AWWSW -- 31 Mar 2009

GenericResource

<dbooth> http://www.w3.org/DesignIssues/Generic.html

<dbooth> dbooth: GR doesn't seem to capture which represetnations are captured, but REST and ftrr:IR do.

timbl: Generic resource puts constraints on representations

if X u:isRepresentationOf G, and r is a representation of X, then r is a representation of G

(X is a u:FixedResource)

timbl: I should probably fix [the dual use of the word 'representation' to mean fixed resource vs. REST representation]
... representation invariant = always the same mime type

dbooth: there are three axes: mime type, language, and time.

jar: so fixed = invariant along all three axes
... Sorry, mistake in the above I think (u:isRepresentationOf etc)

timbl: Should rename 'representation invariant' to 'content-type invariant'

<dbooth> dbooth: So GR has three axes, but it looks like the names used on the third axis (the content type access) should be changed from "Representation" to "ContentType".

<timbl> http://www.w3.org/2006/gen/ont

jar: I was looking for a property that is the conjunction of all three properties - 'X is a fixed resource having as its sole representation one of the representations of G'

<dbooth> dbooth: So the axis should be called 'Content Type', the class should be 'ContentTypeInvariant' and the property should be called 'isContentTypeInvariantVersionOf'.

<dbooth> ... Also it looks like u:isVersionOf should be renamed u:isTimeInvariantVersionOf to be clearer also.

<dbooth> timbl: This metadata might be provided by a link header, for example.

timbl: this ontology (GR) would be found via, say, a Link: header and is meant to explain some particular resource

<dbooth> dbooth: Might also define a property u:isFixedVersionOf, which would be a subproperty of u:isTimeInVariantVersionOf, u:isLanguageInvariantVersionOf and u:isContentTypeInvariantVersionOf.

jar: OK, right now I don't care what it's called. isFixedVersionOf will do
... isTimeSpecificVersionOf ?

<dbooth> Oops, I shoudl have used hte word 'specific' instead of 'invariant', so the above should have been:

<dbooth> u:isTimeSpecificVersionOf, u:isLanguageSpecificVersionOf and u:isContentTypeSpecificVersionOf

isVersionOf = common superproperty

isFixedVersionOf = common subproperty

<dbooth> and finally, the conjunction of all three would be isFixedVersionOf.

<dbooth> timbl: but if you know that x isFixedResource of y, then are you saying that y *is* a GenericResource? What if it only has one language?

<dbooth> jar: So it's a subclass relation.

<dbooth> timbl: Yes, not a proper subclass.

fixed resource is a subclass of generic resource, right?

timbl: Yes.
... Can't distinguish generic resource from information resource.

alan, we're editing http://www.w3.org/DesignIssues/Generic.html

alan, http://www.w3.org/2006/gen/ont too

<timbl> http://www.w3.org/2006/gen/ont

<timbl> http://www.w3.org/2006/gen/ont.n3

<dbooth> jar: How do you determine if something is a representation of a generic resource? what's the truth criterion?

<dbooth> ... if you *have* a rep, how do you decide if it is a rep of a particular GR?

<dbooth> timbl: Then you have metadata and data. When you did a fetch . . . .?

<dbooth> jar: But what if you got it some other way?

<dbooth> ... I want to be able to prove that two URIs name different resources. In principle, what's the theoretical test you could make?

<dbooth> ... Eg for the bible, i could look at the rep and decide if its' a rep of the bible.

timbl: it's one of these open world things

<dbooth> timbl: Typically when you look at the front cover of a book, it says "this printing" and it says 'also available in these languages'.

timbl: Different nonoverlapping author lists might imply that things are different

<dbooth> timbl: You can tell that they're the same, but very difficult to prove different. Like if you have two people. Eg, if you have non-overlapping author/publication lists, then it's a different person, though sometime's there's a mistake.

<dbooth> jar: So it's a much less commital notion than the other models on the table, such as REST and David's ftrr:IR.

<dbooth> jar: The question is whether these models are overlapping, different, the same, etc.

<dbooth> ... David at the start of this call was asking how two of these in particular compare to each other. Need to explore the boundary cases to answer that.

timbl: Roy was happy to say that people are resources *and* have representations

<dbooth> timbl: Roy's REST was pre-SemWeb, so he was happy saying a rep is a rep of a person. Also happy saying a URI for a robot is for a control panel, and also a URI email addr that's used indirectly to identify a person.

<dbooth> jar: But we're only looking at documents here, and he did model documents.

timbl: 302 problem

<dbooth> timbl: I found a bug related to this stuff, it made some assumption from a 302 that threw it into a loop.

jar: Can user-agent affect (some agent's) choice of representation [in each model]?

<dbooth> jar: Another question: can user agent affect the choice of rep?

<dbooth> dbooth: Good question, because 2616 makes clear that conneg can use *anything* in the request to make a choice of rep, but that's not captured in this little GR ontology, which only captures axes of time, language and content type variation.

<dbooth> dbooth: This GR ontology captures three of the most important axes of variation, even if it does not capture all of the variation that 2616 permits.

timbl: One possibility is to say that there are other axes beyond these three

<dbooth> timbl: We still need 3 axes. If you say this is a FixedResource, its independent of not only these three axes, but also all other axes. I kinda like that.

<dbooth> ... So if someone has geographic location of the user, we could add that as another dimension, and FixedResource would have the same meaning.

timbl: Not going to have the ontology in two places (the generic resources design issues document, and the ont RDF file)

alan, can't hear you

<dbooth> jar: For every 200-responding thing, there is an IR, right? Wondering if GRs are meant to be idealized.

jar: Which comes first, the web page, or the generic resource?

timbl: Some generic resources are not on the web, some are.

<dbooth> timbl: For lots of things like wikipedia pages, that page *is* the IR. For other things, there is a work, like a paper is published in multiple places, and they're the same work, whethere we call them the same GR I think is splitting hairs. People who work in libraries would say there is one work and multiple manifestations.

<dbooth> (TimBL departs)

<dbooth> jar: Suppoose yoou have names for documents, and a way to get the document given the name. Then nobody would use http protocol. So if I mention a particular paper, I'm referring to *that* paper, independent of anything the web does.

<dbooth> ... Eg could put an MD5 in the URI, and then you could check to see if it really is the one I meant.

In this fantasy world, the URI names the document, independent of what the web or Internet does.

You could feed the URI to some magic resolving mechanism to get the paper.

So if these 'names' worked beautifully, why would anyone use an http URI?

In HTTP, the protocol is authoritative for the URI.

Names would have idealized persistence.

Resolution protocol: Fetch an octet stream through undetermined means, if md5 is right then ok, otherwise try another

People would like to have names for use in their href=

dbooth: RDF statements are tied to particular times

Given HTTP, no way to tell the difference between a mistake an an intentional change

roles: uri owner, statement author, consumer

in general the consumer won't be able to distinguish "correct" resolution from "incorrect" resolution.

- DRAFT -

AWWSW

31 Mar 2009

Attendees

Contents

GenericResource

Summary of Action Items

Scribe.perl diagnostic output