RE: [WN, SPARQL, ALL] Describe

Hi Jacco,

> McBride, Brian wrote:
> 
> >So a position for discussion:
> >
> >1. The DESCRIBE functionality would be useful for Wordnet.
> >
> >2. It would be useful stand alone, not as part of SPARQL.
> >
> >3. It would also be useful as part of SPARQL, but less so.  
> An example 
> >might be a query that queried multiple Wordnets for 
> different languages 
> >for "chat" and returned dscriptions of each of the word senses.
> >  
> >
> I had missed this mail from Brian, and the last time I had 
> read the SPARQL draft it did not mention DESCRIBE, so I could 
> not answer Guus's question on this during the previous 
> teleconf.  Having catched up, I agree with Brian's point at 
> 1) and 2) above.

:)

>  My position concerning 3) would be stronger 
> than his:  I think any useful implementation of DESCRIBE is 
> so dependent on the vocabulary being queried, that it should 
> NOT be part of the core of a general query language such as 
> SPARQL (all the other functionality of SPARQL can be 
> implemented independent of the data being queried, this would 
> be the only exception) .

Chatting with Andy Seaborne today (that is an acknowledgement rather
than name dropping :) he pointed out that there are some vocabularies
where DESCRIBE might be useful, but where the resource to be described
is represented is not named but is represented by a b-node.  He cited
FOAF as an example.  In this case it is useful to be able to DESCRIBE a
variable in a query.

But I think your point here is that queriers might want some guarantees
about what would be returned as the result of a DESCRIBE query and that
such guarantees are dependent on the vocabulary.  I think this is a
valid point, though it is not one that I feel qualified to make much in
the way of definitive statements of requirements.

I suppose I am wondering whether we might deal with this but allowing
SPARQL to support the general vocabulary in the query language and then
individual vocabularies could suggest/require what should be returned as
the result of a DESCRIBE query.  This would then make the behaviour of
the query server depend on particular vocabularies.  One might envisage
a standard for describing the the results that should be returned as the
result of a describe query which might defined in terms of an equivalent
SPARQL query, e.g. for resource of type X, return the graph generated by
query Q.  I suspect the waters are murky when we have mixed
vocabularies, but might not be too bad.

An alternative line might be that there is no guarantee about what might
be returned and application that does not find the information it needs
returned as the result of a describe query must be prepared to issue
further queries to get the information it needs.

> 
> Having said this, we might want to encourage the DAWG to 
> continue work on DESCRIBE, as it would be useful for Wordnet, 
> SKOS and every other large ontology of which one might want a 
> concise description of a single resource.  I think they will 
> still have an interoperability issue.  I do not buy the 
> argument that the type of functionality offered by DESCRIBE 
> doesn't need to be interoperable.  If a machine needs to be 
> able to do anything useful with the result, there will need 
> to be a minimum set of assumptions the machine can make about 
> the result.  Also, one might expect that different Wordnet 
> servers at least return the same _type_ of information for 
> the same query. So one might want to have an RDF vocabulary 
> in which you can specify the result of DESCRIBE for a given 
> type of resource.

There may be a class of application where just showing the user what the
server chose to return is good enough or what is in fact the desired
behaviour.  

There may be a class of application where the application knows enough
about the vocabulary to ask for stuff that is missing, though in that
case it could have stated explicitly what it wanted in the first place.
In that case using DESCRIBE might just be syntactic shorthand for the
lazy programmer, or a way of saying "tell me what you know" which might
be more than the application was expecting, though again that could be
explicitly coded in the query.

There might be some value for DESCRIBE in handling multiple versions of
vocabularies, where the query to specified depends on the version of the
vocabulary being used.

It could be that the Wordnet and SKOS documents can give guidance on
what a server should return when resources of a particular type are
DESCRIBED, and as I suggested above, there might be a mechanism for
vocabilaries to describe what should be returned.

I'm not yet clear in my own mind what I think about this issue.  At the
time writing it seems to me that the value of DESCRIBE is when
applications don't know what to expect and the server gets to choose,
and also as a short hand for what otherwise be a tedious query to
specify, but I'd like to hear Jacco's and other's thoughts before
continuing.

Brian

Received on Wednesday, 14 December 2005 16:33:29 UTC