Re: Language

At the telecon, I was asked to find out about `expectations' re the
language attribute in RDF particularly at the model level. 

I've been talking to our sales and services guys about user
expectations: they meet lots of users of metadata (Finnish and French
mainly with some other nationalities). Specifically, I've been asking
about the demands of users with respect to the identification of literal
values with locale and language.

The gist (surprise to me) is that there is no expectation of language
attribution to survive in bodies of metadata.

The test I've been using is to what extent is an object literal
considered to be `an element of meaning with multiple textual
embodiments'. This seems a significant question where translation is not
a one-to-one affair and values carry a cultural content.

It's not the way I think of it but metadata seems still to be seen as a
locale-locked resource - either `normalized' (as ever) into "en" and
mediated by localized software or commercially bound within a national
boundary.

One aspect I've been pursuing is the equivalence of metadata triples if
the object fields differ only by locale. If two triples are the same
except for objects l("chat","fr") and l("cat","en"), would these be
expected to be handled as equal in reality by any reasonable product
implementation? The general tendency is to classify such problems as
`not proven'.

Most of our customers and prospectives would willingly model such
equivalence classes as a separate body of metadata where a resource
represents the semantic point and translations are associated as
properties. This attitude derives from the fact that most literals which
can be taken to represent units of meaning (rather than character
strings) belong to limited sets, typically category names. These are
usually reduced to locale-free codes and mediated by interfaces.

We were wondering about the status and role of language in the RDF
model: is it special? This feedback demotes the association of locale
with text down to the level of general property (in the body of the
metadata) rather than as an intrinsic attribute of the quoted literal.
Language is seen by most as a dictionary activity not an issue of
representation. Literals are really character sequences and are terminal
in that sense (in most users' minds).

xml:lang as a special technique of representation is not expected to
feature in RDF directly except in bodies of metadata specifically
addressing localization and even then language would feature mostly
explicitly as predicate codes or property values.

This is just general input, with no specific recommendation attached but
it does come from a group of people actually promoting RDF in the real
world.

-- 
Martyn Horner <martyn.horner@profium.com>
Profium (former name Pro Solutions), Les Espaces de Sophia,
Immeuble Delta, B.P. 037, F-06901 Sophia-Antipolis, France
Tel. +33 (0)4.93.95.31.44 Fax. +33 (0)4.93.95.52.58
Mob. +33 (0)6.21.01.54.56
Internet: http://www.profium.com

Received on Wednesday, 4 July 2001 08:19:00 UTC