W3C | TAG | Previous: 21 Jul face-to-face meeting | Next: 4 August 2003 teleconf

Minutes of 28 July 2003 TAG teleconference

Nearby: IRC log | Teleconference details · issues list · www-tag archive

1. Administrative

  1. Roll call: NW (Chair), DC, TBL, PC, CL, Pat Hayes [PH], TB, DO (briefly), IJ (Scribe). Regrets: SW. Missing: RF.
  2. We did not accept the 21 Jul face-to-face meeting minutes. DC looked at them and did not find them acceptable as is.
  3. Accepted this agenda.
  4. Next meeting: 4 August teleconf. Regrets: DC, DO
  5. Resolved: Meet face-to-face in Bristol 6-8 October 2003 (Mon to Wed) with the expectation that some people may participate by telephone or video remotely. DanC gave regrets for any type of participation.

2. Technical

The TAG invited Pat Hayes to the meeting to discuss the terms URI, identify, resource, representation, and issue httpRange-14.

[Ian]

NW: For technical agenda, we can either talk about (1) httpRange-14 or (2) define resource and representation.
DO: In case of a vote, my proxy vote goes to TB. [DO leaves.]
TBray: Procedural point: If we write up text along the lines we agreed to at the ftf meeting (information resource mentioned), I think several of us can live with that compromise. We can address the issue(s) in detail in a subsequent version of the arch doc.
[Chris]
I agree that information resource really helps as a concept
[Ian]
DC: On "information resources": Roy said there was no such thing. We did not *decide* to make the distinction at the meeting. I think RF is on the record as saying we should not make the distinction.
[Zakim]
timbl, you wanted to mention that Roy had views on this and to proceede without him is a shame.
[Ian]
TBL: Right, I think RF opposed the notion of information resource. RF's absence today is a shame. We invited Pat Hayes (PH) to discuss the use of terms. I think RF might agree that in practice there are information resources, but he would not like to make the distinction in the model.
PH: Please explain RF's position. Is the position that there is no such thing as an information resource, or that the distinction is not useful?
TBray: I think I can convey RF's position. RF and I both observe that the existing deployed base of software has no opinion about what the nature of a resource is. Deployed software doesn't care whether the resource is a mountain or a picture of a mountain. The distinction has nothing to do with respresentational state transfer. While I agree with him technically, I am aware of the angst caused by the issue.
DC: In particular, RF has pointing out that http URIs (without #fragid) exist in practice that refer to robots (not information resources).
TBray: Another example is XML namespace URIs that begin with http and have no frag ids.
PH: Seems like XML Namespace URIs are a good example of URIs that (can) have nothing at the end. It's hard to get ahold of the namespace. You get documents back saying "I am a namespace."
PC: Don't forget the use of Namespace URIs as declared without making available any representations.
PH: Seems frequent to have URIs without representations available; no need to make this illegal. When you get persnickety about nature of resource, you continue to find ambiguities (e.g., resource at given moment in time v. resource at any moment time).
TBray: Suppose we proceed in document by making distinction between information resource and "other types" of resources. TBL has said that ambiguous denotation with URIs is dangerous to sem web. What would need to be said in arch doc to make building sem web sanely possible?
PH: What bothers me is that there is the axiom on the current draft: The claim that a URI must *identify* a unique resource.
TBray: What do you mean by "unique"?
PH: If the axiom could be weakened or removed, a lot of these problems would just go away.
TBL: There's a philosophical debate issue (denotations and interpretations). But there are practical problems when someone wants to use a URI to refer to a page and also to a person. These people haven't been playing with the semantic web.
[Zakim]
timbl, you wanted to explain NS and to also mention Roy's model of all the bits as being representatation of the robot.
[Ian]
TBL: There's a philosophy question (how do we determine we mean the same thing when using URIs). But there's another thing (hair-splitting) about whether we mean a photo or a photo including its frame. I'm worried about neither of these (for the moment). I am concerned when people are expressly referring to two things with the same URI.
PH: For the purposes of today's discussion, I agree with TBL.(But I don't actually agree with TBL) I agree that the current technology doesn't care what the nature of the resources is.
[timbl]
Pat: Current technology not on teh sem web doesn't give a rat what these resources really are.
[Zakim]
tbray, you wanted to ask in what sense it is unique and to ask where the assertion Pat talks about is made
[Ian]
PH: The problem is what's said in the arch doc. The document says important about resources that matter.
[Zakim]
also, you wanted to menation Roy's model of all the bits as being representatation of the robot.
[Ian]
TBray: What language is bothering you in the 16 July Web Arch draft?
2. Identification and Resources "Uniform Resource Identifiers (URI), defined by "Uniform Resource Identifiers (URI): Generic Syntax" [URI], are central to Web Architecture. Parties who wish to communicate about something will establish a shared vocabulary, i.e. a shared set of bindings between identifiers and things. This shared vocabulary has a tangible value: it reduces the cost of communication. The ability to use common identifiers across communities is what motivates global naming in Web Architecture."
[TB reads second para as well]
PH: I don't establish a link to a galaxy by using a URI. Let's define "link"
[Chris]
link is a context of use of uris
[Zakim]
Norm, you wanted to ask how to distinguish between the nits and the real distinctions
[Ian]
PH: URI-makes-link if we think about resources as being networked resources.
TBray: I'm sorry, I just don't see the problem.
PH: How do you link from an imaginary entity to something 100s of 1000s of light years away.
[Zakim]
timbl, you wanted to mention confusion betweb Rs and IRs
[Ian]
PH: You can link the representations, but not the things.
[Chris]
PH just said what I was queued up to say!!
[Zakim]
Chris, you wanted to point out links are only found in resource representations
[Ian]
DC: The doc says "When a REPRESENTATION of one resource..."
CL: You can only have a link from a representation. The link is to a resource, not a representation. (CL: Modulo fragid nonsense.) One knows about links by fetching representations and determining that there's a link. A link IS formed between resources; the link is accomplished via representations.
[Chris]
a link IS NOT* fomed merely by the existence of two resources
a link has to be explicitly established, in a representation
not all representations have links
[Ian]
PH: "shared set of bindings". Can we assume that looking at this from a sem web that "parties" can be software agents?
DC: Yes.
PH: So how do software agents establish a shared vocabulary?
[Chris]
eg I can have an image in SVG that has links, and a JPEG image that does not
[Ian]
DC: The document doesn't say that they "have to", we just observe that they do.
[Chris]
and those could be two representations of the same resource
[Ian]
TBL: Software agents pick up knowledge by being written by humans.
[timbl]
http://www.cacr.caltech.edu/~roy/vermeer/ucyh.jpg "The Astronomer"
[Ian]
TBray: It's safe to assume (by software agents) that same URI refers to same thing.
DC: Both names and what they refer to is bootstrapped.
PH: There's no way to communicate "the thing". You can only refer to it with symbols.
DC: That's exactly what we do.
PH: It works between people in a room because they all see the dog and observe understanding.
TBray: Why doesn't it work on the Web.
PH: Vocab is defined in terms of bindings, not shared URIs. I don't think that's true. Software can do a lot without knowing bindings. It doesn't matter in some cases whether there is even a binding. Only agreement is the agreement to use URIs in the same way (in a given context).
TBray: I agree with PH here - I think the discussion of "shared set of bindings" is gratuitous; we never actually define the bindings. We could delete that phrase; we don't need to talk about bindings at this point in the doc.
DC: But names refer to something. There is tangible value when our views of binding are the same.
[DanC_g]
"a shared set of identifiers on whose meanings they agree"
I like that
[TBray]
+1
[Chris]
I observe that in real life, subsets of communities can agree on the meaning of a given term, but entire communities rarely ever do. hence schools of thought, different factions, political parties, and so on. a canonical set of definitions only goes so far
[Ian]
TBL to PH: On this phone call, we say "Pat's on the queue." Pat is animate, the queue is virtual. But there's no confusion about these things. We've exchanged a huge amount of information, and it would be inconceivable to be confused about what "Pat" means. A vast number of URIs will work that way on the semantic web. E.g., those published by the OWL WG.
[TBL on cost in time of continuing to debate fine points.]
[Zakim]
DanC_g, you wanted to reiterate pat's proposal
[Ian]
[PH proposal: "establish a shared set of identifiers on whose meanings they agree."]
PC: Do we define "identifiers"?
TBray: I don't think we need to define "identifiers"
DC: It's clear that we mean URIs.
[timbl]
http://www.cacr.caltech.edu/~roy/vermeer/ucy.jpg
[Ian]
Proposed: s/a shared set of bindings between identifiers and things/a shared set of identifiers on whose meanings they agree
NW: If we adopt this, does this help clarify what we mean by resources/respresentations?
Resolved: In section 2, s/a shared set of bindings between identifiers and things/a shared set of identifiers on whose meanings they agree
[Ian]
[Second issue is on information resources.]
PH: "The networked information system is built of linked resources, and the large-scale effect is a shared information space. The value of the Web grows exponentially as a function of the number of linked resources (the "network effect")." Whoa. This seems to be talking about information resources.
DC: I agree.
TBray: I don't agree.
TBL: In my terminology, you have a picture/form of a robot; those are information-bearing objects.
TBray: Software can't tell the difference.
TBL: My software can.
[On meaning of "link"]
[Norm]
Stupid is not illegal.
[DanC_g]
but harmful can be, and perhaps should be, promoted to counter-to-web-architecture
[Zakim]
timbl, you wanted to discuss what <http://www.cacr.caltech.edu/~roy/vermeer/ucy.jpg> is.
[Ian_]
TBL: Web works because we have expectations about the same content.
NW: What if you do a GET on a URI and get back an RDF representation that says "That URI refers to a person."
[DanC_g]
I think this "expectation of same content" issue is much more subtle... doesn't work for W3C home page, for example, which has different information on different days.
[Ian_]
TBL: The system would stop. "I'm sorry, I told you that the URI refers to a person; not a painting."
NW: If I own a URI, I don't know why I don't get to say definitively what it refers to.
TBL: It's useful to be able to limit scope to information resources rather than have to call up a person to ask what a URI refers to.
[DanC_g]
we can, and we do go thru doing just that.
[Ian_]
TBL :The Web is built of networked information objects. The identity of those things is defined by what is invariant when you do GET with that URI.
[timbl]
and what is invariant is that that is a picture of an oil painting.
[Ian_]
PH: If you say that what the URI denotes is fixed by the owner, then any URI can denote anything.
[Chris]
yes, and that would be useless, but is still possible
[Ian_]
PH: I don't think that's feasible as a network architecture.
[Zakim]
DanC_g, you wanted to state my position on httpRange-14: I don't think we shoulld resolve it in this version. I think we should make and use the distinction between resources in general and information resources; i.e. those resources that can have representations.
[Ian_]
DC: On this issue, I don't think that we should resolve it entirely in v1 of arch doc. There's a lot of work to be done before we do.
[timbl]
PH: If it were really true that yo had to ask someone what their URI meant, the web would not work. It isnt a working network architecture
[Ian_]
DC: But I do think it's useful to make the distinction between information resources and other resources. It will help the community talk about the problem.
[timbl]
seconded
[Zakim]
TBray, you wanted to say that the information system includes only those things for which people publish URIs, and they're only good citizens if they make representations available
[Ian]
PH: I don't know what it means to build a networked ifnormation system with galaxies.
DC: Is it useful to make a decision about adding "information resource" without RF here?
NW: I'd like to move forward even if RF's not here. He can object.
Proposd action DC: Propose text for architecture document that distinguishes "information resource" from other types of "resources".
DC: I'd like to resolve to include such language.
PC: No chance.
[Others may have said no as well]
DC: I don't accept the action if we are not deciding.
Action TB: Propose text for architecture document that distinguishes "information resource" from other types of "resources".
[On httpRange-14]
NW: I fear that we simply disagree. What's the best way to frame the discussion that will be constructive?
DC: I move to adjourn.
NW: Several people wrote back and said that my summary was flatly wrong.
[timbl]
Uniform Resource Identifiers (URI), defined by "Uniform Resource Identifiers (URI): Generic Syntax" [URI], are central to Web Architecture. Identifier here is used here in the sense of name. Parties who wish to communicate about something will establish a shared vocabulary, i.e. a shared set of bindings between identifiers and things. This shared vocabulary has a tangible value: it reduces the cost of communication. The ability to use common identifiers across URIs identify resources. A resource can be be anything. Certain resources are information resources, which convey information. These are termed information resources. Much of this document discusses information resources, often using the term resource.
An information resource is on the Web when it can be accessed in practice. When a representation of one information resource refers to another information resource with a URI, a link is formed between the two resources. The networked information system is built of linked resources, and the large-scale effect is a shared information space. The value of the Web grows exponentially as a function of the number of linked resources (the "network effect").
[Ian]
PH: Perhaps question producing more contention than it needs to. If an http URI is used with "#fragid", then, it should be that the URI before the "#" SHOULD denote an information resource.
[Chris]
oh, magic hashes again
[Ian]
PH: This gives an opening to folks like Patrick Stickler. But the point is not to have one's cake and eat it too.
[Zakim]
DanC_g, you wanted to doubt that hayes's suggestion helps... I'm pretty sure roy thinks robot#topPart works while robot refers to a non-document
[Ian]
DC: I don't think this suggestion helps. RF might say that <URI#top> refers to top part of robot and <URI> refers to entire robot.
[TB notes that MIME type doesn't say anything about resource, just type of representation]
[Chris]
its not clear that resources even have a type
[DanC_g]
not in general, no, chris.
[TBray]
It's pretty clear that they *don't* have a "type" absent external assertions e.g. in RDF
[Ian]
PC: Imagine I had an XML namespace (paulcotton.name/foobar) describing a bunch of things, and I don't make available a namespace document. And I want to refer to subpieces of the namespace. I can do something like #part1, #part2 to refer to pieces. That seems to make something illegal that folks are already doing today.
[TBray]
paulcotton.name/foobar#2
[Ian]
PC: I use /foobar#part1, /foobar#part2.... I hear PH saying that if I use "#fragid" then there had better be a document available even when the fragid is stripped.
PH: Yes, I was saying that.
[Chris]
Is 'a document' the same as 'an information resource representation'
[TBray]
don't think so, Chris
I think a document is a representation
[Ian]
TBL: The question is whether I can use a URI to refer to a painting, or what magic I have to do to figure out whether the URI refers to a painting or an information object that refers to it. I'd like to be able to refer to an invoice for a robot, and be sure that someone else doesn't use the URI to get the sound of the robot hitting the floor.
NW: That might happen; there's nothing that can be done about it.
[TBray]
NW: shit happens
[Ian]
TBL: But that case is broken. People shouldn't do that. It's damaging.
TBray: We have language to that effect (on ambiguity).
TBL: I want language that says that if you use a URI that refers to a picture and to a person, that that's wrong.
NW: I agree that that's wrong. But I can't swallow assertions related to URIs "with #".
[Chris]
'wrong' and 'inconsistent' are human value judgements and as such, it will be possible to argue for and against them
[Ian]
TBL: Which assertion is wrong?
NW: Don't say that the URI refers to a document.
DC: TBL's argument is rationale, but it's not compelling.
TBL: So the argument that the information content will always be there is not compelling?
[TBL and CL disagree whether consistency is a human value judgment.]
DC: The CYC ontology is coherent, but saying it's web arch at this point seems premature to me. Not every web master has agreed to CYC documentation and agreed to it.
[TBray]
The genie's out of the bottle already, just like qnames in content
[Ian]
TBL: The cost of not agreeing to this point is very high. The language (which one?) will have to be reverse engineered in a year.
NW: I don't think TBL has made the argument in a compelling fashion yet.
[Norm]
what tbray said
[Ian]
IJ: Any summary on this part of the discussion?
[DanC_g]
(I'm OK with gaps in the IRC log; in fact, if people have higher expectations than that, they should think again.)
[Ian]
TBray: no.
[DanC_g]
hey... we got a decision about changing "bindings" to meaning! That's non-trivial!
[TBray]
This isn't supposed to be easy
[Ian]
NW: Thanks to all, especially PH.
PH: I won't make my "crazy suggestion" anymore; it's been shot down. :)

3. Not discussed

Actions related to Architecture Document

Action items related to SVG spec that IJ does not intend to continue to track here:

Identifiers (URIEquivalence-15 , IRIEverywhere-27)

Qnames, fragments, and media types(rdfmsQnameUriMapping-6, fragmentInXML-28, abstractComponentRefs-37, putMediaType-38)

New and other Issues requested for discussion. (mixedUIXMLNamespace-33, RDFinXHTML-35, siteData-36 plus possible new issues)

Existing Issues:

Possible New Issues

Findings in Progress

[Expect contentTypeOverride-24 and whenToUseGet-7 to have made significant progress ahead of F2F. xmlIDSemantics-32 has a stable and mature draft finding and XML Core WG are working toward a resolution of this issue.]

Other issues

Other actions

Ian Jacobs for Norm Walsh and TimBL
Last modified: $Date: 2003/07/28 22:48:23 $