Semantic Web Technologies Workshop 22-23 November 2000, Luxembourg Report by Lynda and FrankN. [[This should be made into an HTML file.]] Firstly, those who went to the Dagstuhl seminar didn't miss all that much. The same friendly faces were there, for the most part. See http://www.corids.lu/ka3/iaf/home.html for information on IST Key Action III, They are interested in comments on the wording of the call as to whether it conveys the meaning that we think it should. They are also interested in 5-10 line scenarios on what "the semantic web" could look like. The main URL for all the slides of the talks can be found at (see mail from Stork). The agenda for the day can be found at http://www.cwi.nl/~media/trip-reports/semweb00/SW_WSAgenda.pdf The one-line summary of the workshop is that OIL has made it in Europe.... The following notes are about the talks that were interesting. Rudi Studer, University of Karlsruhe (Rudi is Dieter Fensel's old boss and is part of Ontoknowledge.) Rudi gave a nice talk, and his slides were nice. [[Copy his own summary from somewhere and link to his talk.]] Rudi chatted to me a couple of times and is interested in future collaboration, basically bringing semantics and multimedia closer together. He has already been talking to Wolfgang Klas (an ex GMD-IPSI'r now at University Vienna). Ian Horrocks gave a very nice introduction to OIL. (Dieter had given the context.) It was just the right level of what does OIL do for you that you want it to do, how does it differ from, e.g. description logics. I think it would make a good reading club piece. Perhaps in conjunction with an OIL paper, or even a talk from FrankH. Frames+DDL+WWW=OIL My question was (Frank's too...) why build OIL on top of RDF and not on top of XML schema? The answer was what I expected, but still don't quite understand, is that RDF supports an inheritance mechanism that is what you would want when building ontologies, but that XML schema does not have this. Another question was whether OIL is going to modularise extensions. Since OIL lite is a consensus of what goes in the core. [[Link to Ian's slides.]] Carole Goble's talk is closely related to what we are doing - although she misses out rhetorics (or some means of communication), space, time and links. Also there are no media (other than text). But her medical application was nice (COHSE). [[check out slides!]] Martin Bryan "Defining and managing semantics and datatypes - is there a role for ISO 13250 Topic Maps?" I'm not sure that I understood what the Topic Maps were, but it was something to do with addressing the problem of combining ontology fragments. http://www.diffuse.org [[link to slides]] R.A. Poell, TNO Notions on distributed knowledge servers. Inference rules are notions, rules can be collected in rule sets, these are also notions. One cute idea was that you could collect data realtime and add it into the "knowledge web", e.g. history of train times, traffic jams. Notes from discussion. One participant (industrial) learned lots about how ontologies should be linked up and put into web, but missed the "reasoning in context". This concerned him. Day 2 Wolfgang Klas "Describing Multimedia" Wolfgang gave a really nice talk, which was (almost) a high-level introdution to mine. It was certainly highly complementary, and I referred to it a number of times from my talk. He is all for multimedia as documents, but I'm not sure if he actually goes so far as SMIL. They have a ZYX document model/language. They have a large medical (heart surgery) application of annotated items which they use to fill predefined templates from a database. He can create compositions of items and save the compositions back into the database. (Sounds like a real hypermedia database to me.) He also mentioned exactly the same types of adaptivity that we mention: user preferences, device dependence and network resource. (Although these appeared through the talk as examples, rather than being stated systematically.) He managed to persuade the medical people that it was worth their while spending time annotating the media elements going into their system, and that Macromedia Director was _not_ a suitable way of archiving material. (This relates to some extent to the notes we made a while ago on creating a multimedia maintenance environment - blue note XXX???) He also stressed the need for integrating multimedia and annotation tools to allow the users (the domain experts) to annotate their material relatively easily (c.f. Marc Davis's annotation tools). E.g. he wants to annotate different parts of a video file with overlapping annotations and without breaking up the file. (Nothing new to multimedia here of course...) (He was saying things we have said in the ACM MM93 structured MM authoring, my thesis and Jacco's thesis. But it is very nice that someone else is now saying these things.) They had a paper in ACM MM 99, which we should definitely read: Boll, Klas and Wandel. He had a slide of placing SMIL, HyTime and other things in a 2D display - semantics and multimedia. ZyX was more semantic and more multimedia than SMIL. [[I need to draw a cute box diagram - no I need to refer to a slide in his talk.]] He, and or Susanne Boll, would actually be a very interesting partner in some future collaboration. He summarised with familiar things such as: presentation neutral composition (what we call presentation-independent); composition of media fragments and ability to adapt. Need for methods and tools. Integrated metadata & management. Specification of description schemas for MM presentations Specification of classification scheme system. And then we had my talk. Which went very well. It went down well, the only criticism was Frank who reminded me that MPEG2 is _not_ a data format. However, Martin Bryan did ask why we weren't using groves to link into the media types. (I think he meant we create a grove for every data type we think we want to address.) However, George Stork, my host, told me twice how much he liked the talk and that it was just what he had expected. I had a number of other positive reactions. I was a bit concerned, since Phillip Hoschka came along and wanted to know why we don't just use SMIL as the "multimedia schema" that I said that we wanted. I tried explaining to Philip that SMIL contains no agreed-upon semantics, other than the DTD, and that is only structural syntax. Only something like RDF or OIL had agreed upon semantics. Perhaps it all boils down to the "semantics" of the different levels, and what the processor at that level thinks it is getting. E.g. a SMIL browser has the SMIL semantics "built in", whereas it doesn't have built-in semantics, in the same sense that RDF has... (Does Phillip understand the Semantic Web initiative? Given his W3C position, shouldn't he??) If the Amuled project ever happens, then I should go back and look at the talks of Peter Fankhauser (Sivia Holfelder), Luca Botturi. Bert Bos gave a demo and talk about his RDF photo annotation system. Looks like something we could play with. It is based on Dublin Core and is available (along with Bert's holiday photos :-) An interesting point is that he wasn't using PNG but JPG to store the images. This is most probably JPG2000, since it allows you to store the annotation within the media item itself. Zdenel Mikovec mentioned using XSL as part of their system for pictorial contents retrieval. A potentially interesting idea was put forward by Atanas Kiryakov on actually trying to link together all the existing upper-level ontologies (Cyc, SENSUS, WorbNet et..). A nice idea - they are looking for project money. Dave De Roure, Southampton, gave a nice talk about agents. It is still not clear to me when a computer program becomes an agent, but he did show a file (see slide XXX) with directories with ontologies for different purposes, e.g. finding agents, how they communicate with each other and hyperlink services. The last invited talk (there were 6 invited talks, the rest were 5 - 15 minute talks) was about business opportunities. OK talk, not much to note. He did have a plea for fragment identifiers for media/data types. He didn't mention C2C (consumer to consumer) e.g. Napster. He expected that the EU should devote more money to annotation tools, since essential for B2B as well as B2C. From industry, stronger move towards standardization (i.e. don't define your _own_ XML DTD each time). Peter asked about file systems. On slide 400 different file systems. MPEG7 already fighting about binary format. (We then left to find a train, via the chocolate shops...) ----- People Frank talked to: Martin Bryan: Talked to him after Lynda's presentation - which he did like a lot. The main aspect of our conversation was MPEG-7 related. He said that he is glad that MPEG-7 decided for XML Schema instead of RDF (which is interesting keeping in mind that the whole WS more or less decided that RDF is the way to go for the semantic web). He does not think that RDF is adequate for describing the complex set of relationships that exist within a multimedia presentation. However, he does think it is important that such data be integratable with the web. Therefore, he suggested early on in the MPEG-7 process that any descriptions should be recorded in XML. The use of XML Schema is required because there are complex datatyping constraints that need to apply to MPEG-7 that cannot be handled adequately using XML DTDs. Whilst these constraints could have been handled using namespaced attributes they are better handled by schemas. He then mentioned that he was somewhat concerned when MPEG-7 started to develop its own language, but when they moved this to XML Schemas, he was quite pleased to see that Schemas seemed to be able to meet the needs of something as complex as MPEG-7 (a real test for it - he shouted). He then mentioned that he suggested in a CEN Workshop Agreement that he and others prepared for the CEN/ISSS Electronic Commerce Workshop as part of the working group in Defining and Managing Semantics and Datatypes, that the datatypes for vector and matrix in MPEG-7 are really good and that one should have a closer look at their solution - it would be great adopting it. This CWA went to the ebXML group as part of the input to the Core Components process. Phillip Hoschka: We talked a bit about the ideas of A4SM - which he liked. He could not believe that up to now the semi-automated annotation of multimedia material during the production is not state of the art (as in digital annotation). Though we had a bit problem when he mentioned that SMIL could be a good tool for doing so - at the end it seems he understood that SMIL is not concerned with the intrinsic content of the media items it synchronizes - though I was not sure if he really agreed on that. David Schweinsberg He works for a company called Steady State Software Ltd. One of the things they are working on is an environment for film annotation during the production (not that they are very far with it, as far as I understood they are currently concerned with the post-production process). However, they are very interested in ways of including the production (beyond the AVID - Ikegami approach) and what they really want is something for including the pre-production information into the annotational process. I pointed their attention to the A4SM site at GMD-IPSI. ----- Action items Lynda tell George Stork that here wasn't enough time to talk to other participants. People to visit: Wolfgang Klas, Ricardo Pell??, Marcus Herzog - Technical University Vienna WEB databases/presentation, Paolo Petta - Austrian AI centre Vienna (all sorts of intelligent HCI) - Vienna. Things to think about - the #100,000.- Information Grid in the UK. There is also a global grid, and a meeting in March in Amsterdam. Joost to contact guy from ILOG about constraint-solving system. If we are doing stuff that is interesting for them they will give us their constraint system to play with. Chat to Kryzstof(?) first before contacting them. He is interested in applications of consraint solving, and in finding out in what areas their constraint solver is insufficient. They have their own backtracking mechanism. http://www.ilog.com Christian de saints Marie Research Projects Manager, csma@ilog.fr