- From: Jim Hendler <hendler@cs.umd.edu>
- Date: Wed, 15 Jan 2003 18:39:18 -0500
- To: Jeremy Carroll <jjc@hplb.hpl.hp.com>
- Cc: webont <www-webont-wg@w3.org>
Note: although this email is meant to be informative, rather than expressing a particular opinion, it is obviously shaped by my own opinions as to what the importance of Owl Lite is. As such, please assume I am expressing personal opinion and not speaking as the chair in the following: >2: the ease of exposition argument does not show that we got the >semantics right. The view of my colleagues is that OWL Lite as the >basis for interoperability means that there should be semantic >interoperability - this means that many tool sets etc. should be >implementing OWL Lite semantics, and lots of different incomplete >OWL Lites will not deliver interoperability. Guess I don't completely agree. For example, of all the dozens of HTML tools on my machine, and the many more running on the web, only one of them (Amaya) can author, display, browse and save HTML documents. Others serve it up, display it, help me author it, automatically author it, scrape it, etc - their job is to make sure that when it goes to the browser it is correctly viewable - but that's not the same as having for each tool to prove it is a total HTML processor in some sense. I think a difference in our world views is that you see a reasoner as the critical application of OWL, where I see reasoners as just one of many OWL tools. I find interoperability grows not from having one or two dominant implementations of a whole language, but by having a tool kit of components that can each do part of the work - parsers, db backends, query tools, classifiers, crawlers, markup tools, ontology extractors, db linkers, portal tools and web mining tools are all being developed for DAML and/or OWL, and the need for interoperability between these is what has driven me in my pursuit of a common ontology language since the mid-1990s. In fact, most of the applications of DAML/OWL I've seen to date use the semantics to help in performing particular tasks, rather than as a full reasoner used as an oracle. Some of these are important to have (and the existence proof of Euler and the NI tools is important - don't get me wrong), but the power of other tools comes from other aspects of the language - here's a few examples: 1) controlled vocabulary for integrating sources - by defining a vacabulary rigorously in OWL, tools are able to maintain consistency in their use/linking to terms. This is critical in, for example, database integration where being able to link attributes between DBs is useful -- it is also important for composition of web services. See [1] a paper Tim Berners-Lee, Eric Miller and I wrote discussing the potential for the integration of applications on the web using RDF(S)/OWL -- and it is not reliant on the reasoning. (note that the extra expressivity of OWL Lite over RDFS is the primary advantage of OWL in this space -- one can state things like which properties are optional v. required, local v. global, some v. all, which is very useful in this space). 2) Portal building - projects like AKT, KAON, Horus, and many others create a virtual portal using the ontology stuff to provide an "object"-like model on which to build - the linking happens in the interaction of knowledge and database. In most of these systems, current web pages are scraped into RDF pages creating large triple stores. Finding appropriate subgraphs that respect some constraints (i.e. find me all papers that discuss a "cancer-causing" property that is of type "locus8bound") is crucial to making this work. While it might seem that a full reasoner that got it "right" in every case was necessary, remember that the stuff that was scraped is generally not 100% correct, so most of the tools deployed to date worry less about being complete than about using some heuristics to find the "best" matches against some rules. OWL is much better than RDF/S because you can use our modeling features in these heuristic checks (c.f. knowing that something has cardinality=1 in a particular field can be used to screen out instances that left that field blank). 3) markup tools - at least some of use still believe that creating markup using tools that assist the human in doing so are an important part of the Sem Web. These tools don't need complete reasoning, but having some simple reasoning can be useful. For example, we have tools which let the user search the DAML library, import the classes they find there, and create menus to make it easier to fill in the properties of those classes [3]. Our tool works behind the scene to make sure the markup that is created (i) observes the constraints the ontology places on the properties, (ii) is legal OWL Lite (if we use Lite ontologies), and (iii) that the linkages and importing between documents is maintained. We are hoping to have an entirely OWLized web site (no HTML at all) running for our lab by end of coming term - and it will use a lot of OWL tools, but doesn't need defined classes, union/intersection, and the other such which would take us out of Lite - so we hope to have markup/retrieval/browsing in an interesting and exciting way, but with little actual reasoning (and none of it complete). Note, as complete reasoners get deployed we may well use them, and I don't mean this to be critical of them - rather I'm emphasizing that we still need and use OWL Lite (and DL) without them. Sorry about the length of this email, but I want to impress on you that I'm not just trying to move to completion -- I really think that we have built a langauge that can address a lot of important problems, that Lite can have a useful role, and that will have a constituency if we can get it out the door. SHOE [4], a language which had less expressivity than OWL Lite (not including its rules, which weren't highly used), was used in many projects and the pages Jeff created continue to be widely hit/cited (type the common English word "shoe" to Google and see what comes up first) - SHOE never had a complete reasoner, and I look forward to the things I can do in OWL by having one, but I don't see the argument that the complete reasoner is the only important application w/respect to interoperability. In trying to convince your colleagues, might I suggest visiting the DAML tools page and looking at the DAML-specific tools [5] -- Murray Burke, DAML Program Manager, has asked all the people he funds to turn their tools in OWL tools, and a number of the others ones in there, not under his control, have already expressed an interest in moving to OWL (and I'm VERY glad to include Jena in that group based on your previous messages). There's a lot going on in this space - which is why I'm optimistic we don't need such a long CR. (by the way, [6], the DAML ontology for the tools, and [7], the DAML document describing the instances use only features that will translate into OWL Lite) -JH p.s. hey Jeremy, thanks - I can use some of the examples above, with screen dumps, in the implementation report :-> Dan C, I also think we should do what Mike D. did, and include an OWL version of our implementation experience - we can jumpstart it from [6]/[7]) [1] http://www.w3.org/2002/07/swint [2] http://networkinference.itcuk.net/presscenter.asp?Item=101 [3] for example the SMORE tool, http://www.mindswap.org/~aditkal/editor.shtml, which is also described in the paper at http://www.mindswap.org/papers/SMORE.pdf [4] http://www.cs.umd.edu/projects/plus/SHOE/ [5] http://www.daml.org/tools/#projectbycategory [6] http://www.daml.org/tools/tools-ont [7] http://www.daml.org/tools/tools.daml -- Professor James Hendler hendler@cs.umd.edu Director, Semantic Web and Agent Technologies 301-405-2696 Maryland Information and Network Dynamics Lab. 301-405-6707 (Fax) Univ of Maryland, College Park, MD 20742 240-731-3822 (Cell) http://www.cs.umd.edu/users/hendler
Received on Wednesday, 15 January 2003 18:39:28 UTC