Re: SKOS Reference Editor's Draft 23 December 2007

On 9 Jan 2008, at 11:32, Simon Spero wrote:

> [Greetings from Sunny Chigwell]

Greetings from Sunny (yes, really :-) Manchester!

> I'm still really, really uncomfortable with the breakage of   
> skos:broader, primarily in regards to the loss of transitivity.
>
> BT relationships in all compliant thesauri must be transitive.
>
>   Each of the relationships should lead to hierarchies that are  
> amenable to a logical test through reference to the basic types of  
> concept represented by the terms. (NISO 2005, §8.3)
>
>
> Of course, for polyhierarchic structures,   BTG* should not follow   
> BTP, but that's a different issue :)
>
> If skos:broader is to be (a) intransitive and (b) useful, there  
> needs to be a very explicit specification of what the relationship  
> does mean; this will probably require a complete non-monotonic  
> semantics.


I had a look through the minutes of the F2F [1] to try and remember  
why we had made this decision. First of all, let's be clear about  
what we're currently saying in the document. Apologies if this strays  
into granny/egg territory, but I think it's useful to check that  
we're all talking about the same thing, particularly when we're using  
terms like intransitive. The draft says:

[[
Note that skos:broader is *not* a transitive property.
]]

So, given

a skos:broader b
b skos:broader c                                   (1)

we cannot, in general, infer that

a skos:broader c                                   (2)

That is not to say that this is never the case. I would use  
intransitivity for the stronger case. That is largely a matter of  
terminology though, but I just wanted to make that clear.

Looking through the meeting record [1], it seems that one of the  
objections to the transitivity was from an IR perspective. Ralph  
summed this up as:

"The primary use case for broader/narrower is in information  
retrieval. For such cases, the application of OWL reasoning on  
transitive relations would weaken search results in many (most?)  
cases, therefore the SKOS specification does not demand that broader/ 
narrower are transitive."

I guess this wording didn't make it into the draft.

My concerns with the transitivity are mainly wrapped up with the  
issue of whether we were defining a knowledge representation language  
or a data model for KOSs. If broader is transitive, there are  
implicit consequences about the relationships between concepts that  
can be drawn from a collection of statements. But then again, we have  
stated that broader/narrower are inverses, which also then admits  
inference.

> As an  example of a valid hierarchical relationship and test  (NISO  
> 2005, Figure 6 ) uses:
>
> cacti   ∝ succulent plants
> SOME succulent plants are cacti
> ALL cacti are succulent plants
>
> The example given for an  invalid hierarchical relationship is  
> (NISO 2005, Figure 7 )
>
> cacti ∝ desert plants
> SOME desert plants are cacti
> SOME cacti are desert plants
>
> A  relationship that does not obey these properties is not  
> hierarchical and should not be labeled as such.  The relationship  
> is associative, and can be modeled in SKOS using the appropriate  
> construct (skos:related).    The error is in the data, not the  
> standard.


These examples are clear, however, they are essentially examples of  
subclass -- one can interpret "succulent plants" as describing the  
class of things that are succulent plants. Similarly for "cacti" or  
"desert plants". Thus your reasoning (SOME x are y) is based on  
instances of those classes. This is relatively straightforward if the  
concepts are describing concrete concepts, and such an interpretation  
fits naturally with a transitive broader. But what if the terms are  
more "conceptual"? For example, if I look at the term "Leisure" in  
the Unesco thesaurus

http://databases.unesco.org/thesaurus/wwwi32.exe/%5Bin=affiche.in%5D/

There is a NT Sociology of Leisure. What's the corresponding  
statement here in terms of ALL/SOME? Does this pass the test you  
describe?

I should make it clear that I don't have a particular hard line on  
this and wouldn't be deeply unhappy to see broader as transitive.  
Also, I'm not too clued up on thesaurus standards -- I'm coming at  
this much more from the formal KR end of things, so am very happy to  
be told what's what by those in the know in terms of application  
need :-).

Cheers,

	Sean

[1] http://www.w3.org/2007/10/09-swd-minutes.html

--
Sean Bechhofer
School of Computer Science
University of Manchester
sean.bechhofer@manchester.ac.uk
http://www.cs.manchester.ac.uk/people/bechhofer

Received on Wednesday, 9 January 2008 17:06:10 UTC