Some notes from the W3C conference on RDF and Semantic Web. The W3C has more on the semi-official page. Definitions: Q = Question; C = Comment.
Disclaimer: These notes are a rough, personal account of what took place at the meeting. They are provided for the use of those who were not able to attend but are not meant to be a full or completely accurate account of the proceedings. Many portions of the meeting are not documented or may be misconstrued. All text within is a rough paraphrasing except where in quotes (""). Corrections and additions are appreciated.
Starting off with personal introductions. Meeting run by DanBri and Ralph Swick*.
Most specs continue on after the Recommendation step -- RDF is a bit odd in this.
How do we deal with confusing things in the press? It was hard, but then we got Janet Daly. ;-)
We need ways to interface with Web services so that documents will be updated with updated status information.
Problem: Dynamic work may also go out of date.
We had some of this with the RDF spec, and we were worried not only about technology but also about the social system and the vocabulary used for it.
Anyone can publish a note (Member / Invited Expert -- don't want to be overloaded). An important option. W3C not so good at the research phase -- may be the problem with RDF Schema.
How do we deal with updating schemas? Do we transform the old data or create a second (new) ontology. Problem: Older data may not be as finely detailed as new data. Same problem with syntax (what happens when new distinctions appear?).
Eric Miller: left OCLC (long time), first hour on the job. Four years, lots have happened. No one can keep up with code (except maybe AaronSw ;-)). How do we get to the next level?
Bringing together people under the loose umbrella. Locking down work and building systems.
Power of the Semantic Web -- relational links with descriptions, things in context, distributing content in a new way.
Two parts: Finishing up the core. (Building a foundation. Richer vocabularies, richer queries built on top.) Doing advanced development. (technology influences standards influences technology. all sorts of important work being done that hasn't been done in the past.)
Members and non-members will work to make things a reality.
Brian McBride: Taking a walk through the charter. Asking for feedback. Found RDF-IG irritating. All the discussion goes nowhere -- no answers. Update M&S (possibly split), update RDF Schema (plenty of comments).
New work: explain how RDF fits into the rest of the world (especially XML Schema). Needs implementor feedback, pointing to improvements, clarifications. NOT: build a new syntax, build a new model. About fixing current problems. Must be backwards compatible, "forwards compatible" (i.e. DW*). Synergy with ISO Topic Maps, XML Protocol (serializing directed graphs).
Deliverables: Issue Tracking (current list incomplete, need to fill it in) errata page. Next step is the Call For Participation, especially implementors. Need a glossary to be nailed down. Lots of great stuff in the mail archives to be trawled for issues, mail to Brian McBride*. Volunteers should split up the archive to document the issues w/ pointers and discussion. RDF-based issue management tool would be great. ;-)
Q: Will we unify the Model and Schema namespaces? No, breaks backwards compatibility. Q: But, will we carry on the quirky parts of the specs? We need to treat the specs with resepct, yet there are goofy parts (like aboutEachPrefix). C: We've convinced people to use RDF, we don't want you to go change it. Please don't change the model! TimBL: People may have actually used aboutEachPrefix -- perhaps we should run a poll to see if people have used it. Seems unlikely to be implemented. [ Some confusion over RDFS domain and range. ] C: If we change the meaning of domain, will we change the RDF namespace. We should teach a lesson to toolsmakers who've hard-coded namespaces.
Q: Where is reification on line from quirky and non-quirky? We need to clarify where it hurts interoperability and modify that which makes RDF expensive to implement. Working Group can't discard what they don't understand. We still have lots of things to clarify (namespaces, etc.) and WG has to tackle straight on. WG gets to set precedent in this respect. Can also change local part. Eric Miller: Interesting suggestion from Sergey -- we could profile RDF and issue human compliance regulations.
Daniel Brickley*: It's important to separate model from syntax. C: You can extend the model, and add things (ID for statements, ID for model) but make extensions not replacements. Jos de Roo*: N3 is good way of doing notation, reification. I've found nothing better. Graham Klyne* Scope creep, is adding N3 spec too much? JdR: There's already something in the spec -- little example formats. GK: Oh. TimBL: We can charter another group for it.
C: When I look at RDF, there was lot of design discussion and result was simple. (which is why we have Semantic web activity) Something useful might be to revisit Semantic web in context of uses and things at a higher abstraction level. Important for RDF to provide clear model and semantics to provide information about other XML. We could define "inner" RDF core and support richer things built on that for specific purposes. Perhaps the other thing will go away. Sandro Hawke: Second the creation of a model and a "trivial syntax".
RRS: You can have multiple syntaxes. We made XML normative. We'll learn if there's more/less for the core to do and work out the boundaries. TimBL: Perhaps we can change language so it won't trip people up so much? Seeing a lot of nods. DanBri: Some of us thought we were using the work resource correctly.
Sort of new, idea is to have staff resources from members, etc. to do some implementation, experimentation beyond the charter. MIT has submitted proposal to DARPA DAML to do work with TimBL's formal logics. Little more structure than tossing things out on the IG. Trying to apply to W3C website, process, etc. -- attempting application. DAML is one part.
Part of DAML project. Work with agents (agents are our middle name). Trying to help move up the Semantic Web vector. Building on top of existing work, integrating with OIL, etc.
Has some projects built with DAML from a homework assignment.
No ability to be definitive in RDF. X has only two children. Added a LISP*-like structure with null terminator. Supports cardinality, class expressions from the OIL effort. Using information to define meaning of type properties. Extending schema to add datatypes -- coming soon, taken a lot of discussion. adding rules to DAML (working with RuleML)
not one big ontology. need of interoperability between ontologies so relationships can be specified. specified as RDF, also model-theroetic and axiomatic versions. Added properties for reasoning engines. future versions should have rules
more than just languages: work with ontologies, developing mapping between UML* and DAML. trying to add DAML to PowerPoint, Web, creating services for/with DAML and using DAML to describe the services. doing translation between ontologies; query processing; inferencing; report generation and signed XML (trust).
developed lots of tools, about 130 ontologies. running a DAML crawler. Please submit your ontologies, etc. Have ontology building tools: oiled*, protege*. Ontology analyzer: chimera*. viewers: DAML viewer*, Palm DAML*. DAML has been used to categorize talks, todo list in software project management.
Good start, only been running for six months.
Q: How many are real ontologies and how many are just demoes? Now mostly demoes, expected to shift towards real ontologies.
RDF is just a DLG model/syntax.
Lots of questions about XML, URI: What is a namespace? (RDDL makes it a collection of resources; designed to be RDF compatible but uses XLink syntax). XPointer: the fragment ID syntax for media type "text/xml". XLink maps into RDF (arcrole -> predicate; href -> object; role -> rdf:type of object). XLink is easy for Web developers.
RDDL combines human-readable, with machine-readable. use xpointer "raw name" syntax. Uses URIs to define things. RDDL uses namespace URIs as roles, which is incompatible with rdf:type. Ignoring that, XLink can be seen as an RDF syntax. XLink lets you define class of object, typedNode only allows subject.
Hard for people to deal with even namespaces. Forced them into using namespaces. Able to use plain old XML as RDF by assuming rdf:parseType="Resource".
No anonymous nodes -- there's XPointer. It's not artificial, uses XPointer and allows reconstruction of the document. Rathole: What are we naming? Can we do that. TimBL: Bug is that anonymous nodes aren't in the model. RRS: It's in the model, the parsers just do it wrong.
How do we go from qname -> URI? Just add a pound sign. Problem: What about current URIs? -- the only one with a pound sign. Add two pound signs? Looks funny. Don't? Then you don't know whether the namespace had a pound or not. Essential problem that we can't refer to concepts properly (i.e. XML schema, etc.). Needs to be done ASAP to save future implementations.
XMTP: XML (RDF?) syntax for MIME. XSet: RDF syntax for EBNF grammars (of XML, etc.). Creating RDF schemas for XML documents.
Doing work to make Web more accessible...
Lots of acessibility tests are not doable by machines (i.e. connecting text with images and finding whether the words are useful). Don't want to repeat those tests except when necessary.
Connecting inaccessible items with accessible ones. Marking sections of pages as inaccessible. Finding other sections that are related.
Meaning is important -- mapping words and phrases, marking things as the same, what does a style signify?
Syntax isn't important -- RDF becomes triples, tools translate everything out of syntax, syntax needs to be common so that it's easier to interop.
Not just about words -- accessibility isn't just for the blind.
All about tools. Most people working EARL haven't done much work with RDF so it's a good test of utility to outside world.
A language for saying things about the document.
Needed something reasonably standard. Can say what's not done and let other people fix them. Written as a conformance language.
TimBL: What about human access to machine-readable documents? All sorts of way to serialize it. SW would be great for accessibility -- but wait, we're messy with types in RDF. If we say something is the name, we forget about the conceptual title which has lingual representations. The Bible is only title: Bible in English. We use xml:lang but maybe we should find a way to allow it to be automatically referring to the concept which is part of every language. Will some simple model work now make things easier later? We may be shooting ourselves in the foot.
Eric Miller: I've spoken with library cataloggers and they want to annotate existing objects. Ability to say that X is useful for some Y. Also, want to be able to layer information on top of things to make it easier to navigate. Difference between physical manifestation (Bible in English) and the abstract manifestation (The Bible) and wanting to not duplicate work between different manifestations. Anyway, this demonstrates power of RDF.
Ron Daniel* of Interwoven*: We don't know the accessibility problem, so it's hard to model. We model things differently based on the problem. We don't know if an abstract title even exists. Beware of scope creep -- I'll say it when I see it. We don't necessarily need to change the model.
Graham Klyne*: Tim was talking about how to model, Ron was saying this might change. We have the tyranny of ontology: if you make a decision and have to stick with it then you may be cut off form the future. The Model can do it, but it may be difficult to evolve. This is an interesting problem with RDF. EM: This is a problem about everything that includes description! It's not an RDF problem. TimBL: It's a testcase.
TimBL: We need to solve this problem now. DAML found it annoying to have to use the RDF, RDFS namespaces. They wanted to create a new set of things and declare them equivalent. So, tool providers, use daml:equivalentTo so we can make these historical jumps. GK: It may be more than just DAML equivalence, it's model equivalence. When we model things in different ways we may want to demonstrate equivilence.
CM: We need a way to talk about limited equivalences. If you do X, then you fufill WAI-Y but not US code section Z. Or vice versa. C: Different rules sets to do things. GK: real-life example is incompatibility between XML Schema datatypes and IETF work.
I had to turn my graph examle into actual code. So I wrote it in N3 (note, URI may not be actually on Web). [ Short N3 description ]
So CWM (pronounced "coom"?; in Wales means a valley which is a closed-world) can merge the graphs. (demoes) CWM can also take a number of rules, and defined DAML properties, and can spit out conclusions. (demoes) I can reify this (demo) find all implications (demo). I did this because Dan Connolly dared be I couldn't make these conclusions at the AC meeting so I programmed it for the next AC meeting. EM: I love these dares!
Problems come when you turn it back into the RDF. Uses parseType="Quote" which extends "just a weeny bit". Called Notation3 because RDF M&S was the first, the RDF strawman was the second and this is the third.
I got pestered with questions by EM and TimBL and turned rather red but it all worked out OK in the end.
TimBL: Always be sure to use rdfs:comment, not XML or N3 comments -- that way people reading things through processors know how to use things properly.
Dave Beckett demoed Redland and showed how it could parse RSS and store things in a database. Before he could demonstrate much more the fire alarm went off (we suspect it was CM's fault!) and the meeting had to end as we all evacuated the building.
Tim grabbed his backpack and ran off before I could get a chance to say much to him. Everyone else broke off into discussions with each other.
Continued tomorrow...
Part of LogicError. Powered by Blogspace, an Aaron Swartz project. Email the webmaster with problems.