Comment on URNs, namespaces, and registries "finding": Some unfair characterizations

This comment is with regards to the TAG finding "URNs, Namespaces and
Registries" (Editor's draft) in circulation.  It appears to have
become the precedent for arguments against the use of additional URI
schemes - even where network resolution is completely [1] orthogonal.

The conversation leading up [2] to the raising of the "issue" left out
a strong motivation for coining non-HTTP URIs for RDF content:
non-dereferencable names

HTTP URI dereference (as a mechanism) and (formal) denotation are
completely orthogonal.  Mixing them up without the right amount of
"care" has dire consequences.  HTTP dereference is not without expense
and persistent suggestion that it *should* always be performed during
RDF processing seem to fall along certain assumptions (ala "Fallacies
of Distributed Computing").

Functional computations (such as URI dereference) are typically left
out of many predicate logic dialect syntax due to consequences of
computational tractability and (perhaps?) decidability.  it seems
dangerous to forcibly suggest that URI schemes which are orthogonal to
any network dereference (by design) should not be used as identifiers
for model-theoretic representations (such as RDF and OWL).

The fact that there is no advantage in using a particular scheme where
network resolution is irrelevant should not be misconstrued as a
(circular) argument for *only* using a particular scheme.

Section 4.5 ("Erroneous appearance of dereferencability of
identifiers") of the finding seems to seriously underestimate the
network effect of the assumption [3] that HTTP schemes should be both
used ( in RDF ) and dereferenced.

Ironically, this section also paraphrases a well-founded reason for
using a scheme which has no suggestion of "dereferencability": neither
the human nor a web agent makes a false assumption that useful
information can and should be "dereferenced".

Why agitate a well-known fallacy of distributed computing for reasons
completely orthogonal to any network benefit of HTTP resolution?

With regards to the AWWW best practice regarding re-use of schemes:

"A specification SHOULD reuse an existing URI scheme (rather than
create a new one) when it provides the desired properties of
identifiers and their relation to resources."

It certainly cannot be claimed that HTTP *always* "provides the
desired properties of identifiers" when (for example) it has no
syntactic support (or semantics) for certain identity management
capabilities that motivated the LSID and UUID specifications
(versioning, non-colliding identification, etc.)

Indeed, UUIDs (and other such URI schemes) can be used in HTTP URIs
but this is at best a naming convention and not licensed
(syntactically and semantically) by any of the schemes in isolation.

If the "answers" given are "Rarely if ever" and "Probably not" (i.e.,
not authoritative) then at the very least the scenarios where the HTTP
scheme does not offer any obvious advantage should be better
represented so as not to perpetuate misconceptions regarding that
(grey) area where RDF overlaps with HTTP.  At best, some clear
indication (for the purposes of education[4]) of good practice within
mechanisms which identify network locations of relevant OWL/RDFS
assertions would very instructive: rdfs:seeAlso, owl:imports,
rdfs:isDefinedBy etc..

[1]http://lists.w3.org/Archives/Public/public-semweb-lifesci/2007Aug/0050.html
[2]http://lists.w3.org/Archives/Public/www-tag/2005Mar/att-0056/March152005#item07
[3]http://www.w3.org/DesignIssues/LinkedData.html
[4]http://esw.w3.org/topic/HCLS/WebClosureSocialConvention

Received on Friday, 17 August 2007 04:08:05 UTC