Re: Usage of literals in rdf-primer figure 1 and 5

Hi Chris--

Thanks for the comments.  I don't disagree with the idea of encouraging 
people to think about using URIrefs in place of literals.  However, I 
think the Primer goes as far as it needs to in this regard.

In the first place, I think most RDF novices need to be *explicitly 
introduced* to the idea of using URIs for what are conventionally 
represented by literals (like "Dr."), rather than this use suddenly 
appearing (let alone appearing "exclusively" throughout the Primer). 
Hence, the Primer mainly (except for the "fast forward" in the intro) 
starts out using URIs (actually URLs) to identify something that most 
Web users would already expect them to be used for:  a Web page; and 
using literals for other things.  The Primer then explicitly points out 
the extended use of URIs to identify other things, like people and 
predicates, and then discusses the potential value of using URIs instead 
of literals to identify things.

In the second place, using a mixture of literals and URIrefs to describe 
things is the real world, and will be for some time, if not forever 
(URIrefs are just names for things;  people continue to use combinations 
of names and descriptions employing literals to describe things). 
Requiring the exclusive use of URIrefs for literals in the Primer would, 
for example, preclude presenting any of the example applications in 
Section 6 (none of which use URIrefs exclusively for these sorts of things).

Simply using URIrefs (e.g., for a conventional title like "Dr.") doesn't 
magically improve interoperability;  it has to be a *common* URIref; 
one whose meaning is generally shared.  For example, whose URIref for 
"Dr." are you suggesting should be the one that appears in the Primer? 
In this case, until better community consensus arises (and maybe even 
after), I'd bet that a literal like "Dr." will serve fairly well to 
indicate this title, given its fairly-universal use right now (certainly 
better than a non-standard URIref).  The same holds for city and state 
names, zip codes, and so on (some of which have values that are 
explicitly intended as unique identifiers).  Everyone making up their 
own URIrefs for such things as titles won't help very much.

I also think some of the best practice in this area needs to evolve, 
rather than the Primer talking as if it already existed (let alone 
appearing to dictate it).  Take a Zip code, for example.  Is it best 
modeled as:

* a URI like usps:01730 (with "usps" representing a prefix officially 
defined by the US Postal Service) as the value of a predicate like 
ex:postalCode?

* a typed literal like "01730"^^usps:zipcode as the value of the same 
predicate (in this case, the USPS is defining a datatype for Zipcodes)?

* the same typed literal, but as the value of a USPS-defined predicate 
usps:postalCode (or a corresponding internationally-defined datatype and 
predicate)?

* the original plain literal, but as the value of the USPS-defined 
predicate?

Note that the title "Dr." in the first Primer example is the value of a 
predicate from a specified vocabulary (the SWAP contact vocabulary), so 
intended meaning of the literal value is not totally undefined, and, as 
the examples above suggest, I think the identification of the predicate 
has a place in this discussion.

There's a proposal for a Semantic Web "best practices" working group 
floating around (see message 
http://lists.w3.org/Archives/Public/www-rdf-interest/2003Nov/0055.html) 
and I expect that group, should it actually start work, will contribute 
to the understanding of some of these issues.  At this point, I think 
the Primer is doing its job if it makes novices aware that there's an 
issue here, without trying to suggest that all the decisions about best 
practice have already been made.

--Frank


Chris Bizer wrote:

> Hi,
> 
> a small but maybe didactically important comment on the usage of  literals
> in figure 1 and 5 of rdf- primer (also cited in rdf-concepts).
> 
> It is stated that Eric has the personal title "Dr." using a literal.
> Following common RDF modelling principles, an URIref should be used to
> identify the concept of "Dr." as an academic title.
> 
> I think I might be confusing for a RDF novice to be taught in section 2.1
> and 2.2 to use URIs to identify the creator of a webpage and then to have
> examples in the same document, that do exactly the opposite.
> 
> The same problem appears to the example in figure 5 where the properties of
> an address (City, State, ...) are described using literals, which is deadly
> for any graph merging.
> 
> I think it is important to use URIrefs instead of literals in the examples.
> These examples will be copied into hundreds of introductions to RDF and will
> otherwise confuse beginners.
> 
> You could also think about changing the sentence "In fact, a URI can be
> created to refer to anything that needs to be referred to in a statement,
> including ..." in section 2.1. into "In fact, a URI *should be used* to
> refer to anything that needs to be referred to in a statement, including
> ..." in order to make good modelling practices clearer.
> 
> 
> 
> Chris Bizer
> 
> 
> 
> 

Received on Tuesday, 20 January 2004 09:37:19 UTC