Re: translations for new properties

Charles McCathieNevile wrote:

>
> On Sat, 08 Oct 2005 10:57:17 +0200, Bernard Vatant  
> <bernard.vatant@mondeca.com> wrote:
>
>> Your "descargabile de la red" is more explicit than the english  
>> original, which seems to
>> have been problematic also for the german translator. The same in 
>> french  would be indeed
>> "téléchargeable sur le Web" rather than "addressable".
>>
>> But since the range of skos:symbol is  dcmitype:image should not we  
>> stick to the definition given by DCMI?
>
>
> I think not, for two reasons:
>
> Alistair admitted to being "lazy" in writing the property, but the 
> fact  that we have a lot of clues about what he meant to say, in my 
> mind,  justifies us doing a good translation of what he meant, in 
> order to  provide more guidance for people relying on the non-english 
> version.
>
> The second is that the presence of real text, not just a note about 
> the  range, is that there might be something more than the minimum 
> that can be  inferred from the range. For example we might read the 
> note and thus not  be surprised when the property is "upgraded" to use 
> a range of something  more tightly defined than dct:image

Yep, quite right.

At the meta-level, there's a piece of work to be done here (perhaps by a 
future incarnation of SWBP WG) describing classes of change to RDF 
vocabularies. It's related to the VM work (which I wish I'd had more 
time to spend on...).

case 1: making a range more restrictive
eg 2005 you say "the rdfs:range of eg:myproperty is xyz:SomeClass" and 
then in in 2006, you change the schema to assert that the rdfs:range is 
pqr:SomeOtherClass", ... and the latter is a subclass of the former. So 
before you were saying "whenever you see a true usage of  eg:myproperty, 
you can be sure it is pointing at an instance of an xyz:SomeClass". Now 
you're saying "whenever you see it, you can be sure you're pointing at 
an instance of a pqr:SomeOtherClass".

As you point out (and the RDFS spec tries to say too[1], folk shouldn't 
simplistically assume that the rdfs:range of a property means that any 
instance of the range class is an applicable value. They need to know 
more about the meaning of the property. What the RDFS semantics 
guarantee is just that, if you see the property with some given resource 
as its value, you know the type of that resource. Not that all other 
things of that type could have been used.

We need some notion of "fairness" here. Depending on what exactly the 
text definition, examples, mainstream practice etc etc amount to, the 
community will react in various ways. Eg. "hey, unfair, your examples in 
various specs led me to believe the property was applicable to these 
things which did fall within the formal range of the property" versus 
"ok, that's reasonable; the new range more accurately captures the 
commonly understood meaning of the term". In the Dublin Core scene, some 
of us are wondering how a restriction of the range of dc:creator etc 
would be perceived. My guess is that it won't be perceived well... but 
that's another thread.

case 2: making a range less restrictive
This is a bit different. It basically removes the justification for data 
that was inferred on the basis of range/domains. That new data might 
still happen to be true, but claims in the schema are no longer 
available as 'evidence'.

There are a lot more variations than I've listed here, too...

cheers,

Dan

[1] http://www.w3.org/TR/rdf-schema/#ch_introduction

Received on Sunday, 9 October 2005 21:18:46 UTC