Web composition related work | WAP and the Web | GM OnStar system (ubiquitous
Web composition related stuff
Aurora (IBM Almaden)
Goal is to provide task-oriented access to Web services for disabled users (e.g.
allowing the blind to use a speech interface to contact a Web service). The approach
is to construct a programmatic representation of the service, let the user interact with
that, and use WBI (IBM's transcoding-proxy framework) to translate between the
programmatic representation and actual Web transactions. PatML (an XML-based pattern
description language, presumably developed by the authors?) is used to express the rules
by which an activity on a site (e.g. choosing an element from a popup list) is translated
to the formal XML representation. The XML-based descriptions can be rendered as
simple HTML, WAP, text-to-speech, etc.
The "composition" element comes from the fact that Aurora lets the user
interact with any one of a number of semantically-equivalent sites (e.g. bid on same item
at Yahoo, Ebay, and OnSale). For each semantic domain there is an XML representation
(e.g. "bidding sites", "stock purchasing", etc.) and within a domain,
there is a ruleset per Web site. Anita Huang (main author) claims that grouping
semantically-equivalent sites into a single domain and using a particular XML DTD
per-domain has been helpful.
WebEngineering.org (TeCO, Univ. Karlsruhe)
Their desire is to make the Web look something like CHAIMS (I think!), where everything
is a component and all components export well-known behaviors via a method-call interface.
Pages can contain multiple components. Because pages today are not authored
this way (but are instead authored using simple layout-only authoring tools), authors
propose to develop tools to parse existing HTML, "extract components" from
existing pages, store the component info separately and then augment it with method
information. They do not appear to have actually developed these tools yet, and they
don't expect that the tools will be 100% automatic.
The main difference between their approach and ours is granularity: their grain of
addressable entity is the component, which can be quite small and maybe dozens on a page,
whereas our grain is the whole autonomous service. I think our approach fits better
with composition of existing services since it's easier to reason about the
behavior of the overall entity over time and track changes in the service interface than
it is to identify "components" inside the entity and track changes in individual
components over time.
Two totally distinct sets of topics: (a) technology issues (HDML vs WML, etc); (b)
overall site structure and UE vs Web-specific UE. Panelists:
- Gabriel Montenegro, Sun (pilc)
- Josh Cohen, Microsoft
- Dave Raggett, W3C/HPL
- Bruce Martin, VP Prod Dev, Phone.com
- Johan Hjelm W3C/Ericsson
- Rohit Khare, 4K Assoc (will be a Mobicom 2000 panelist too)
The panel was explicitly not intended to discuss whether it was justified
for Phone.com to do a separate forum rather than deal with IETF/W3C. The moderator
was embarrassingly apologetic for phone.com, saying that "there is a $5b company
there, and their advantage and size might have something to do with a proprietary
technical advantage" -- sick. Happily, Bruce Martin deflected this.
- Will we be stuck with a "WML ghetto" (now) or "XHTML ghetto" later?
- "XHTML is only part of the answer...need to focus on user experience...must be open
to multiple higher-level formats to enable new models including VXML, etc..."
- Use XML+XSL, etc..."It's cheap and relatively simple" to do multidevice
targeting if you've thought about your site as structure rather than content. See
paper in Mobile Access section from IBM Tokyo on guided transcoding. XHTML is a fine
presentaiton language but not a good structural one.
- The bigger problem is driving the user experience on these devices. It's folly to
believe that "if existing Web authoring tools 'supported' XML, we'd get WAP
- Content providers tend to want to optimize for the platform (typically they fight
against cross-platform standards since these reduce the predictability of presentation to
the end user). This is cultural and can't be solved only by technology, and it's
getting worse since users will have more choices for presentation rather than
- There are other UE lessons, such as effect of latency on user's perception of site
trustworthiness, that are equally important. One panelist claimed that this is a
reason to deprecate proxies--you lose control of the UE. I disagree, it depends
what the proxy does.
Things to think about:
- Do we even have a "UE language"? Maybe the interaction model
can be separated from the data model in a declarative way...this is something we should
think about in the context of composition--both involve designing one's own interaction
with a bunch of content.
- This forces site designers to ask why users would come to their site in the first place,
and design the interaction around the task.
Questions from audience
- Suppose I want to serve something simple, like directions to my office, to a phone.
What else do I need to provide? It should be trivial to use this
information as input to something else, e.g. so user can use it to get driving directions
or traffic from their favorite site, etc.
- I wanted to ask about how all this versioning stuff will fall out (HDML, WML 1.1, WML
1.2, I heard a ref to WML 2.0). Phones have long design cycles--will I replace it
all the time, re-flash it, should I now expect buggy software updates to crash it, etc.?
But I never got to ask...the panelists were too busy discussing among themselves to
take questions. (At most 2 questions were asked)
- "Where we are now" is irrelevant. "$5B valuation" is
irrelevant; witness Netscape. What matters is coming up with a long term solution
for a device-independent Web. "WAP-wide web" is provincial and arrogant.
(Murray Maloney, Muzmo Comm., organizing member of WWW-9)
Mobile Web marketing (???, PriceWaterhouseCoopers)
"Mobile internet convergence" includes WAP, iMode/jMode, etc. - anything that
gives Internet access to non-PC's
- WAP seen as a transitional technology until devices get better
- 50 kb/s GSM/GPSR on the horizon for end of this year
- Typical cell phone lifetime is 27 months; hence, 27 months after Jan 2001 (by which time
all shipping phones will be WAP-enabled), every cell phone user will be WAP-enabled
- 55% of iMode usage in Japan is games/entertainment
Ubiquitous computing related
GM Onstar system
This is a "platform" for delivering voice, data, etc. services to a driver
while driving a car. The UI consists of one or two buttons, microphone, speaker.
Current implementation: you press the button and are connected (via cellular)
to an "advisor" who is your personal assistant. There is also a big red
button that connects you to 911. The platform can tap into various car systems in at
least two ways:
- Use other car devices - radio, etc. - for I/O. This enables use of programmatic
data services in the future, in addition to human-assisted. There are also sensors
connected to various things like airbags, etc. so the system can automatically detect if
you've had an accident and call 911 for you.
- Limited control path into the car. E.g. if system detects airbags have
blown up, the remote advisor can direct your car to unlock its doors so that assistance
can get in, flash the headlights and horn to get attention, etc. I had some concerns
about security here (imagine someone figuring out how to hack this!) but I was assured
that their security was "extremely strong and extremely proprietary". Boo,
Interesting points about OnStar:
- Driver's environment dominates design choices: common simple UI's, minimal distraction
from main task of driving, etc. - a good model for cell phone services and other ubicomp.
- Currently: cell phone + GPS + integration w/vehicle systems (both sensors and
actuators!). Services provided by human advisors on call. Future: platform for
delivery of services to driver. Some services will be automated (eg personalized
Internet-based content delivery with voice triggering).
- Note, GM controls both the network and the platform! So, if you want to offer your
own service through OnStar, you have to go through them. They claimed OnStar was an
"open architecture", which seems to conflict with this. When I asked the
speaker, he capitulated and said they do intend to keep their lead as a service provider
(read: stranglehold on the network...just like the phone companies.)
This last point was ironic because Larry Lessig's keynote made exactly this comparison
between wired/cell phone companies and the Internet: in one case the network is centrally
controlled and you have to get cooperation from the operator to deploy new services, in
the other case the network is not centrally controlled and anyone can
offer a new service without anyone else's permission or cooperation.