The group met in Boston with the aim of advancing its draft documents to Last Call… er… it didn’t quite go that way. However, the overall news is very positive.
A problem that has dogged the development of POWDER since even before the establishment of the WCL Incubator Group appears to have been solved. Expanded upon by Dan Brickley
during the XG process, it comes down to the fact that we want to make generalised statements about lots of resources at once when RDF is predicated on triples with a single resource, identified by a URI, as its subject. It has been suggested before that OWL might provide a way forward but a clear notion of how this could help has been as elusive as any other solution. Now we think we know – and it has the air of ‘is it really that simple?’ about it.
The benefit of a gathering like TPAC
– indeed the whole point of the event – is that you can ask other people what they think. So in the corridors and restaurants of the conference hotel views were sought from the likes of Eric Prud'hommeaux
and David Booth
with Dan Brickley
being the recipient of a lot of IM traffic. But it was during the first day of the WG meeting that we were very pleased to be joined by several ‘observers’ including Max Froumentin
, Fabien Gandon
and Tim Berners-Lee
. This was just the kind of input the group needed – although it felt a bit like anaesthesia-free surgery at the time. It was the next day when we were able to analyse all the input we’d received, begin to identify aspects that require further elaboration… and to work out a new timeline.
The new model is best described using an example – that we want to assert that all resources on example.org are mobileOK.
We first define the set of all things that are mobileOK (actually, we don’t – that’s the job of the Mobile Web Best Practices Working Group
but you get the idea). Then we define an OWL class with the property restrictions that apply to the resources to be described, in this instance, resources that have the property of being available on example.org. If we now assert that the class of resources on example.org is a subclass of the class of resources that are mobileOK, OWL inference can tell us what we need to know. Recall that we always
start with a particular candidate resource in mind, say, http://www.example.org/index.html. Is that an instance of the class of resources on example.org? Yes. What do we know about it? Well… now we know it’s mobileOK.
But hang on, what about attribution? As the WG members emphasise over and over again, the absolute critical factor in POWDER is that you always know who
has made the assertion that a particular thing is true. You can’t decide whether you trust an assertion if you don’t know who’s making it. A classic job for reification!
Defining a class of things that are mobileOK is uncontentious, as is defining a class of things that are on example.org. The triple in which trust is important is the one that says that the example.org class is a sub class of the mobileOK class – so that’s the one that will be the focus of the reification data with our usual foaf:maker, dcterms:issued and wdr:validUntil properties.
In the week following the meeting, some basic testing will be carried out at NCSR Demokritos
in Athens, after which we’ll be able to put such examples online. Meanwhile all currently published documentation (i.e. anything published before November 8th 2007) must be treated with extreme caution – the forthcoming revisions are very substantial.
A few further notes:
- The group must make significant efforts to engage with, and seek input from, the OWL community.
- We need to try to make sure it is always clear when creating a Description Resource which class is a sub class of the other.
- We will be doing a lot of testing to ensure that standard OWL reasoning does always produce the desired outcome (expect a full Test Suite as part of the POWDER document set).
- Description Resources are about adding semantic data to the Web and all the normative documentation will be couched in Semantic Web terms. However, we do expect to be able to devise an XML-based approach that can be used to model simple DRs. The structure of such an XML instance will look superficially similar to the example DR in the current documentation and it will be possible to use a GRDDL engine to generate a DR from it.
- There was some discussion about ‘the protocol part’. We expect to just use rel="meta" (cf. rel="powder") on link elements and need to look at services that can return triples about whatever resource includes such a link plus a triple pointing to the full DR data.
- Tim BL in particular was concerned about the effect on processing efficiency of using things like ‘path contains’ and regular expressions to match against URIs in the Resource Grouping side of things. The group will take this into account in the light of the use cases for POWDER, however, as a first pass, we remain of the opinion that such flexibility is important, even if DR creators are encouraged to only use such features when absolutely necessary.
- The group’s timeline has clearly slipped. We’re now planning to reach Last Call by the end of January with test suites completed, LC comments answered and the Primer finalised around the end of April (so we can seek transition to Proposed Recommendation). We plan to hold at least one outreach meeting in the Last Call phase (see previous event).
- It follows that we expect to apply for a relatively small charter extension of 3 months to give us to the end of June to complete our work.