Uncoupling Identity and Resolution

My previous post outlined my understanding of Norman Walsh's postion on URIs and Identifiers.  I indicated in that post that I believed there are several cases where it is desireable to uncouple identity and resolution, and why my personal view is that http:URIs are not ideal for such cases.

I do not dispute that the requirements embedded in these cases can be addressed using http:URIs. I simply assert that they can be better served within a more complete naming architecture that explicitly accomodates pure identifiers... that is, identifiers that are explicitly uncoupled from resolution protocols. 

The current state of Web architecture does not support pure identifiers, and in fact I believe it is fair to characterize the prevailing attitude towards them among Web and Internet architects as hostile.  Certainly discussions of the merits and demerits in various public fora have often been (over?) heated at times.

Resolving Pure Identifiers

Does this mean that pure identifiers are never subject to resolution? No. It means that a pure identifier must be explicitly bound to a resolution protocol by a process or within a system designed to exploit that identifier.  This is, of course, a far more constrained circumstance than one mostly encounters on the open Web.  What are the means for such resolution?

1. Binding of a pure identifier and a resolution protocol can be done by convention, as is the practice with DOIs, for example. Such conventions require special knowledge that must be established and maintained within a community of use.  The commercial benefits of the use of DOIs to help publishers manage intellectual property has helped to overcome this barrier, though the success is modest and fairly closely contained within the publishing community.

2. It can be done through the use of plugins for standard software such as browsers.  During the URN battles within the IETF, which coincided roughly with the emergence of 'standardized' browsers, much hope was pinned on the use of such plugins to help get URN usage off the ground.  These hopes live yet in some parts of the library community (a number of European libraries have coalesced around the registration of National Bibliography Numbers, or NBNs, as a registered URN namespace), but it is a hope that (in my judgment) is unlikely to be widely realized.

Even if there were no organizational impediments to the use of plugins, getting users to install such plugins in a widespread way is unlikely.  In fact, the installation of plugins in large organizations is often strictly controlled for security reasons, and hence unlikely to be done widely on the open Web.

3. The architecture of the Internet includes a registry for URI schemes, maintained by the Internet Assigned Numbers Authority (IANA), which in theory could be deployed so as to ease the recognition of registered URI schemes within Web.  For a variety of reasons related to lack of compelling business cases, security issues, ideology, and perhaps the convenience of systems developers, very little of this capability has ever seen the light of working code.  The procedure for registering a new URI scheme was revamped as recently as last year (2005), suggesting that some in a position of influence still consider this important, but at this stage of Web technology development, the prospects of widely deployed software that can use arbitrary URI scheme declarations sensibly seems remote.

4. Finally, labelled identifiers are often easily recognized and parsed from open Web data.  ISBNs are a good example of this, as they are widely recognized by end-users, have a public syntax, and are typically labeled with the "ISBN" token, making them easy to identify in unstructured data.  Oh... and there is a business case for their use.  Easy pickings.  Never mind that the identifier system itself has flaws... it works well enough, often enough, to generate value.   Given that there are many legacy identifier systems in the non-web world, and there is great confusion about how best to 'webulate' such systems, it is comforting that, as long as well-formed and labelled identifiers are used, they will be findable.

This final approach may constitute the most reliable path to recognition of a given class of pure identifier.  It does not require agreement on the part of Internet architecture gatekeepers, and it is market driven.  So if there is in fact a market need, there is some chance that it can be filled by entrepreneurial zeal or even community-mindedness.

-----

Image: Lady Washington, in Tacoma Harbor, with Rainier-san as a backdrop (August 27)

Comments

[Jonathan Rochkind]

Hi Stu. Finding this post when looking for a clearly stated argument as to why it's neccesary to support 'pure identifiers' (that are not canonically resolvable)... I don't actually find one here, despite the promise of the introduction! Most of your post is talking about _how_ 'pure identifiers' might be used and implemented.

But many in those 'overheated debates' you mention would argue that there ought to BE no 'pure identifiers', all identifiers ought to be canonically resolvable. Can you point me to a good argument against this, a position you allude to, that "I do argue, however, that there are circumstances in which resolution should be explicitly uncoupled from identity". What are these cases?