Minutes: W3C Technical Plenary

3 March 2004

These are the minutes of the fourth annual W3C Technical Plenary held on 3 March 2004 at the Sofitel Royal Casino in Cannes-Mandelieu, France. In addition to the day-long event, thirty W3C Working Groups and Interest Groups held face-to-face meetings over four days at the same location. The public plenary consisted of six sessions. This meeting record is transcribed from the IRC log and is not verbatim. Please send corrections to the W3C Communications Team.

The first annual W3C Technical Plenary took place on 28 February 2001, the second on 27 February 2002 and the third on 5 March 2003.

Also available are photos of the plenary week and Cannes-Mandelieu from Kevin Lawver (AOL), Libby Miller (University of Bristol), and Richard Ishida (W3C), and plenary survey results [W3C Member-only link].

Plenary Day Agenda

two to three hundred attendees seated in the meeting room

Lisa Seeman and Janet Daly seated on stage during Session 3

Noah Mendelsohn speaking at a floor microphone with others in line
Photos: Masayasu Ishikawa

Session One

Steve Bratt speaking on stage
Photo: Masayasu Ishikawa

Session Two

Architecture of the World Wide Web and Hot TAG Topics

Norm Walsh, Roy Fielding and David Orchard seated on stage during the TAG presentation
Photo: Masayasu Ishikawa

Moderator: Stuart Williams (HP). TAG participants: Dan Connolly (W3C Team), Paul Cotton (Microsoft), Roy Fielding (Day Software), Mario Jeckle (DaimlerChrysler) Chris Lilley (W3C Team), David Orchard (BEA), Norm Walsh (Sun Microsystems), Stuart Williams (HP), Ian Jacobs (W3C Team)

Extensibility/Versioning [slides]

Stuart Williams: How should MUST UNDERSTAND get deployed?

David Orchard: In conversation with Paul Biron, we observed it doesn't work with multiple schema documents. MUST UNDERSTAND which schema?

Stuart Williams: It may prevent you from understanding other pieces that you could have understood.

Al Gilman: Regarding two-phase commit, you have an optimistic interpretation, but there is a transaction implied. You don't complete it when you don't understand.

David Orchard: A node must look at all subnodes and must understand all MUST UNDERSTANDs in them to proceed.

Henry Thompson (W3C): In designing XLM Schema, a number of Members said we should solve the versioning problem before publishing version 1, but we didn't. It doesn't solve the versioning problem. It's a well known unsolved problem, and we're in good company. The value of this draft finding is that we're picking out small areas where we can say "these techniques are recommended." The dimensions of analysis are good. Distinctions between circumstances where you have a new schema with an old app is very different from when you don't know the new schema with an old app. There is good work [W3C Member-only link] by Eduardo Gutentag and Arofan Gregory in ebXML. Your finding has focus on the "we can't assume that you know the new schema" case. If you have independence of one element from another, then MAY IGNORE becomes viable. The categorizations are a huge step forward.

Paul Biron (Health Level 7): You had a couple of techniques on your slides for schema support, substitution groups. One more that I like - partial validation. We went to a lot of trouble in the Schema spec to permit partial validation. We didn't get it 100% right but input is useful.

David Orchard: I asked the Web Services Description Working Group - When WSDL does an import of a schema, can I do so with partial validity? I think we should do more in the area of partial validity.

Paul Biron: Perhaps the semantics of MUST IGNORE can be tied to parts of the schema.

David Orchard: Errata?

Paul Biron: I don't think so. A Note maybe.

Micah Dubinko in IRC: TAG team :)

Paul Downey (BT): With thousands of Web services, the natural way of versioning is to copy and corrupt them. If the Web worked that way, we wouldn't have a Web.

Norm Walsh (Sun): XML Query/XSL Working Groups work on the Data Model is building useful data models from partially valid input, relevant to these discussions. We're looking at this in WSDL, but it goes beyond that. If you're building for future, you need to anticipate change. Re XML 1.0 to 1.1 it's a major change that needs to be communicated. A major thing missing - Having put up minor versions, we need to indicate how these versions relate. We have a notion of backward compatibility, but the idea of partial validation is attractive.

Al Gilman in IRC: I am interested how much David Orchard's discussion of partial validity and interoperation resonates with what I think is the agenda for language blending.

Philippe Le Hegaret in IRC: Al, that was my thought when I replied on the html-cg suggesting that WS people might be interested in it

Paul Cotton (Microsoft): If we talk about extensibility and versioning and only focus on schema and ignore the operators that will pass over that XML, then we're only solving a small part of the problem. If we get a schema extensibility mechanism, and a function falls apart, we haven't solved the larger problem.

Dan Connolly (W3C): Re how to implement MUST IGNORE, one way is to use a new method. There's a pattern - use unused bandwidth in the syntax. XML is great that way because there's a lot of redundancy, whereas with binary there isn't much wiggle room.

Henry Thompson in IRC: agrees with Paul Cotton - another dimension we have begun to discuss is that of the kinds of operation which will or won't break in the face of what kinds of extension or new version. Interested parties are referred to a position paper [W3C Member-only link] by Eve Maler for the Schema Working Group from 1999 (!).

Michael Sperberg-McQueen (W3C): Re Paul Downey's question, there is work by Eduaro and Gregory that is based on using Schema type derivation to provide a backtrail and document the relationship between versions. I also want to record uneasiness with the IGNORE rule - it's underspecified.

David Orchard: I agree. You could have a semantic only change with no syntax, and that isn't compatible. Some will interpret it as meaning "ignore the elements you don't understand", some will interpret as "ignore the tag you don't understand" and the rest won't know the difference. Working on syntax versioning is useful - don't want to boil ocean. It's also over specified - IGNORE is a nice explicit rule, but it isn't the only possible rule. Partial solutions are a lot better than no solutions. But these solutions are relative to a particular kind of processing. To allow data reuse, with declarative semantics, at design time we don't know all the runtime processing expectations. It's very different for a pretty printer to ignore data than a realtime processor.

Web Identifiers [slides]

fantasai in IRC: realtime processor -> purchasing order processor

Al in IRC: I got a contact at SDSC to create domain maps, that is to say schema+operator vs. schema+operator comparative description written in FLORA, a constraint programming language. Bertram Ludaescher (PDF)

fantasai in IRC: MSM 'transformations of a tree' probably means transforming that tree to a different tree of the same kind

PStickler in IRC: If you want to know what a URI actually denotes, rather than guess from some representation, just use URIQA ;-)

Al in IRC: MSM, reflow of a graph into new tree. Infoset as graph is more authoritative than 'serialization' into tree order.

MSM in IRC: fantasai, true. but for everyone who wishes to use XML as the standard format for all valuable data, virtually all processing will take XML trees as input and produce XML trees as output. (Al, s/tree/graph/ if you wish, I'll still stand by it). the upshot is that 'xslt processing' is not a special case of XML processing, but the archetype of it

dom in IRC: [please use "Nickname, " when talking to one another rather than "Nickname:"

chaalsNCE in IRC: If you want to know what other people think URIQA means, you can ask the existing web.

Al in IRC: fantasai, I agree with you. But I think that it is important to make the tree/graph distinction because the minority of use cases where the order conventions of the tree-ification needs to be redone make the Web complete and humane. Can't create a spec that is actually followed with enough reliability without saying 'graph' rather than 'tree.' Principle of lowest level language. Graph is the more primitive geometry, is the absorbing class, and we need to baseline here.

dirkx in IRC: If you want to know what a URI actually denotes, rather than guess from some representation, just use DDDS and uri.arpa. to find out :-) :-) :-)

chaalsNCE in IRC: Seriously, the issue about how the terms are used (when that differs from how the creator of the term thought it would be used) seems important to me...

Jeremy Carroll (HP): Re QName to URI relationship, one of the drivers is the social use is that we need short names because URIs are too long.

dirkx in IRC: chaalsNCE, not sure. you have a URI -> and you can use it as an Identifier, as a Name, or dereference it to get the Resource or get information about the resource; i.e. the URI architecture list of operations.

Roy Fielding (Day Software): Entity references, relative URIs, etc. Yes, it's a long standing issue.

Jonathan Marsh (Microsoft): What did I miss? Where did fragment identifiers go.

Roy Fielding: RFC1738 was done in 1994, 1630 was done in 1993. Between 1994 and last year, fragments have not been in URIs. They have been in URI references.

dirkx in IRC: (since RFC 2396)

Roy Fielding: A community decided they didn't want people to use fragids. The effect was to remove them from the URI spec. But schemes don't have anything to do with fragids, because they're defined separately from the URI resolution process. Rectifying that has been part of the goal.

Rotan Hanrahan (MobileAware): Regarding the meaning of "resource"... The Web page that demonstrates the weather could be today's weather. It's temporal. Might also be en français. Same resource, but may also have different URIs. Could I say yesterday's weather has a particular URI?

dirkx in IRC: Ah - and also define de-referencing :-)

Chris Lilley: You build a consensus on what the URI is supposed to mean. If it says today's weather and it gives you a Barbie doll, it's not correct. If it says "today's weather" and it's 12:01am somewhere in the world, then it's a gray area.

gk in IRC: (I like that - Chris Lilley - "resources are like angels dancing on the head of a pin" (goes on, you can't see them but you can talk about them)

PStickler in IRC: If the URI denotes "today's weather" and the representation is a picture of a Barbie doll, then the representation is less useful, but that has no effect whatsoever on the denotation of the URI!

DanC_jam in IRC: +1 re "you look at the representations and build a consensus"

Stuart Williams: This is a great time to wrap up. [laughter]

karl in IRC: hmmm.... I wonder if a Barbie doll can make my weather for the day

PStickler in IRC: Perhaps the web authority wished to express via a picture of Barbie that "today's weather" is "hot"... ;-)

Session Three

Lightning Tech Talks

Moderator: Janet Daly (W3C Head of Communications)

Micah Dubinko, Steve Ross-Talbot standing, and Lisa Seeman during lightning talks
Photo: Masayasu Ishikawa

DanC_jam in IRC: yeah for lightning talks!

chaalsNCE in IRC: Lightning talks - cool thing.

Make Incubator Mandatory

Jonathan Robie: [summary of 3 minute talk] Encouraged to be controversial. Proposes Mandatory Incubator. Add an incubator phase to the process. You don't need a standard to play with ideas. Prepare use cases.

kplawver in IRC: Not all use cases are important. The group must decide.

holstege in IRC: (so requirements may not be inclusive enough)

David Booth (W3C): The WSDL new charter is a scheduled priority charter and has emphasis on meeting its schedule.

Steve Bratt: Our incubator process is for a maximum of one year.

Art Barstow (Nokia): I suggest simply requiring Working Groups to do requirements (and AC approval) before specification work.

David Orchard: [comment missing]

Jim Larson (Intel): Some process is good. Too much is a burden. I think this is overkill.

Dan Connolly: My new Working Group will focus on similar topics. Should requirements be part of the Recommendation track. We put it in our set of specs.

timbl in IRC: Good idea.

Steve Bratt in IRC: [after session] Incubator proposal under consideration is not proposed to be part of the current Rec track process. It is a complement, with an end in itself. Some subset of Incubator notes may go onto the Rec track, most will not I suspect.

pdowney in IRC: [after session] anything that speeds up the process is good since there is often a window of opportunity for standards, something the W3C may have missed in Web services.

No Semantic Web without Trust; and vice-versa

Daniel Dardailler (W3C): [summary of 3 minute talk] Relates an experience for looking for info on the Web (for example a hotel in Brussels). I always ended up in the booking service, not the hotel site itself. Similar when looking for artists. Similar when looking at University admissions. I want a Semantic Web markup for "I'm the real thing!" I know that the Semantic Web works well on the intranet (within w3.org for example). However that is because we don't have a trust problem. I'd like to be able to tell the hotel manager what to do. I end up looking at the yellowpages.com entry. Do we need to push for PKI technology?

gk in IRC: IST iTrust project

timbl in IRC: We have the tools.

Graham Klyne: I have pasted iTrust (EU project) into IRC log

Daniel Dardailler: We also may be involved in another EU project, QUATRO.

Jeremy Carroll: There is a way to go in this area. PKI is only one bit.

karl in IRC: glazou except if the domain is hijacked at a point ;)

DanC_NCE in IRC: the idea of "I'm the only web site authoritative on the business establishment at (lat, long)" might be something we could build from DNS/whois stuff. hmm.

Tim Berners-Lee (W3C): We've played with this a bit already.

karl in IRC: Trust really relies on social community. Trust is very dependent on the context.

Daniel Dardailler: I think it is the first thing we have to do for the Semantic Web to scale onto the Internet.

DanC_NCE in IRC: re timbl's comment, see Trust

bijan in IRC: A trust link

Jonathan Robie: I'm wondering what the hotel could provide to me in order to establish trust

ivan in IRC: the only uri I found on itrust on the web but the information is not very rich...

timbl in IRC: DanC, could you explain the delegated trust things we did in test/crypto?

Daniel Dardailler: If I know the address I can find it in the yellow pages. Why can't I do the same on the Web?

PStickler in IRC: Printed yellow pages are a trusted intermediary. Why not web-based yellow pages?

JacekK in IRC: will the real YP please step forward?

Dan Connolly: You can say that I'm the only place with this GPS location.

bijan in IRC: Jen Golbeck (a grad student at UMD/MINDSWAP) is in fact organizing a Trust workshop at the International Semantic Web Conference.

Colas in IRC: why don't you just use the web site of the yellow pages of the country to get the hotel coordinates?

Jeremy Carroll: A Web site that says "I'm the Web site" may be true or false. A Web site that claims to represent multiple sites is probably not the hotel Web site. That is my trust algorithm.

tvraman in IRC: but the yellow pages is also a middle-man. What you have today on the Internet is 1000 middle-men claiming to be the authentic yellow pages (moral- a man with one clock knows the time, a man with a million watches is perennially confused).

gk in IRC: There's another project looking at trust-without-PKI, the SECURE project. They're particularly interested in ubiquitous computing (cf. PowerPoint slides).

Colas in IRC: yellow pages is an official, trusted middleman

PStickler in IRC: There can be social/legal ramifications for making claims of identity that are false.

GlenD in IRC: JacekK, "real" is determined by who people choose to trust (as it should be)

DanC_NCE in IRC: trust in general is very hard, but "I'm the web site for this place on the planet" is a much smaller problem.

Rigo Wenning (W3C): This is a typical trust issue (like passports). The IETF has tried this for years. Can we do it in six months? Not sure but we should try.

vibrania in IRC: It is a legal responsibility to be the root signatory, who wants that?

Yoshio in IRC: Trust should be given to assertions, not to people, in its first place, I think.

JacekK in IRC: GlenD, so I trust some metasite to tell me which YP is the real YP? Then I trust the YP to tell me which hotel is the real one... Is there meta-meta before the metasite? etc.

Jeff Mischkinsky (Oracle): People pay Google to become the authoritative site.

DanC_NCE in IRC: [IRC continues after session] see also geourl

timbl in IRC: We have played with combining RDF rules with digital signature and building a delegated trust system from scratch instead of from PKI so that it matches the real trust structure.

gk in IRC: TimBL, cool. Do you have any demo/links that I might mention at the forthcoming iTrust conference?

DanC_NCE in IRC: gk, demo and corresponding slides

gk in IRC: DanC, thanks.

bijan in IRC: Yoshio, Actually, there's interesting interactions between the two.

Colas in IRC: in most countries you cannot buy a domain name just like that. in france, to get a .fr, you must provide a filing in the business bureau

bijan in IRC: I.e., you might (by default) trust what someone says because they are a trusted expert on the topic

GlenD in IRC: Jacek, clearly you need grounding at some level. For instance, you might trust the DNS to map a given name to a directory you trust, etc.

Colas in IRC: so I would trust anything in .nl, not .com

glazou in IRC: Colas, how does someone know he/she can trust a given TLD like that?

Colas in IRC: if it is run by a government I guess. So it is a subject for ICANN, not the W3C

glazou in IRC: Colas, *.nu ?

Yoshio in IRC: but trusting someone in a transitive manner is hard or dangerous thing, I think they say all people in the world could be connected in less than 7 jump (FOAF).

bijan in IRC: Yoshio, see the site I posted (trust.mindswap.org). There are some surprising initial results.

Colas in IRC: glazou, you can have as much trust in .xx as in the government of country .xx. would I trust .tv officials... that's a good question :-)

GlenD in IRC: I always trust TV.

bijan in IRC: And that's pure interpersonal trust on one axis. You can add lots of other considerations

JacekK in IRC: bijan, can you give a URI to the surprising initial results? 8-)

Colas in IRC: I mean, if you cannot trust .nl officials, I would not trust their hotels either

bijan in IRC: JacekK, I believe this paper (PDF) around page 5.

Implications of XML 1.1

holstege in IRC: the character formerly known as noah

Richard Tobin (University of Edinburgh): You said existing schema types could not validate new characters and that this would cause more people to use 1.1. [part of comment missing]

Noah Mendelsohn: I believe many people are thinking they won't see 1.1 content apart from in the places that use these characters.

MR: Are slides available? This is one of the many interop problems introduced in XML 1.1. I believe profiling has done the community a disservice.

Noah Mendelsohn: whatever the merits of XML 1.1 are, the situation is that 1.1 is a Recommendation. Let's go into it with a positive attitude.

David Orchard: We have a versioning problem. There has been some extension work done. Forward compatibility problem. What does a 1.0 processor do with 1.1 features? What we ought to have done is built into 1.0 a forward-ignore.

holstege in IRC: i am very skeptical that a "must-ignore" rule on characters in element names would be a viable option

Michael Sperberg-McQueen: [compares clean XML to clean HTML]

pdowney in IRC: wonders if it is possible to deprecate a rec

DanC_NCE in IRC: yes, pdowney, 7.7 Rescinding a W3C Recommendation (untested, as far as I know. ;-)

timbl in IRC: Must Ignore would just drop new characters? I guess that shows that the choice ignore/stop is not always obvious.

Glenn Adams (Extensible Formatting Systems): This is already a problem [part of comment missing]. We should anticipate XML 1.2, since Unicode is still changing.

Noah Mendelsohn: I think XML 1.1 anticipates that sort of change in Unicode, obviating a need for 1.2.

Glenn Adams: It is hard to predict the future that way.

davidF in IRC: [IRC continues after session] GD, there is already a problem with encoding, and accepting all Unicode chars in the future will be hard

veillard in IRC: must-ignore on characters within the document content seems an heresy to me. you can't trust the parser anymore ... arghhhh

dom in IRC: the difficulty is that the characters are part of the syntax, and ignoring them change the semantics, ie, you can't just ignore the element that says "pay 2000$ to Do[Noah Character]".

vibrania in IRC: Glenn said XML 1.0 already has issue that encoding is open

I'll name that tune in..."

maxf in IRC: costello!

Steven in IRC: my funny valentine

glazou in IRC: Ian Jacobs singing!

danbri in IRC: i like this account of choreography... (in terms of subtle pattern similarities, basically)

Glenn Adams: sounds like difference between content and style

Dan Connolly: just one global model, or several?

bijan in IRC: PStickler, choreographies describe protocols, not interfaces

Steve Ross-Talbot: [gives examples of global models]... There is more than one model just as there is more than one "my funny valentine."

Glenn Adams: I too like this... in a similar way to the "liking" which Dan did, but with a slightly different implementation.

David Booth: Can you contrast with orchestration?

Steve Ross-Talbot: In choreography, there isn't someone telling people what to do. In orchestration there is a conductor.

bijan in IRC: I am now quite enthused about chore in general and ws-chore in particular

timbl in IRC: So is BPEL4WS orchestration?

bijan in IRC: timbl, yes. BPEL4WS describes things from the point of view of one actor.

Michael Sperberg-McQueen: One interesting challenge is finding a notation. Your Working Group faces a similar challenge. Where are we in the search for notation?

John in IRC: but an orchestra also have a score - even if it's John Cage

Steve Ross-Talbot: There are existing notations (including 2 Turing awards winners)

DanC_NCE in IRC: Michael Sperberg-McQueen used the word "adequate"... I've seen "adequacy" used in a technical sense, but I don't really grok. Anybody care to recommend a reference?

Henry Thompson in IRC: Tony Hoare and Robin Milner

timbl in IRC: Robin Milner, the 1/4-calculus man

Henry Thompson: that's 'pi' calculus, for those of you who saw a 1/4 [discussion of the channel's encoding follows. The channel is UTF-8 but the IRC client sent Latin 1.]

GlenD in IRC: I thought TimBL was making some cute joke for math geeks way beyond my understanding...

danbri in IRC: aside - Is there work on ws-chor -related logging? ie. conventions for documenting what did happen, in a way that's compatible with the ws-chor approach for describing patterns that might happen?

bijan in IRC: danbri, I believe so. More precisely, choreography is a global perspective, that is, you have a bird's eye view of what's going on.

JacekK in IRC: is choreography deployment time and orchestration run-time, but otherwise the same?

bijan in IRC: JacekK, no

Henry Thompson in IRC: Well, what was asserted was that choreography is in principle lacking in central coordination, ever

vibrania in IRC: Choreography usually has a choreographer - and then lots of rehearsal

Stuart Williams: Are choreographies explicit?

Steve Ross-Talbot: not sure what you mean?

Stuart Williams: birds don't crash - emergent behaviour.

Ken Laskey (MITRE): seems like we are talking about timeframe of control.

DanC_NCE in IRC: nifty, davidF. [link]

Steve Ross-Talbot: SWIFT decides on a particular day that the world will change and live with the migration problems

dbooth in IRC: Isn't the "global model" defined by a document somewhere?

bijan in IRC: It seems perfectly possible (limb going on here) that from that perspective, you can describe a fairly centralized system

DanC_NCE in IRC: that was a great "what is the WG up to?" presentation. set a high bar. [bijan and danbri agree]

bijan in IRC: I sat in on the chore f2f and, really, I think it's great stuff. This was not my impression prior to this. I'm a total convert.

XForms Validator

veillard in IRC: 2 hands on the Online XForms Validator

chaalsNCE in IRC: I discovered the Xforms validator from a spanish-language accessibility list - so the message is getting around...

DanC_NCE in IRC: yes, the HTML validator grew because I was tired of answering those messages back in 1993 or 1994.

ht in IRC: Ditto for online XSV -- editors were deluged with "what's wrong with my schema" messages

tvraman in IRC: it was covered on places like OReilly News

Dan Connolly: Can I see a demo?

Micah Dubinko: written in python in libxml2. only uses GET. [shows demo]

Karl Dubost: very cool - please contact Olivier Thereaux (W3C)

Rotan Hanrahan: I raised this with HTML validation people. It should give you a score at the end of validation.

DanC_NCE in IRC: hmm... score = 1 - 1/errorQty

dom in IRC: Qty or Density?

Rotan in IRC: Too simple. Should rate errors according to severity.

olivier in IRC: number of errors is NOT a reliable factor (cascading errors)

dom in IRC: a random idea for measuring score for the validator - makes a comparison between the page as it stands and what tidy makes out of it [discussion of Webcams and video in IRC follows]

Micah Dubinko: In XForms you can use XML, but in this case I serialized to GET.

Using RDF and content morphing to promote the de-segmentation on Web content, universal design and accessibility

Lisa Seeman: An example of an annotation is relative importance.

Matt May (W3C): Annotation is a powerful mechanism. Equally important for people to still think about how they put their site together. I heard the undercurrent of "we'll get accessibility for free."

chaalsNCE in IRC: and as well as the dangerously tempting "we don't have to worry about our main site" I heard "you can give people your real Web site and allow people to read the simple language version without changing your lovely prose style."

Al in IRC: My reading of what Lisa said other places in her talk is that the annotations are done by media-critic experts and the view-recasting transforms by disability-segment experts. But the alternate-views transform processor is linked from the origin site and the original business entity retains review control over the transform results.

chaalsNCE in IRC: I think the point is that you don't get anything for free - if you use the tools they can help do something interesting, and this out-of-band method can give some flexibility beyond what ordinary user-land technology can do at the moment. I don't think there is anything that forces the user to rely on the original business' version of the transform. If someone else provides a better one then it just becomes a copyright argument. On the other hand, as Matt points out, the original content provider can't just assume "some body" will do the annotating - since quite possibly his brother "no body" will do it...

Al in IRC: in the earlier trust discussion it was clear that most consumers will prefer a representation that the merchant will vouch for.

chaalsNCE in IRC: Sure. But in a small community of users with specific preferences there might be a further optimisation done by a thrid party. There is a potential copyright issue, and the original provider is in any case smarter to make sure they offer the best access to their material...

Session Four

Adventures with Mixed Markup Language Documents

seven people in line at the floor microphone
Photo: Masayasu Ishikawa

Moderator: Debbie Dahl, Chair, Multimodal Interaction Working Group. Organizer: Glenn Adams (XFSI) Timed Text Working Group. Panelists: Mark Birbeck, Masayasu Ishikawa (W3C Team), Rhys Lewis (Volantis) Device Independence, Henry Thompson (W3C Team) plus additional TAG and Semantic Web reps

Debbie Dahl: I'm very much a learner here, I'm going to be very interested to hear what the panelists have to say... [panelists introduce themselves and their interest in the topic]

Scott McGlashan: We use VoiceXML for interactive dialogs, we do speech synthesis and speech grammar, and have to mix that with VoiceXML using XML schemas. Challenges included markup but also deployment issues. Now looking at version 3, and we want to integrate other languages with it.

Chris Lilley: We noticed lots of people were using scripting with SVG, and some other markup, with scripts to transform this into SVG for rendering. How do you do rendering, or add stuff to the DOM, without having extensions interfere with each other?

Henry Thompson: Whatever one may think about application/xxx+xml, I don't think anyone argues that it scales well to x+y+z+xml. How important is it to know early what namespaces must be understood in order to process a document? Secondly, 5 years ago, when XML Schema was chartered, one of our requirements was to support multiple namespace documents in a modular fashion. We delivered on that but we now have a bunch of experience with the range of namespace mixing that wasn't available to us when that design was chosen. The Schema Working Group needs an inventory of examples to support our design for the next generation schema language, so we can do a better job.

Mark Birbeck: My big interest is in using the DOM as a programming environment so I need to be able to bind in functionality dynamically, at runtime. E.g. DOM3 load and save, or validation modules [shows Forms Player being used as a presentation player]. One can imagine extending... building blocks... dynamically at runtime. How do you make an XPath query that spans what might be in a different parser's tree. DOM3 implementation registry... events rather than methods... so we are starting to make our specs of the events that something understands, rather than of the methods it supports. We need to distinguish editing a document and running it.

Chris in IRC: DOM3 implementation registry seen as important, event driven rather than method-driven architectures

Rhys Lewis: We try to define a language profile that authors might use to present on a wide range of devices, and communicating the device capabilities that will render those documents. How to communicate that, what should that look like? I'm here because I have a need for building mixed-markup language profiles.

Al Gilman: I have a history in building large systems like GPS. I know just enough to be dangerous. The experience of people who need the Web to be accessible has been one with mixed language resources, e.g. we have to include ECMAScript and "retire document.write()".

Chris in IRC: document.write delanda est

Al Gilman: Authors use scripting to make dynamic documents simply because they couldn't get the static markup working. Another crusade... Web content is semi-formal, composed interpretation of content m/c doesn't understand and markup the m/c does understand. A non-linear non-orthogonal factorization. We want to reinforce natural language with timing (see Timed Text Working Group), pronunciation (SSML), and other languages/tools that let us put interpretative assistance for people who need that. Can we make these concurrent aspects of how you strengthen natural language content? We need language fusion techniques to allow people to mix these things freely.

Masayasu Ishikawa (W3C): I'm here to take a few photos from the stage. In order to do that, I was told I had to join a panel. We are working on XHTML... eXtensible... designed to be modular. I'm editor of a Working Draft, An XHTML + MathML + SVG Profile. Most of this draft is about DTD (messes) to combine namespaces. I also worked on Ruby. (gives example). It's good to have a markup schema for Ruby. I implemented it in DTD, Relax, XML Schema, but none of them can fully describe this small markup language. It has only 6 elements, but you need to use multiple schema languages together to describe it! [gives example of mixed namespace HTML + SVG, + RDF + MathML + EGIX + .... [document]

mimasa in IRC: Implementing the Ruby Module.

Dan Connolly: I edited HTML 4, participated in XML Schema, XML on SemWeb Coordination Group, and the TAG. This mixed namespace stuff is TAG issue No. 13 "what's the meaning of a document of a document with mixed namespaces". There are 3 issues - mixed UI, XML functions (what goes first, [processing model]) and issue 35, RDF + HTML. I think one of the reasons W3C has a leadership position is our ability to keep in mind that our work is part of a larger whole.

Question 1: 1) Namespaces - Solution to the problem? If XML namespaces are the solution, what's the problem? Diversity of vocabulary? Ownership of vocabulary (US/UK English, you say tomato, I say tomato).

Dan Connolly: I'd say they are the start of a solution. The problem is people all over the planet trying to communicate with each other. The Web helps people do what they were trying to do before the Web came along. People do it lots of ways, if you let them name their ways you can start to communicate. And now we can see the very hard problem, the nature of the beast, communicating, is very hard, so we're of course going to get adventures.

Henry Thompson: Namespaces are the solution, but only to a very circumscribed problem. When I teach namespaces... the Recommendation solved two problems... (1) different people used the same tags for different purposes, and we needed to be able to distinguish them...

Chris in IRC: (and different attributes, too)

Henry Thompson: and (2) W3C wanted to use XML without having to lay claim to huge swaths of reserved names. Separate otherwise confusable names one from another, that's all it does. That allows you to get yourself into arbitrary amounts of trouble, which we're now trying to deal with.

Al Gilman: A market opportunity. A given dialog usually takes place in the context of a set of a given set of vocabularies. This is a way we talk to others in giving context. Namespacing is a piece...people have this need... reuse groups that are smaller, we wind up wanting to have this kind of structure in the markup.

Question 2: Content Types. What is content (media) type of a mixed language document? Does this help user agents or processors? e.g. what does application/mathml+smil+svg+xml mean? Does is make sense to associate content...

Masayasu Ishikawa: I tested how browsers treat various types [test page]. In general browsers don't distinguish application/xhtml+xml, text/xml, and application/xml. (There are differences between text/html and the others.) So media types are not very useful for this purpose in my opinion.

Dan Connolly: That is a software installation problem. Content has a media type, OS dispatches a pieces of software, whose software gets to run?

Mark Birbeck: We need them at the moment, but the current world is of the monolithic browser. There's an issue of validation... but I think that's going to be solved at some point. We now know we need more than one schema, but it's not a hard problem. But to send down a document that in advance you don't know what it does, and to process it. A mime type assumes there's a monolithic application that supports a single type. I think we've reached the end of what we can do with the monolithic browser.

sean in IRC: lets re-invent opendoc, and OLE linking (not)

Mark Birbeck: It's impossible for any browser vendor to keep up. There's no point in having an ever-increasing list of mime media types, we need to go the opposite way. Build in support on the client end for the different namespaces, a has-feature attribute or something

holstege in IRC: +1 on not assuming 1-1 mapping between document type and application

Chris Lilley: People pick something they know will give them an operating environment, e.g. DOM, event model, so they'll use a +xml type but they need to say, "I need these various pieces." E.g. if you're editing SVG in a form, you need forms but not SVG display. SMIL has a feature-test feature.

gk in IRC: From my recollection of the IETF discussion, the +XML convention was never intended to be used like this. But for any finer-grained use, separate descriptions (e.g. Content-features) should be used.

Chris Lilley: The order is significant too, in mathml+smil_svg+svg, so it's intractable. Yes, and the combinatorial explosion of concatenated types would never work.

timbl in IRC: q+ to say that it is useful to know at the top level what the outermost element namespace is, because that tells us what sort of app it is. Then the modularity and namespace mixing is very different for hypertext applications (html, svg, smil) and web services applications ro semantic web applications. (if there is time) [no time remained]

gk in IRC: TimBL, content-feature can be used to indicate top-level namespace. I had a work-in-progress to register this, but it fizzled out. But RFC2913 gives an idea.

Question 3: Composition - Internal or External. Should specific compositions be defined internally or externally? XML schemas are internal.

Michael Sperberg-McQueen: You can compose schemas dynamically, why are they said [on the slide] to be internal composition?

Al Gilman: I believe using schemas you can derive a structured type and expose it in your namespace, you can import everything and import one namespaces,but only at the schema level.

Dan Connolly: if you change/reduce the problem is can be solved in some cases; RDF has a different style of composition. Henry had said, if the elements are semantically independent, can ignore each other, and RDF defines that to the extreme. The object is that you can define namespaces independently and merge them later, if you just want to state facts and merge them look at RDF.

Masayasu Ishikawa: When you combine various markup languages, not all will be under your control. E.g. in my mixed example, we can't assume that everything in the world uses the same schema language, e.g., W3C XML Schema. So maybe you'd end up writing the schema yourself, or we need a way to combine different languages.

Norm in IRC: Namespaces for everyone

Glenn Adams: In many cases it's necessary to have to anticipate within the design of a module how external composition might reuse those parts and what kind of composition rules have to occur. Another area is the amount of knowledge of the insides of modules an outer module needs.

Question 4: Composition Semantics. Should composed language absorb constituent vocabulary. How to determine identity vs. near identity of vocabulary? And how to declare constraints on composition?

Chris Lilley: For SVG animation we were subclassing and had to copy, but for XLink we could refer, because didn't customize it. Re compositional, we had a metadata element (in SVG) and an SVG must ignore the content ("not get upset by it"). Equally we have a foreign object element - here's a rectangular region, something else is going to draw in there. So we have a hand-off at the rendering level. We also don't assume we're the document root so we can be embedded in something else. These are examples showing how to be composable.

gk in IRC: that work isn't in URI form, but with RFC 3553 it can be so.

Mark Birbeck: This is another example where you have to prose [comment missing] apart from what the problems are. E.g. suppose you have a choice of 3 SVG processors. You want to create your document so it dynamically binds in an SVG processor, and you might mind which is picked up. So we need a trigger, e.g. the namespace.

Chris Lilley: You have a choice of multiple SVG processors, one is dynamic and one is higher quality but static - requiredFeatures and DOM hasFeature helps select the right one. So there you wouldn't want to get rid of namespaces, in fact I want every feature to have its own namespaces... but this proliferation of namespaces, you don't want it when you're authoring. I need to come up with proposals, e.g. like C++ "using". We are trying to address validation, user authoring, dynamic building of documents at runtime, and there is no single solution for all of them I think.

Question 5: Possible Next Steps. Propose TAG issue(s); if you care, then volunteer more than your opinion, write a white paper... organize interested parties

Henry Thompson: We're in a much better position now to analyze the problem space, than some years back. There's more than one problem here, more than one requirement on the top of people's lists. I'd like to see a focus on tabulating distinct usage scenarios and requirements and analyzing the dimensionality on which they differ.

Dan Connolly: The big problem is interesting, but solving small parts of it looks like a bigger bang for the buck for me. Tag Issue 35, there's a session on Friday morning. The observed rate is closing one TAG issue every 6 months. But some have been around for 10 years or more, so that's speedy!

timbl in IRC: I don't think that the mixed namespace can be done in general, but can be done within an application area. Has been done [comment missing] RDF. Could be done for HTML, SVG, Math, ... hence, we split TAG issue 13 to separate application area issues.

Al Gilman: I was very interested in a conversation between Device Independence and Multimodal Interaction earlier this week. It was quite constructive. They're trying to cover a fairly diverse space of delivery contexts and have minimum number of spare parts and broken rules so they are decomposing the Web page better, and each part is better characterized. Things are happening in SVG, in MMI, that people should be aware of.

chaalsNCE in IRC: RDF gets away with it by using namespaces purely as ways of distinguishing basic names - all the different namespaces are assumed to be RDF (which isn't what happens for SVG mixed with MathML, mixed with HTML...)

Debbie Dahl: I'd like comments from the audience now.

David Orchard: One thing we have to realize about the intersection between media types, mixed namespaces, and fragment identifiers. My guess is that we've outlived a chunk of the Internet media infrastructure. With XML and these complex interacting functions, this roughly flat string for the media type just doesn't work. Say there's a resource identified, it emits things and you want to filter on that, and you don't know what state it's going to be in based on the media type. There is an example with XInclude, you don't know when a frag id applies.

Chris in IRC: fragids and secondary resources, effect of media type registration. composition of nested xml functions

David Orchard: I think talking to the TAG, and to me this is a really important architectural issue. People should hold the TAG's feet more to the fire on solving this issue.

sean in IRC: The problem as I see it is we are applying tools like schema which work at the syntax level to semantic issues

Mark Birbeck: Broader than that do everything as late as possible, so you need to know what your next processor in the chain can do.

Chris in IRC: Everything assumes it is last in the chain, though .....

Mark Birbeck: e.g. if the next process can't do XInclude, do it now, else do it later

sean in IRC: Defining and mixing context free grammars isn't so hard. But combining unknown semantics is undefined.

gk in IRC: What can a proxy do? The design of CC/PP had a story for that.

Charles McCathieNevile: I think many authors don't know when they are using which namespace, they just pick up a tool and don't see the underlying code. I think the issue of what happens in authoring is really important

Mark in IRC: On CC/PP ... I definitely agree that there could be a solution to a lot of this there.

Jacek Kopecky (Systinet): I'd like to react to the first slide, "Namespaces a problem or solution?" I don't have an answer, but the proliferation of QName symbol spaces is another similar issue

chaalsNCE in IRC: Millions of mixed namespace documents are generated daily by Microsoft Office.

Jacek Kopecky: Is this connected to the problem of namespaces? if not, should it be, and on what level? Do we need namespaces for QNames?

Dan Connolly: The TAG has looked at this... My position is that the Web works best when there's just one Web, everyone can point to the things using URIs... So the things identified by QNames can also be identified by URIs. The TAG is definitely noodling on that [Editor's draft]

chaalsNCE in IRC: RDF treats foaf:knows, dc:title, and foo:wigwamForAGoosesBridle exactly the same, which meets user expectations. Treating html:p and svg:rect the same doesn't work in a scalable way that meets user expectations.

Rigo Wenning: The Privacy Activity wants to give a service to the rest of the W3C community by inventing the P3P generic attribute to bind to an arbitrary piece of XML. But it's difficult. I joined W3C in 1999 and there was heavy discussion on namespaces, it hasn't stopped since. What does it mean if I have a p3p attribute on my element foo?

gk in IRC: Rigo asks about generic attributes (P3P generic). Is this similar to hints at a generic rdf:about?

Rigo Wenning: It depends on what you do with your XML. What we'd like to do now is, because P3P actively provides this kind of service, if you want to use P3P please do it like this. If someone uses our namespace we'd like to constrain them, if you use it, only in this-and-that context. There's no way to constrain the use of a different namespace in an XML document. Could you perhaps noodle around about this, as DanC says.

Al Gilman: You can model semantics as constraints. That's not articulated in the current architecture, which leaves us spinning our wheels. The interaction isn't orthogonal, you have to have ways to articulate what comes from the source language and what comes from the application. [gives clarification about schema import]

Noah Mendelsohn: I think there a number of overall idioms in combining things. One is "isa", putting an equation into HTML doesn't make your outer document an equation, it's "using". Then there are container formats, like MIME and SOAP. I'll carry things around for you, but they mean what they mean, you tend to process them separately. There is a trap - the main meaning is not always the outer element. E.g. a purchase order wrapped in SOAP, or SOAP that has buried in it a purchase order. In a word processor, that equation might have been delete! One can't process without the outer context.

Chris in IRC: templates like XSLT are a third type

gk in IRC: is-a, uses, containers .... (where have I seem this before ;-)

Jon Ferraiolo (Adobe): W3C has been very good at defining orthogonal document formats but hasn't figured out how to integrate them. Action happens best when there's strong vendor and consumer interest. There's a lot of interest in combining the W3C's popular document formats. A lot of mobile interest in HTML + SVG. Community interest in XForms + HTML, + SVG. The next step is to get vendors and members to work together. A workshop where people say what they are thinking. A baby step, maybe, but might be getting us going in the right direction.

Jonathan Robie: If you have something that works on a particular kind of document, it assumes that the root is that kind of document. There's no one right thing to do it. E.g. databases now define XML views of the database table but people put XML inside the database columns! If you do a query on that, there might be document nodes inside those cells. We have to fix that.

Chris in IRC: XML views of databases where records themselves contain xml ....

ht in IRC: JR's headline concern was the problem of the document node

Chris in IRC: hopefully not all escaped

Debbie Dahl: Thanks everyone, I think that session certainly raised some issues.

Session Five

Making Test Suites Work for Working Groups

Henry Thompson, Tantek Çelik, Ian Hickson and Dominique Hazaël-Massieux on stage during Session 5
Photo: Masayasu Ishikawa

Moderator: Dominique Hazaël-Massieux (W3C Team). Panelists: Ian Hickson (Opera) and Tantek Çelik (Microsoft), CSS Working Group; Henry Thompson (W3C Team) XML Schema Working Group; Jeremy Carroll (HP) OWL Working Group, Patrick Curran (Sun Microsystems) QA Working Group; Mary Brady (NIST) XML Core Working Group

Dominique Hazaël-Massieux: [Poll of people involved with test suites - large number put up hands]

CSS Test Suite Documentation [content]

Ian Hickson: Tests should be valid! We found 18 test suites were invalid.

MDubinko in IRC: +1 for validation, he says knowingly

Chris in IRC: not all tests can be valid e.g. tests that include things like RDF that have no schema

dbaron in IRC: error handling tests are the exception

veillard in IRC: right you actually want more test for errors than for correct input Q

fantasai in IRC: If you read the CSS test documentation, you'll see that failure tests are recognized and recommended.

Ian Hickson: Tests should be simple. Avoid interference from parts not being tested. Make tests easy to author; this encourages others to submit tests. Test what you have to test, not other technologies. Test all features once on their own, but also in combination

Chris in IRC: corollary - not all combinations can be tested in a finite time

Ian Hickson: Results should document themselves. It must be obvious if they fail or not.

gk in IRC: I'm wondering where specification testing leaves off and implementation testing starts. (Or are we talking verification suites here?)

Chris in IRC: I can't stop myself from thinking "testing all HTTP/1.1 features combinations" :-) Separating the inter-test navigation and the tests themselves (as the CSS Working Group did ) is also very valuable.

CSS test suites: lessons learned [slides]

Tantek Çelik: A small number of people worked on the CSS1 test suite. Implementations use tests more than they use the spec. Features with more tests tend to be more interoperable. It turned out that tests with many tests seem to be more interoperable than those with few

veillard in IRC: test coverage -> quality of final implementation, especially for "legacy" CSS1 is a very interesting feedback

Tantek Çelik: Interoperability may be limited to only what was tested. On the other hand, features may be expanded by what was tested. Tests influenced implementors to expand their implementations. An example -- the CSS1 test suite tested for white-space on inline elements, which was not allowed by the spec, so implementors implemented it. And as a result, CSS2.1 expanded the spec to allow it.

Tantek Çelik: Freeze the suite at a given moment. Completed test suite results in better interoperability because people pay attention when it's marked done. You can still tweak it afterwards anyway.

Tantek Çelik: [shows CSS Selectors test suite] New purpose - demonstrate interoperability to exit Candidate Recommendation. This suite uses a test report template. Features are rows. Columns are languages. This revealed a problem - certain features had no test. [Shows example report; colors and layout show success or failure] [result report hyperlinks to the actual tests] This gives a good overview of how well an implementation does. Beware of specs with few tests. The at a glance report is essential.

Rich in IRC: hopefully nobody in the room is color blind

fantasai in IRC: That's why there's alignment used as well

Henry Thompson: Very briefly, things your grandma would know about testing. Because our domains are so different, the kinds of tests are very different. CSS or SVG suite is going to assume that a human is involved because the output must be visibly correct

Chris in IRC: except when used for regression testing on the same implementation

Henry Thompson: It is hard to do the comparisons automatically. Other things are trivial to test automatically. In XML Schema we focussed on things that could be automated so we have >20,000 tests! No one has looked at them all, that's fine. The focus of this kind of suite is on utility for the developer, not the Working Group. Some companies forbid using their products to run test suites.

chaalsNCE in IRC: So if we had done more listening we would not be (temporarily) losing aural styling from CSS

glazou in IRC: doing automatic test report like the CSS Selectors' one is possible if scripting and CSS computed style is implemented in the tested implementation

joeclark in IRC: IIRC the speech media group will ultimately take care of "aural" media styling. This does not imply the whole concept is useful to screen-reader users.

glazou in IRC: This brings a new idea, blatantly stolen from Karl - the results of the tests for a given implementation should be called the "implementation profile" for that implementation and made public by the implementor when it claims conformance to the spec.

chaalsNCE in IRC: Joe, right. Hence "(temporarily)". And yes, without testing, we lose it for reasons that are orthogonal to whether it is useful to anyone, which seems the wrong reason...

joeclark in IRC: speech CSS may be useful for some application other than screen readers. Cf. 'reader,' but that's one of my little hobbyhorses.

Test Driven Specification Development [slides]

Jeremy Carroll: [describes the process used by the RDF Core Working Group]

chaalsNCE in IRC: There have been people suggesting that the spatial stuff in Aural CSS would be interesting for game developers.

Tantek in IRC: joeclark's 'reader' reference

jhendler in IRC: this is an example of autogenerated tests and results page (thanks to Sandro) for OWL

ht in IRC: glazou, wrt implementation profile, absolutely, that's the schema has done it from the beginning, going forward we've augmented the mechanisms to say that if two profiles differ on a test or sub-family of tests, and they both assert on inspection that they're right, thats _prima facie_ evidence that the WG needs to intervene

pdowney in IRC: applauds the notion of test driven specifications

glazou in IRC: ht - could become a mandatory criterium for all specs imho

Tantek in IRC: Michael, IMHO, no, but often tests/results have the side effect of affecting the spec through errata.

NIST work on W3C test suites

Mary Brady: XML core has >2000 tests. DOM has 3500+ derived directly from the spec. XSL-FO has 2700+ tests. For XML Schema, NIST has produced 23000+ tests. XML Query 1800 tests.

sean in IRC: To make the tests normative implies you really have a formal semantic model.

Mary Brady: You need to organize yourself to work collaboratively. How to deal with bugs, how to communicate and have a Web presence. [example of DOM test suite]. Start with atomic tests. Then it is easy to determine what is failing. Then feature interaction tests. Map tests to the spec. Test coverage is a matter of how much. You are never really done. Encourage people to submit tests. The suite is more than the tests - packaging is important, categories [XSL FO for example]. We helped base decisions on the spec. You could offer feedback if there was a problem with a test

Mary Brady: You must test the test suite.

Mary Brady: Automatic test generation works well, reduces boredom. We need ways to define interesting tests. Different formats for each Working Group means it is hard to share harnesses. If I had to do it all over again I would use a test markup language.

Unidentified speaker: 1. Test cases should have HTTP URIs. 2. Tests should be available and runnable over HTTP.

A Kinder Gentler QA Framework

Michael Rys: Are you making the tests a normative part of the spec?

Jeremy Carroll: If there is a disagreement between spec and test then it is a bug.

Michael Rys: We make a distinction between specs, test writers and implementors. I think only the spec should be normative or all components, but not just the tests!

Jeremy Carroll: I agree.

dbaron in IRC: In the case of 'white-space', nobody noticed the test was wrong for at least a few years, so the implementations were all changed to follow the test. That doesn't mean the same thing would have happened if the mistake were caught sooner. (Also, the change in question was the removal of a simplification.)

Arthur Ryman (IBM): What are the IP rights issues with test sets?

fantasai in IRC: TC, The change went through the whole approval process, through discussion within the CSS WG and publication later as a normative part of the spec

Tantek Çelik: Microsoft contributed the HTML4.01 test suite under W3C license.

Henry Thompson: There are precedents. The W3C document license, software license, or either, have been used. There are other variants. We are currently trying to get a uniform letter of grant.

Rigo Wenning: Change control remains with the Working Group. if you use the document license vs. the software license, it is like open source so change control moves out of W3C. So the software license is better since change control remains here.

Mark Birbeck: I like putting the tests in the spec. It would then change the nature of the specs. For example in XForms, select1 normally gives a drop box but actually it does lots of other things, generating events, responding to interaction etc. You need to specify how to test that. For instance in CSS, what is 'green'?

Chris Lilley: (I know what green is.) SVG differentiates between test cases, and harness. This is important so that test cases never get changed.

Mark in IRC: Can I change 'green' on my "colour blind browser" ... or would it fail these 'tests'. ;-)

Tantek Çelik: It also helps re-use.

sean in IRC: is that NTSC green or PAL green? [long IRC discussion follows]

Session Six

Can I Really Get Good Web Access Without Carrying a PC and a Big Screen?

a photo of the screen during phone demo: business man entering data next to a picture of the face of the phone showing the interface to what he is doing
Photo: Masayasu Ishikawa

Moderator: Scott McGlashan (HP) Voice Browser Working Group. Demonstrations: Dan Zucker (Access) Multimodal Interaction Working Group; Michael Johnston (AT&T) Multimodal Interaction Working Group; Ewald Anderl (Kirusa) Multimodal Interaction Working Group. Panelists: Rhys Lewis (Volantis) Device Independence Working Group; Guido Grassel (Nokia) Device Independence Working Group; Michael Johnston (AT&T) Multimodal Interaction Working Group; T.V. Raman (IBM) Multimodal Interaction Working Group


Scott McGlashan: [asks the audience who thinks it is possible to get good Web access - 1/4 of hands are raised] Go Digital program on BBC had an interesting discussion about this topic [plays an excerpt of a person predicting the death of the Web]

Rhys Lewis: Goal of the Working Group is access to a unified Web from any device in any context by anyone. CC/PP allows us to adapt content to end-user capabilities. The Device Independence Working Group started by looking into problems that need to be solved and proposed some possible add-on modules that would help in other W3C specifications to get to universal access.

Rhys Lewis: There are two ways. 1) "write once, browser anywhere" but then do we settle for a "lowest common denominator?" 2) Make a different version of every document and adapt it to all the possible devices. But that makes the authoring effort excessive. The Device Independence (DI) Working Group is thus working on affordable authoring allowing authors to choose what works best for them.

Rhys Lewis: What underpins authoring for DI? Presenting possible extensions to the familiar technologies (XHTML, XForms, SVG, CSS...). Where might the limits be? How far can you go to support all these technologies in different devices? The question is whether the abstractions are appropriate in a given modality.

Paul Burke: The Multimodal Interaction and Voice Browser Working Groups show it is possible to get access with a large set of devices through their demo. Device Independence is much harder. What is multimodal interaction (MMI)? (listing goals from MMI) Why not use a "mainstream" spec for MMI? MMI is about making other modes of I/O mainstream. [giving examples where new types of control and interaction are needed] [lists current documents in progress]

Rhys Lewis: [introduces demo using a PDA - a video of MMI vision, about a business person traveling to a meeting in France. Mr. Jones travels a lot and his multimodal companions use Text To Speech for comments. The PDA speaks about trains schedule. Tickets bought get printed on his Bluetooth printer... takes the taxi, goes to the train station, gets a stock alert in the taxi, gets an email on his phone.] Now getting the equivalent of the video in real life. Presenting the technologies, the type of features, the hardware/software involved. All the software is client-side, directly on the hardware. Does not involve server-side software. [PDA reads aloud what is shown on the screen. Demonstrator speaks his choices, and the form on the PDA gets completed] MMI is needed in this case, if you want to be able to see the seat you're going to choose. [Until then, everything in the demo could have been done through a cellphone.]

Rhys Lewis: Next demonstration is getting a train ticket, done through a smartphone. MMI is important here, because if you're in a hurry you may have not the time to use the visual interface. You would rather be doing that through voice commands. [talking interactively with a service, getting info on the small screen of the smartphone... smartphone detects printing kiosks and proposes them to the user]

Rhys Lewis: Next demo shows the next step in the journey; back in your home town, using a public Web kiosk and not using any keyboard or mouse.

Rhys Lewis: Next demo. MMI shows how voice, touch and ink can be used together [Public kiosk showing an animated face of a woman welcoming the user]... demonstrator speaks to the kiosk, asks for nearby restaurants, and gets the answers on a map... asks for a review of a given restaurant by pointing to it with a pen and asking aloud... he then writes with a pen, selects places and gets the phone numbers of the places spoken to him aloud.. he asks for zooming by writing on the screen... again, prints the information]

TV Raman: When talking about combining namespaces in XML there are 2 different levels - the syntactic level (avoid vocabulary collision) and the semantic level - how to architect a browser dealing with various XML namespaces. There is another dimension to this, the Multimodal aspect of it, input, not just output. And the processing model also. This means that you need to integrate the processing model.

Unidentified speaker: In my Working Group, we're using events to close the loop. Integrated syntax without integrated processing isn't much value. So we need to go beyond XML Schema for integration. Speaking from podium about more powerful mobile devices, bigger screens, more powerful processors. Better browsers, efforts to improve usability. Authoring content in this context means authoring with multiple output in mind. The kiosk demo shows that application authors have to realize their users are on the move. That brings new opportunities, but also new requirements. E.g. give me restaurants in my current neighborhood. This makes the apps and the platform more attractive to the mobile user on the road.

Jeff Cousins (IBM): Can you have the Web on a small device? The demos showed applications on a small device, not the Web itself, right?

Panelist replies: Yes, we can have the true Web on a small device. Better screens are all around us on mobile devices, and that makes a big difference. The 2nd component is re-rendering technologies. I can use my portable handheld to browse the Web, using this technology.

Another panelist: Efforts to represent a single-browser-targeted information is just a taster, to get people interested. The goal is to create critical mass, but the next step must be to consider the mobile device at authoring time.

TV Raman: Close the loop, to bootstrap content for the mobile.

Tantek in IRC: The reason the Web took off is because it lets you easily link to resources already out there. Think back to 1992-1993 -- the Web took off not because HTML was wonderful, but because you could link to existing Internet resources, before there was Web-appropriate content. Even before you put content on the web, you could get to content.

Ken Laskey: Further to bootstrapping this is an impressive array of capabilities, but the user is high-end business travellers. You need use cases where you can show large-market utility, e.g. repairmen. I am not sure your apps don't cover this, but your presentation needs to move away from high-end examples.

chaalsNCE in IRC: A real use case - A deaf man who manages an oyster farm for people in a different state...

Scott McGlashan: We may just not be seeing those apps yet.

Steven Pemberton: My mobile phone is already more powerful than the computer I first used to access the Web. People complain they can't browse the Web from their phone, so we do a mobile profile. Better perhaps to just wait 2 years?

Panelist reply: Device Independence is trying to address this over time. Adaptation from high-level stuff to device-targetted data streams is meant to insulate the author. Server-side adaption now, client-side later.

olivier in IRC: (by the time mobile profile is a rec phones implement the original technology)

Another panelist: Yes, we need mobile profiles for mobile browsers, but faster specs would help too. Application boot time hasn't changed as machines have gotten faster, because apps get bigger at the same rate.

Another reply: It is not either-or. We can make the existing specs run today.

Jonny Axelsson (Opera): Making phone-specific content isn't helpful. We want to make rich content, then specialize as best we can. WAI is useful -- accessible content is mobile-readable content.

TV Raman: I agree. WML is a mistake. Two webs, two universes of content was a bad design. Big screens are not the answer to all problems, my pocket won't get any bigger. Emphasize one Web for content. Then distinguish delivery media in terms of size, power etc. Don't put 8-bit floats in the spec for today's devices, because by the time the spec is agreed upon, the devices will handle 14-bit.

Jeremy Carroll: My app of choice is a teenager going to a party.

Panelist: You can use the laser printer via wifi if you can find its IP address! [applause]

Session Seven

Querying the Web

scene from the back left of the plenary room: people seated
Photo: Masayasu Ishikawa

Moderator: C.M. Sperberg-McQueen (W3C Team). Panelists: Paul Cotton (Microsoft) XML Query Working Group; Dan Connolly (W3C Team) RDF Data Access Working Group; Andy Seaborne (HP); Jonathan Robie

Michael Sperberg-McQueen: [invites audience to stand and stretch. Half do.]

Paul Cotton: I suggest all XQuery commenters to do pushups.

Querying the Web [slides]

PStickler in IRC: Virtual TriX atop an RDF datastore queryable using XQuery... ;-)

areggiori in IRC: I agree.

Paul Cotton: XQuery/XPath has 1100 Last Call comments, has any group had more?

Danny Weitzner: Patent Policy

RDF Data Access Working Group [slides]

Dan Connolly: Our first face to face meeting is in April 2004.

RDF Query

Andy Seaborne: [presents on RDF Query]. Querying published data. Putting RDF together from multiple sources. Modularize data. A simple query language.

Andy Seaborne: Slide 2 is a diagram "View as RDF" [diagram showing an RDF Application accessing an RDF Store, A relational database from an existing App, and some inferred RDF]

Andy Seaborne: Slide 3 - "RDF Query" - many things. RDF => XML/XHTML, RDF => Application, RDF => RDF

Andy Seaborne: Slide 4 - Museum Example 1 - Find all such that...

Andy Seaborne: Slide 5 - Museum Example 2 [shows results linked from slide]

Andy Seaborne: Slide 6 - RDF Data Access - Query+Protocol

Querying RDF with XQuery [slides]

ht_sof in IRC: I can't find the W3C Web site version, but here's one version of Jonathan's paper.

PStickler in IRC: However ;-) TriX provides for a nice canonical XML serialization that would be ideal for use with XQuery...

libby in IRC: libby and danbri came up with squish; based on guha's rdfdb ql

areggiori in IRC: PStickler, yes TriX is one way to do express (and query later) canonical RDF

Yoshio in IRC: I wonder how to deal with the reification?

danbri in IRC: yeah, that tradition of rdf query was all was src'd from Guha's work, originally in a QL'98 paper 'enabling inference'

areggiori in IRC: one nice thing of XQuery is the result "construction" part - not sure useful for simple strawman RDF-query

Ivan Herman: You said data access concentrates on SOAP or GET protocol. If I have loaded a bunch of data, local, using Python. Do I get anything out of the work?

Andy Seaborne: I think there is a value to W3C specs. You should be able to use the language independent of protocol.

Paul Cotton: You can use Web services on a single machine.

Steven Pemberton: On XPath and XQuery, we got flak that XForms doesn't fit on a mobile device. Are you creating a mobile profile?

Tantek in IRC: mobile profile of XPath?

gk in IRC: Yeah... I am using some similar ideas to generate an XML document format from an RDF database.

Paul Cotton: I've been asked that. There have been places where we've considered that, but no, we don't have that on our list.

Paul Cotton: Any SQL engine in-lines functions.

Jonathan Robie: To me, calling something a function, or using a BNF, is only a difference in syntax. I don't see why one is easier than the other.

Bijan Parsia (University of Maryland): The language in the XQuery functions is richer than you need.

Jonathan Robie: I don't quite understand.

Bijan Parsia: Using XPath is more complicated than it needs to be for my purposes [comment missing].

Massimo Marchiori (W3C): The biggest critique is that XQuery is so complex. I think this is skewing the real perspective. Ora Lassila said that RDF is not the syntax, it's the model. The same thing applies here. XQuery might not have the same underlying model. You should either allow direct application of XQuery, or use a language which matches the model. Looking for something simpler.

sean in IRC: like scheme?

Dan Connolly: I don't understand the complexity of XQuery to be in the syntax.

Richard Tobin: Is query practical on any large database?

Jonathan Robie: I've only tried it on toy examples.

Richard Tobin: Is there some suggestion you could use the to query "the whole Web"?

gk in IRC: Hmmm... RDF implemented naively in an RDB loses a lot of structure that SQL queries can leverage. I wonder if anyone's doing anything like the Lore ideas with RDF, taking advantage of structure in the data, where it occurs?

Jonathan Robie: I know how to do some things I can explain. We'd have to sit down and discuss your problem.

Steve Bratt: [Thanks sponsors IBM and Sun. Asks that everyone please fill out the survey [W3C Member-only link]. Thanks the meeting planners (Coralie Mercier, Amy van der Hiel, Alexandra Lavirotte), the scribes (David Booth, Dean Jackson, Liam Quin, Steven Pemberton, Dominique Hazaël-Massieux, Henry Thompson, Alan Kotok) and the program committee (Glenn Adams, Jim Hendler, Rhys Lewis, Scott McLashan, Noah Mendelsohn, Stuart Williams, Janet Daly, Daniel Dardailler, Steve Bratt).]

timbl in IRC: Thank you, RalphS, for audio.


W3C Communications Team
$Id: plenary-minutes.html,v 1.18 2004/03/25 16:29:25 slesch Exp $