:11
To: Miles, AJ (Alistair)
Cc: SWD WG; public-esw-thes@w3.org
Subject: Re: [SKOS] ISSUE-26: "Minimal Label Relation" Proposal



Hi Alistair,

>  
>
>   
>> By the way this 'counting' does not take into account the
>> weight of the seeLabelRelation. To me, you could remove it, 
>> this would have the benefit of not anchoring your link to one 
>> concept exclusively. This anchoring can have some advantages, 
>> but it also seems very restrictive. 
>> If you were to have 2 labels linked that belong to different 
>> concepts (e.g. if your label relationship is 'antonym') would 
>> you attach the labelRelation instance to both concepts?
>>     
>
> The proposal has deliberately weak semantics for 
> skos:seeLabelRelation. There is no requirement to use it all. There 
> are no constraints its cardinality in either direction. Also (from 
> [4]):
>
> "...there does not necessarily have to be any correspondance between 
> the lexical labels of a resource, and the labels involved in a label 
> relation, to which the resource is related via the 
> skos:seeLabelRelation property"
>
>   
I had indeed understood this. But it was not very explicit from the 
examples alone, and I wanted to have confirmation of what would happen 
in the antonymy case.

>>
>>> Hi all,
>>>
>>> Here is an alternative proposal for resolution of ISSUE-26
>>>       
>>>
>>> I've called this proposal "Minimal Label Relation" because
>>>       
>> it is very similar to Guus' "LabelRelation" proposal, but
>> with a bit less ontological commitment.
>>     
>>>   
>>>       
>> Why?
>>     
>
> Note I am referring to Guus' "LabelRelation" proposal [1], not Guus' 
> "Simple Extension" proposal [3].
>> Antoher question, purely formal: is this new proposal of
>> yours deprecating the more general proposal you launched in [2]?
>>     
>
> I prefer the "Minimal Label Relation" proposal [4] to the 
> "LabelAnnotation" proposal [2].
>
> So many proposal names :)

Argh. So if I would try to clarify the information found at the top of 
[0] (yes, zero ;-) then your "minmal label relation" [4] competes with 
your previous "Label Annotation" [2] and Guus' "Label Relation" [1]. And 
then the best of the three should be compared to the "Simple extension" 
by Guus [4] which proposes a less systematic approach to the 
Term-as-class option

Notice that all the proposals relying on sort of reification of the 
label link face the important problem raised by Jon in [5]: asserting 
links between labels should somehow be contextualized, and I'm not sure 
you could do with so little semantics for your seeLabelRelation 
property. Indeed, since attaching a LabelRelation to a concept is not 
optimal as I've tried to show, I would propose to contextualize it by 
attaching it to the ConceptScheme instance itself...

Notice also that this contextualization issue may also occur for the 
term-as-class part of Guus' "Simple Extension" [3]. Perhaps a better way 
is just to ignore this problem for now, and to postpone it till we solve 
the related SemanticRelationshipContainment issue [6]. I really don't 
like how all these issues are related...

Cheers,

Antoine

[0] 
http://www.w3.org/2006/07/SWD/wiki/SkosDesign/RelationshipsBetweenLabels/
[1] http://lists.w3.org/Archives/Public/public-swd-wg/2007Feb/0195.html
[2] http://lists.w3.org/Archives/Public/public-swd-wg/2007Mar/0092.html
[3] 
http://www.w3.org/2006/07/SWD/wiki/SkosDesign/RelationshipsBetweenLabels/Prop
osalThree
[4] 
http://www.w3.org/2006/07/SWD/wiki/SkosDesign/RelationshipsBetweenLabels/Prop
osalFour
[5] http://lists.w3.org/Archives/Public/public-swd-wg/2007Feb/0187.html
[6] http://www.w3.org/2006/07/SWD/track/issues/36

Received on Wednesday, 20 June 2007 15:09:21 UTC