W3C home > Mailing lists > Public > w3c-rdfcore-wg@w3.org > May 2004

Re: comments on Web Architecture First Edition

From: Roy T. Fielding <fielding@gbiv.com>
Date: Fri, 7 May 2004 16:47:56 -0700
Cc: public-webarch-comments@w3.org, w3c-rdfcore-wg@w3.org
To: Pat Hayes <phayes@ihmc.us>
Message-Id: <F68ECF5F-A080-11D8-A255-000393753936@gbiv.com>

I have no doubt that there exists some inconsistent use of terms
within the webarch document.  However, the way to fix them is to
point to the individual inconsistencies using factual statements.
I have no respect for arguments that are based on taking single
examples out of context and making claims that they somehow define
the entire space in exclusion of all other potential examples.

> The key distinction here is identifying the thing - what Tim calls 
> "name" -  versus identifying the location - "address"; a difference 
> which is made vivid when the thing moves to a different address. OK, 
> good point, one worth making in an architectural context: but not what 
> I am talking about. I am talking about 'names' (not in Tims sense) 
> which have got NOTHING WHATEVER TO DO with addresses of any kind, 
> which are entirely removed from any architectural or computational 
> sense of 'naming' at all, and which are not 'identifiers' in the same 
> sense (because they cannot in general be "dereferenced" in the sense 
> that Tim is talking about): the kind of names and naming that were 
> around in human society ever since apes started grunting to indicate 
> things in the external world, the kind of names that pre-dated the 
> computer by several dozen millennia. Just FORGET about computers and 
> networks and architecture and the entire WWWeb for a second, if you 
> can, and just think of LANGUAGE and REFERENCE. How WORDS refer. That 
> is what I am trying to talk about. That is what the Sweb semantics is 
> an idealization of. It has got almost NOTHING directly to do with 
> networks.

Sometimes, yes.  I fail to see why the EXAMPLES that are provided in
the document are somehow DEFINITIVE for the entire space of resources,
or that the document must give equal weight to all examples of resources
and their interconnectedness.  You are simply assuming "for all" in 
cases
where no such assumption should be made, and not specifically 
identifying
cases where the document does make that assumption and should be fixed.

> [...]
> > In particular, in sense (C), but not in sense (D), there is a
> > presumption of a computable or effective process which can be applied
> > to the identifier to provide access to the entity identified;
>
> no, I don't believe there is any such presumption in the webarch
> document.
>
>
> Oh, surely this assumption is implicit throughout the document. It 
> says for example that resources are connected by networks, that they 
> can be accessed and operations performed on them by network protocols, 
> etc. ; less directly, the idea that resources have a unique identifier 
> has no rational basis for anything other than nodes which must be 
> somehow addressed on a network in order to support transmission 
> protocols.

Where does it say that all resources are connected by networks?
Where does it say that all resources can be accessed?
Where does it say that all resources have a unique identifier?

Those are obvious contradictions with what has been written in
rfc2396bis, and if they exist in the webarch document then I want to
know where so that we can fix them.  I don't see them, but at this point
I've read the document too many times and my eyes tend to glaze over.

> [...]
>
> > 2. Hunting down what is meant by "resource".
> [...]
> > The latter (D) interpretation seems to be insisted upon by the cited
> > document <http://gbiv.com/protocols/uri/rev-2002/rfc2396bis.html>
> > which reads:

It has been changed since then.

> > Which could be paraphrased as "A resource can be anything, and
> > everything is a resource".
>
> Well, then, it is hard to resist asking the question, why did y'all 
> feel obliged to (mis)-use a word when there already were perfectly 
> good words you could have used, such as "entity" or even the plainer 
> "thing" ? Grice's maxims come into effect at this point in the 
> reader's mind: they aren't using the ordinary word, so they must 
> intend a special sense...try to figure out what it is....

Do you really think that those words would be any less likely to be
misinterpreted?  In any case, entity is typically used to refer to
an embodiment of some type, and not all "things" are resources, so both
examples are more overloaded given what we want to use resources for in
the SemWeb.

We have already talked about why more of the rationale behind "resource"
is not in rfc2396bis.  To some extent, I have improved the definition
since then, as the number of obstructionists have decreased over time.
However, my recollection was that you could not come up with a 
definition
that was any better, nor any useful recommendations for better 
terminology.

> Is wood something which can be connected to a network-spanning 
> information space by a link?

Yes.

> Is wood something on which operations can be performed by means of 
> network protocols?

Yes. (assuming the network protocols are only one part of an overall 
system)

> Is wood something that must have a unique identifying URI? (What would 
> that claim even MEAN? what could it POSSIBLY mean? )

No.  I have no idea what you are talking about here.

> > The only example given in the document is disturbingly vague at
> > precisely this critical point: the resource is the "Oaxaca Weather
> > Report". But what KIND of thing is that,
>
> It's a resource.
>
>
> AAARGH. WHAT DO YOU MEAN BY "RESOURCE" ???
>
> That answer is VACUOUS since you have not said what you mean by this 
> damnable word. (It does mean NOT mean resource in the ordinary English 
> sense of "resource". )

Yes, it does, though you seem to have trouble believing it.  Some people
use the term "resource" to mean people, others to mean "oil", still 
others
to mean "land", etc.  The one thing that the ordinary English sense of
the noun "resource" has in common is that a resource is a source of
supply or support within the context of a given system, whatever that
system might be.  A system like a "hypertext browser" is going to have
a very different set of resources than a system like a "theorem prover".
Nevertheless, the things that they operate on can be termed resources,
can be identified by the same system of identification (URI), and over
the past 10 years I have yet to encounter any other word that can
express the same sentiment in a reasonable number of characters.
Whether or not all systems can make effective use of all URIs is
completely irrelevant to whether or not a thing can be a resource
to one system.

In short, if you feel that "resource" means something other than the
above in the ordinary English sense, then please feel free to explain
what your sense of the English IS rather than simply complaining about
every definition that has ever been proposed.

> There is a real need to explain what you mean when you use words in an 
> unusual way. This is not just a quibble: until you SAY what you mean 
> by this highly non-standard usage, the entire document is literally 
> USELESS.
>
> I find it hard to believe that we are even having this conversation. 
> If the TAG group were a bunch of students in a high-school English 
> class they would get an F immediately for not understanding this basic 
> point.

I am surprised that a person working on the semantic web would assume
that all communities would use the same term in the same way, let alone
in the way that you claim (but have never actually shown).  I require
that the terms be defined and used consistently within the scope of
a given set of specifications, but you reject the definition AND then
complain that its usage is inconsistent with some imagined definition
based on the Web examples.  The inconsistency is in your assumption
that the examples are somehow covering the entire scope of what is
being described by the Web architecture.  The examples provide a way
of relating the concepts within the architecture to a familiar set of
behaviors and interactions within one system based on that architecture.
If you can think of additional examples that show a wider variety of
resources, preferably based on real systems rather than potential
future notions of what you think the SemWeb is about, then
please send them as your comments.

In fact, your entire tirade is basically taking one sentence,
reinterpreting a few of the words in a sense that artificially
bolsters your argument, and then claiming that those reinterpreted
definitions somehow restrict the semantic web.  I don't believe
your paper tiger has any bite, and if you took the time to read
the document without prejudice then maybe you could help identify
the places where the architectural principles are bogus (there are
a few) or where the prose overstates a practice as being universal
when it, in fact, is only applicable to one of the systems that
use the Web.  Those are valid and helpful comments.  Claiming that
you have a special understanding of the English language that differs
from that used in the document and  invalidates the work we have
put into describing the Web architecture, is not only unhelpful --
it is obviously wrong and I have shown you, many times, where your
claims about the meaning of words are a complete fiction.

> I was not asking about Web architecture; I was asking about the 
> MEANING of the WORDS you use to describe it. As I said, almost the 
> entire document makes perfect sense under the reading that you (and 
> Roy) reject, the one under which it would indeed be a document 
> describing an architecture, and did not try to simultaneously be a 
> tract based on an ill-thought-out (and unpublished) 
> operationalist/pragmatic philosophy of language which apparently is 
> based on the idea that reference and physical connection are identical 
> and which assumes that every entity has a unique True Name. If you 
> really believe this stuff, I think you ought to write it up and send 
> it to a journal, or at least put it up on a Blog somewhere where 
> people can comment on it.

Again, I have no idea what you are talking about.  Nobody on the TAG has
ever assumed that every entity has a unique True Name, though some
people (myself excluded) would like to require that every name have one
true interpretation, and most people agree that it is better to use one
name rather than several different names to refer to the same resource.
That does not prevent the assignment of multiple names to a resource,
nor does it ensure that names that have been assigned are used in the
way that they were intended, but it does state a preference that folks
should not be intentionally inconsistent in their use of names.

I assume that your comment on an unpublished philosophy of language
is based on prior messages I sent to you about the background of REST.
I assure you that what you are complaining about above did not originate
with me and has nothing whatsoever to do with REST or its background.
Since none of that background appears in this document, I'll just put
your comments down to an arrogant assumption that you can evaluate my
work even though you have never seen it and do not, in fact, have
any experience in related fields.  Maybe if you weren't so prejudiced
against finding any meaningful insight in the webarch document, you
could find time to comment on what is actually published rather than
on your assumptions about something you have never seen.

> It is not written with enough precision to know whether it is 
> consistent or not. It SEEMS to contain inconsistencies, in the sense 
> that words and phrases  are used apparently with different and 
> incompatible meanings in different parts of the document. I have 
> documented this phenomenon in detail.

You documented the phenomenon?  Why don't you simply document the
instances where usage differs from definition?  The only things you
highlighted were simple examples used to relate the definitions to
the real world.  One solution would be to add more examples.

> What you have written is likely to be very useful to people who are 
> not interested in reference or descriptional semantics. I wish it were 
> written clearly with that (large) audience in mind. In its present 
> form, however, the document is not only not helpful, it is positively 
> harmful, to clarifying the relationship between SW semantics and Web 
> architecture. I would describe it as corrosive, a major disaster. I 
> think it, or, more properly, the confusion that it embodies, fails to 
> clarify, and in places authoritatively reiterates, has been and 
> continues to be a constant source of misunderstanding, controversy and 
> potentially bad design decisions, and that until this mess is cleared 
> away we will never get a sensible coherent account of the foundations 
> of the semantic web.

That is quite possible.  Not a useful comment, however.  It would
probably be more coherent if we wrote several different documents,
as suggested by another commenter, each one with an individual author
that can maintain consistent use of terminology throughout their
document.  In the long run, though, I doubt that would help attain
any sense of consensus, and in any case we barely have enough spare
time to collaborate on just one document.

> > Trying to home in on your intended meaning by searching the document
> > for uses of "resource" gives the following:
> >
>
> > [[The World Wide Web is a network-spanning information space of
>
> > resources interconnected by links. ]]
> >
> > I take it then that a resource is something that can be connected by 
> a
> > link to another resource. I presume also that "link" here means more
> > than simply a reference to something,
>
> why?
>
>
> Because that is the usual meaning of the words "link" and  
> "interconnected" . Reference is not in any sense a connection. I can 
> refer to things that do not exist, or have ceased to exist, or which 
> might exist but do not yet exist: I cannot be interconnected to them, 
> by links or anything else. I can refer to things that I cannot be 
> connected to by virtue of physical limitations, they being light-years 
> away. I can refer to abstractions such as numbers or sets, for which 
> it would be logically incoherent to even suggest that one could be 
> connected in a network sense. I can refer ambiguously (in fact, I 
> would argue that this is the normal case) but I cannot be ambiguously 
> connected.  I can refer in one possible interpretation and refer 
> differently in another, but I do not know how to get possibly 
> connected in an interpretation. And so on.

That is, of course, a hopeless misinterpretation of the word connected.
Have you ever seen the BBC series (shown on PBS as well) on the history
of the world, called "Connections" (James Burke, 1978)?  For that 
matter,
just look it up in any dictionary:

     http://dict.die.net/connection/

I completely reject the notion that "interconnected" implies a physical
connection of any kind, just as I completely reject the notion that a
hypertext link (a reference that uses a URI) is in any sense a physical
notion that requires that the referent exists.  There is no basis for
your claims.

> If you do not understand this then I despair of explaining it to you. 
> Surely it is OBVIOUS that referring to something is not being 
> connected to it? Do you think you are linked, in the sense implied by 
> "a network-spanning information space of resources interconnected by 
> links", to , say Santa Clause, or Nero, or the great nebula in 
> Andromeda, or to my as yet unborn great-grand-daughter, or the base of 
> the natural logarithms? Do you think that simply by mentioning 
> something you are thereby enabled to perform operations on it, using a 
> kind of network version of Voodoo?

No.  Nobody here has ever said that all resources are connected to all
other resources, nor that all resources are accessible by means of
network protocols, nor that making a reference somehow guarantees
that operations on it will be successful.


Cheers,

Roy T. Fielding                            <http://roy.gbiv.com/>
Chief Scientist, Day Software              <http://www.day.com/>
Received on Friday, 7 May 2004 19:49:09 EDT

This archive was generated by hypermail pre-2.1.9 : Friday, 7 May 2004 19:49:13 EDT