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.
Doing Web metadata since '95, building software that matches the generality of RDF. Redland "contains the infrastructure" needed for RDF. Not having to do syntax work "was a pleasure, I can tell you that".
Repeat of RSS demo, used Aaron's feed.
Big deal: crawls web site with links and extracts RDF data and categorizes, status. Uses "closeness" to the catalog. Also uses cataloging from other sources, provided in RDF.
Introducing a new query and inference language. Needed a language in/for RDF. Most current approaches has built-in semantics and no support for plain RDF Models -- just look as one big heap. Want to separate into separate overlapping subgraphs, perhaps based on sources.
What does triple do? Native support for: namespaces, models, reification, etc. Give it an expression:
http://...123[age->"32;spouse->http://...256[age->"31"]]. Reification, similarly (wrap with angle brackets) and it stores in the DB.
Can do namespace declarations, abbreviations, create a model block, etc. Direct mapping to Horn Logic. Yesterday, we realized it's very much like N3.
TimBL: yeah, this is great... but if you state a fact in a model, does it store it in the model or does it store the reification of the existence of the statement in model1. Umm, yes.
Q: Can you store existential facts? If they have a URI.
Ontology editing and knowledge acquisition tool extended for RDF. Very close to RDF schema. Easy to create classes, subclasses, instances. Can output as rdfs, rdf, and UML-like graphs. Can do queries that return complex expressions w/ conjunction, disjunction. System can deduce things and decide bottom concepts. Primary focus is extensibility.
TimBL: Does it still say everything is a subclassof Thing? No, it uses rdfs:Resource. Sergey: Why is this a problem? TimBL: Because it adds no information! Speaker: Perhaps in a future version. TimBL: Why do I need it? Sergey: It's stating that we do use RDF Schema, specifically. TimBL: It bothers me that these systems put out all this unnecessary output. There are triples "flying around" that we don't need.
Jonathan Borden: Wouldn't it be good to codify that any literal can be a resource -- rdf:Literal rdfs:subClassOf rdfs:Resource; then a daml:Thing daml:equivalentTo rdfs:Resource. TimBL: My assumption is that the literal is a short and for data: -- a convenient shorthand.
Squish is a strawman syntax for RDF. A lot like Guha's RDFdb language. Basic idea is to keep things simple so that RDF could be accessible to normal people with scripting and web pages. Testing a "mass-market thing". Does straightforward subgraph matching. Takes a bunch of clauses and uses question marks for variables. Limited support for testing. (i.e. greater than, less than) (demoes)
Built on top of JDBC. Can be used with any existing interface. Can be very inefficient because it only needs triples-matching.
Does an interesting demo of a calendar app -- very interested in calendar apps. Also showed schemarama a system where queries specify existence constraints.
Showed RDFWeb. Conclusion: very easy to hack up things with Squish. JDBC interface makes it real simple. This implementation isn't so good but it could be.
Bob Schloss* of IBM*: Are there facilities to deal with bags, etc. in Squish. A: No, there aren't any good ways to get stuff. Same thing's true of reification. Q: Does it know about sub-properties. A: No. Q: Anonymous nodes? A: No. Q: It makes up names for anonymous nodes. A: Yup.
Very similar to Redland*. Trying to tackle problems of DCMI. See element sets and qualifiers in many languages, etc. Adding domain-specific semantics. Wanted to specify audience. Stored in "registry".
demoes We can do queries. Highlights results of full-text queries. Can do queries by subject and object. Working with several groups in European Union* which is defining vocabularies for their projects and wanted ways to graph back to the Dublin Core. Making registries so that this kind of material can be annotated (officially and unofficially) which can then be filtered.
Trying to find ways for dictionaries to coordinate -- refresh, reroute, etc. A way to have Inter-process communication between the triple stores.
Bob Schloss: Is it required that it is an RDF Schema to go in there? A: Yes, that's what the projects have defined as the schema format. C: You can store other facts but they won't be displayed on the site. A: We expected several skins to be placed, so we used XSLT for formatting so you can choose the formats. Q: Is the XSLT on the RDF or some XML. A: It's on RDF.
It also powers the search engine on the Dublin Core site.
Uses modularized DTDs to represent rule communities. Some of the rules work like RDF triples. Ecommerce wants to formalize English-written business rules so semi-formal specifications are allowed. More complex constructions can contain formalized rules. It can all be converted to RDF 1.0 syntax.
Born of dissatisfaction with SiRPAC syntax. Added more resource-centric model instead of statement-centric. Can add properties to resources. Treats a statement as a resource but maintains difference between statement and reified statements.
Q: It converts integers to integers, does it do the same with XML? A: No, it treats them as literals. [ confusion about whether this is in the spec, delayed until later test-cases ]
Everyone knows websites are tree graphs, but now we have RDF graphs. How do you save websites as RDF? Created WWWPAL system to work with the Web graphs. It will all be open-source. Also has an XML format (XGGML*), which XSLT can convert to HTML reports. Also can be combined with log analysis (LogML*).
Database stores atoms, statements, attributions. The server is a small shim over the RDF database. Annotations server provides the home for annotations, since it doesn't exist on the Web and client won't exist forever. Annotations server looks through the data for an annotation and gives it a local URI where the anonymous nodes are. Someone then can do a query for a URI and get back the annotations.
Interesting Blue-Sky conversation with Eric Miller. Resolved that the goals for SWAG* are essential to the Semantic Web and that the connections between different examples made for wonderful demoes.
A system similar to RDF work, but slightly different. Calls a "model" a "corpus". Uses triples notation, but seems to do it in a different way. All nodes have no global significance until they are associated with a higher-level system. They all live within the scope of the file. The system can do processing of these statements, but does things in the style more of C, with rules returning symbols, instead of the more declarative XML style.
Warning: All these URIs may change.
Redland started off with SiRPAC, needed a C one so used libwww, anyway used tests on the W3C site to see whether the triples worked. Lots of problems, lots of errors -- don't even know what the correct result is. Terrible interoperability. Results and expected values combined from all over the Web.
Lots of problems with large files, performance. Many leaked memory -- lots of performance problems. not a performance test.
TimBL: Do all these parsers have serializers? If they do, one of the tests would be to have the parsers iterate on their own output.
At Extreme Markup Languages there was a connection between XTM and RDF. Had three or four teleconferences about the connection but this is the next step. Subject in TM is a notion -- a thing -- not necessarily a resource. C: It is a resource. A: I was trying to push your button. DanBri: Is topic a subject? A: No, a topic is the computer representation of a subject -- a surrogate.
A topic map is a collection of subjects which can have semantics (topics?) on top to qualify them. Topics can exist inside a document or inside a computer structure. [ mass confusion on whether a topic is a resource ] Topic maps have associations (like statements) and scopes (models perhaps?) which are special topics which connect to other topics.
Topic is an instance of either a topic or another subject. A topic can only have one subject. A topic is the type of the topic which it is the instance of. Very few semantics in the standard (except for types, which snuck in because of user requests).
Wanted to stress the point (which they believed has not yet been made in RDF: that an information resource (like a document) can indicate its subject, or can be a subject. [ some confusion on this topic; Aaron thinks he understands -- seemed difference between data: URIs as opposed to documents which represent abstract resources ]
Subject Identity Point is a resource that either constitutes or indicates a subject. Topics that share indicators share subjects. A subject is considered "published" if it is available for public use (ISBN, UDDI, part numbers). Previously called "public" resources. Topic can have zero, one, many base names, base names can be explicitly scoped, base name can give rise to variants. Scopes become namespaces for topics. Two topics in same scope are the same. Topics don't have scope -- only topic characteristics (like name).
Ways to merge: subject-based merging - multiple topics with same subject; name-based merging - multiple topics with same name in same scope (can be fine-tuned).
Questions on what syntax we should use. Seemed general agreement that model should be removed. (TimBL: Logicians have exploded twice.) Some thought that the RDF spec should provide translation from logic to RDF. Others said don't use logic because there's no logic in RDF. Some said both, but keep them separate.
TimBL: The RDF spec delgates to the URI spec which then delegates to all sorts of other specs (went through URI delegation). So when you use a resource, you should remember this and use things which work with that type. HTTP GET should actually be more like RENDER because you get back a version of the document -- an enity -- a bunch of bits and a MIME type. After you have all this, then you can use the fragment identifier. It's at this point the RDF spec can tell you what the resource is.
Bob Schloss*: [ really good way to make point ] We need to be clear about what our spec is saying, and factor each spec as far as possible and each spec should use the terms that will make sense to most people. [ cheers from the crowd ]
Eric Miller: The names were hard to decide on. Lets keep the names but provide mappings to other people's ideas.
Sergey Melnik*: You can have a world with many identifiers for the same thing and still have a mechanism that for a query (head of a project at MIT, invented the WWW, etc.) and still get a result on the Semantic Web through something like a join (Tim Berners-Lee's email address) and getting the most probable match. I have some examples that scraped from the RDF websites and used different IDs while still getting the same thing.
Dave Beckett: It's been interesting, I'll try my best to do prototypes/hacks/better-than-hacks/tools but I can't commit.
McBride: Very interesting, I think i'm into topic maps
Klyne: I thought we might be more accomplished, but this was interesting.
Tod Matola: interesting
Libby Miller: great fun, need to make usable for mass audiences
Aaron Swartz: we need to take it farther and keep it going
Frank Manola: if we continue this and they coalesce into activites we're off to a great start
C: Pick a place that's not so cold.
Sergey: I'm continuing working on killer apps and i hope that when the spec is on the web I won't feel ashamed to refer to it. (C: that's optimistic)
DanC: I've been having fun with XSLT and Python, he has tools, and looks forward to Palm->Web connection, been catching up with formal systems, time for more outreach, but that prospect scares me.
Michel (Topic Map Guy): This meeting is interesting and I have much more to learn and hope we can sync with RDF.
Other TM guy: we're all groping together, and this is a remarkable moment. no, I don't mean it quite that way! I find calendaring very interesting, and that's part of how HyTime swtarted. in topic maps, we want to know how to say scope of time. We've been punting on that so far thinking HyTime at last.
C: I came as a complete newbie, and I came back (and that says something[laughs]) but it brought a gleam to my eyes.
C (RuleML guy): I think one of the most important issues is how to evolve RDF metadata. before I start to do my own documents, I need security and without it I won't start. if i have a way to evolve, i'd start right now. i think there are ways to learn a lot for RuleML, etc.
Mossimo(?): keep things small and clear, don't try to do everything and nothing at the same time. keep a heartbeat. provide usecases -- challenges and be specific so people can build them and see our status.
C (ben grosso?): I echo the RuleML statement and I still think that we should have an RDF core that's humble and well-defined rather than the original Semantic Web visions take over everything.
Mike Dean (DAML): I liked it
Sandro: It's nice to be part of the community with other people, these are difficult problems but smart people. How do we share things with the open-source community and others? How do we feed the fire?
Danny Weitzman* of W3C's Technology and Society domain*
Art Barstow: Let's congratulate Dan. [ standing ovation, Dan turns bright red] now maybe he'll give me that money he owes me.
Marita ??? Anotea: it was great
Borden: thanks, i've learned a lot, let's go forward
Rick Jeliffe: Thanks to the W3C for getting me out here, i'll be updating my tools, i think that rdf development needs to become more abstract and not get distracted by the syntax first.
Jos de Roo: I'm glad to see the faces and direction behind them. i hope it continues.
Bob Schloss: I love connecting faces and addresses, seeing demoes, i feel we need to integrate with XML framework. and i hope we will move to the education and outreach stage. i've been approached by book publishers and i don't have time -- does anyone want to? (DanBri: at least they approached you -- i had a book published with my name on the front and i was never asked. [laughs])
Pudik: formal grammar, collaboration
DanBri: if people had the interesting hallway conversations i had it will be great, we need these projects to get going.
TimBL: it's great to see how it's taken off and how we're working frantically. i find it difficult to figure out what i should be thinking about -- how high and how low. should i be coding? working on an application? writing an article? what level should I do it at (ontology, inference, etc.)? low-hanging fruit or something really hard? I'm not comitting myself [laughs] i need the isAPhotoOf property [laughs, danc jumps up to do it ]. it's so exciting and the first thing that has me writing code again in ten years.
EricP: i've been hearing voices of doom as to whether we can take off. it seems the climate has changed so that our goals are universal. lots of people have similar goals. it's hard to stop it at this point.
Eric Miller: having been part of rdf/metadata for so long, we're at a powerful inflection point. about to do something really exciting. seeing the dialogue and quality of people with resources (people) is incredibly encouraging. you've made my second day at work! [laughs] I'm so excited about working with you and moving the web to the future. everyone take a bow.
TimBL: when it comes to taking bows, there was a great sigh of relief when eric came to relieve Ralph who was doing a lot of work and administration and research and metadata work when he wants to do collaboratioin work [lots of applause]
RRS: I should have given eric the last word. i always forget how invigorating this, and while i wish we had the tech, we don't. the problem is trying to figure out what we learned. on the verge of important thresdhold. took longer than i thought but at the point. People doing things we didn't expect is soon, sort of starting. looking forward to it.
More is continued tomorrow.
Part of LogicError. Powered by Blogspace, an Aaron Swartz project. Email the webmaster with problems.