Re: On intentions of Naming Authorities and Referers

On Apr 2, 2008, at 10:34 AM, Williams, Stuart (HP Labs, Bristol) wrote:

> Some musings that I failed so stop myself writing down... they seem  
> to me worth sharing... but YMMV.
>
> I have some nagging thoughts around what David would call URI  
> Declarations and the intentions of two parties:
>
> - parties that act as naming authorities and bind names to things;
>   specifically URI names; even more specifically HTTP URI names.
>   Typically by arranging that an origin server response when  
> questioned
>   with a given name (http URI). Also, by creating 'anchor' points in
>   documents which publish.
>
> - parties that make references to things using names;
>   specifically URI names; even more specifically HTTP URI names.
>   Typically by making references using URIs in documents, the  
> publication
>   of which involves an act of naming - though these parties are  
> generally
>   not providers of all the names that they use to make references.
>
> The general question that seems to be on the table is how an  
> observer of a reference (made using a URI) is to determine the  
> intended referent ie. the referent intended by the publisher of the  
> reference.

Well... this would be sufficient, but not necessary. You may be able  
to make statements about something, even if you don't know what it  
is.  The implication from 200 that the referent is an "information  
resource" might be something of that sort.

> Though critised by Hayes and Halpin[1], the "follow-your-nose"  
> philosophy of the web is that the intended referent of a URI name,  
> wherever it appears, is whatever the naming authority for that name  
> intended it to be a name of.

I don't think this has anything to do with follow your nose. Even if  
I don't give my domain name an IP address, I still get to say what I  
think my URIs are supposed to mean (denote).

> In effect the implict contract in using the web is that when a  
> party creates a reference using a URI name to refer to something,  
> then what the name in the reference refers to is whatever the  
> naming authority for the name intends that it refer to  
> (acknowledging the chorus of "...that's fine, but how would anyone  
> know *what* that is?"). Thus, the question for the observer of a  
> name becomes not "what did the creator of the reference intend it  
> to refer to?" but "what did the naming authority for the name  
> intend that it refer to?"

Well... ultimately I would think you really do know what the author  
was trying to say, even if they have made a mistake. The point of  
naming is that if sender and receiver both understand a name the same  
way, then they don't have to communicate its meaning (definition,  
whatever), they only have to use the name. A naming system is  
successfully deployed to the extent that they don't misunderstand  
uses of the name. An "authority" that has communicated with both  
sender and receiver may help to nail the connection.

> the intention of web architecture being that the answer is the same  
> in both cases.

The answer *should* be the same if no one makes a mistake.  
Architecture can help prevent mistakes, but it can't make anyone use  
a URI in the right way.

> The means to obtain an answer (if any) needs to significantly out- 
> live the relevant authority - whose personal capacity to answer  
> such repeated enquiries will diminish to zero overtime :-)
>
> The GOFHTW (good old-fashioned hypertext-web) has thrived largely  
> without naming authorities giving explicit expression to their  
> intentions. Often (leaf-delegated) naming authorities do not even  
> realise that they were acting in such a role ("I just put this  
> document on the web") or that they have or had any obligation to  
> make explicit statements about what they have published - indeed in  
> general there has been no such obligation on the GOFHTW.
>
> The GOFHTW has evolved and been 'successful' without requiring such  
> expression... why is that?

Excellent question. I would say:
   - because it is used by people, not by automated agents, and  
people are forgiving
   - broken links are usually quickly repaired because web sites  
(unlike libraries) are 'live'
   - most assertions have "href" as the verb, which is so sloppy that  
it's difficult to be wrong
   - because of its low expected semantic service level (librarians  
don't use URIs)

> It seems to me to have relied on the intuition of human consumers  
> of references who upon 'following' a reference (a hyper-link) are  
> presented with a rendering of a representation of (the state of?)  
> of the referenced thing (or are redirected to something related to  
> it). In the main it has not been neccessary to share definitive  
> assertions about what in fact is being referenced - only an  
> understanding that whatever the intended referent is, it is the  
> same for each use of the same name, and it is the same for all  
> observers[*] of references. Statements such as "http:/ 
> weather.example.com/Oaxaca" is a good place to look for information  
> about the weather in Oaxaca can be made. Maybe it would be  
> reasonable to inuit that the referenced thing is a source of Oaxaca  
> weather reports - though that is left unstated by the naming  
> authority. On the GOFHTW human consumers of references content  
> themselves with their inutitions arising from the 'decoration' that  
> surrounds a reference (link) and what they are presented with if  
> they follow it.  Many, though by no means all, documents on the web  
> are self-referential and may provide a human and potentially a  
> machine readable account of themselves, possibly amongst other things.
>
> The 'things' (ie. the referents of names) of the GOFHTW are also  
> the 'things' of the NFSW (new-fangled semantic web).
> That is, the NFSW is not merely layered on top of the GOFHTW, but  
> intertwined with it. The intended referent of a reference made  
> using a URI in a semantic web document is the same as the intended  
> referent of a reference made using the same URI in a hypertext  
> document (or a pdf or... ) - or at least (FWIW) I assert that that  
> remains the implicit contract of using the web.

I wish this were true - that following a link always got me to a  
document reflecting what the "URI owner" means by the URI. But the  
infrastructure - both servers and clients - is not so reliable. You  
have to assume that this is the case.

> Which then leads us back to the same dangling question... "what  
> does (or did) the naming authority for a given name intend that the  
> name refer to?".
>
> On the GOFHTW self-reference and human inuition where adequate for  
> sufficiently resolving that question in most cases. In cases where  
> this is insuffient, frankly, there is no reliable technical  
> mechanism. One can try to ask the naming authority (maybe email a  
> question to the relevant webmaster... but over time I'd expect a  
> lack the knowledge and an induced loss of patience to prevail :-)).
>
> The naming authority could try to publish some definitive  
> information about the names under their authority that might help.  
> Jonathan surveys some suggested approaches for obtaining such  
> descriptions at [2]); make some narrative assertions (for humans)  
> or some RDF assertions (for machines - and some humans :-). However  
> we are in a free floating world of symbols grounded by symbols  
> with, in some cases, symbols grounded in social documents  
> (specifications) which capture some level of common social  
> agreement (sometimes weak agreement!) about the intended use of  
> particular names.
>
> I think that the best that we are likely to be able to do from  
> descriptions is:
>
> - detecting when some different names are being used to reference  
> the same thing (without being definitive about what that same thing  
> 'is');
> - detecting when what is being said about a thing is inconsistent -  
> ie. that there is no-thing for which the set of assertions being  
> considered could possibly all be true.
> - infering some other things that can be deduced from what has been  
> said (classifications, closures...)
>
> Of themselves these are useful things to be able to do... but they  
> are way short of a machine being able to determine say, that the  
> referent of a given URI is the 'actual moon' 'that orbits' 'the  
> earth' 'inhabited by' 'us' [all of those quoted strings themselves  
> being symbols whose referents need to be understood in order to  
> understand the description of the original referent]. Pat Hayes' by  
> now infamous PatHayesAbout document[3] (oh Pat I see that you've  
> rearranged the way [4] responds to accesses) is appealing in that  
> it establishes a number of invariants of the person which it  
> describes. However I almost failed to notice that it is replete  
> with a number of other names for which no similar attempt is made -  
> thus CYC I guess, by induction - and there Pat would have had me -  
> though much later onced I'd had warmed to that style of description  
> and sprung the trap :-)
>
> Jonathan/Alan seems to speak in terms of there being a strong  
> obligation on naming authorities to give an account of what the  
> names they deploy are used to refer to. They may go further to  
> speak of such accounts as definitive. I believe that is at best  
> hard and probably impossible. That's not to say that descriptions  
> are useless.

The obligation is conditional: You only need to do it if you want  
people to be able to speak clearly about your things (or rather using  
your URIs), and have people understand you for more than a few  
months. There are plenty of successful naming systems, such as that  
for minor planets, and some of them are embedded in URI space, so you  
can't say that the task is impossible. But setting one up is really  
difficult, and some semantic web literature borders on being  
disingenuous on this subject.

If you're only interested in ephemeral communication the job gets  
easier (as far as 200s are concerned), and if you assume nothing  
changes faster than the rate at which what you say about it gets  
forgotten, you can infer all sorts of things about a document from  
the content of a response. We (AWWSW) have the option of formalizing  
this, if we so decide.

> That a particular description may be entailed in the deployment of  
> a name by a naming authority (eg. by association through a link  
> header or one of the other mechanisms in [2]) may lend it some  
> particular weight (eg. its a "URI Declaration"), but it most  
> respects it is just another description like any other that you  
> might find. You still have to evaluate whether it is worthy of more  
> regard than a description of the same thing given elsewhere.

I agree.

> Regards
>
> Stuart
> --
> [1] http://www.ibiblio.org/hhalpin/homepage/publications/ 
> indefenseofambiguity.html
>    "It places the responsibility for deciding the relationship  
> between referring
>     and accessing at the wrong end of the communication channel,  
> that of the person
>     who hosts representations accessible at the URI, not the user  
> of the URI."
> [2] http://esw.w3.org/topic/FindingResourceDescriptions
> [3] http://www.ihmc.us/users/phayes/PatHayesAbout
> [4] http://www.ihmc.us/users/phayes/PatHayes
> [*] This neglects such purtibations as might be induced by a change  
> in ownership of a domain name, or the wholesale reorganisation of a  
> site that results in reuse of some of the names therein.
> --
> Hewlett-Packard Limited registered Office: Cain Road, Bracknell,  
> Berks RG12 1HN
> Registered No: 690597 England

Received on Thursday, 3 April 2008 20:55:36 UTC