The use of OWL
by Bijan Parsia
(By OWL, I mean OWL-DL, by which I mean a suitably large fragment of OWL for which there is comprehensive reasoning support.)
I am sometimes asked, even by myself, what the glaring successes of OWL are. What’s the killer app, or the essential app, or just the app its well suited for. I was just asked this last week by some IBMers up in Cambridge.
The answer is tricky. The question is not entirely complete. One thing we certainly don’t want in the answer is anything derived entirely from (posited, future) network effects. We want some intrinsic value. Since OWL is pretty durn expressive, it is tempting to point at things you can or cannot (straightforwardly; note the KEY qualifier) say in OWL. But that key qualifier is crucial.
One aspect of OWL that stymies is its open world assumption. The OWA entails that you have to say a lot more explicitly in order to nail down certain conclusions. For example, in an OWL knowledge base (KB) where I know that Mary is either tall or short, I don’t necessarily know which. Similarly, I might know that she has three children, but there may be no “Mary parentOf xxxx” statements in the KB. Worse, a KB with the fact that Mary has (only) three children and that Mary is the parent of Monique, Carol, Esvandiar, and Megan will not be contradictory, since, for all we know, Monique and Megan are the very same child.
So, validation of this sort is definitely not a strong point of OWL. If that’s what you want, you are just in the wrong place (though, with certain extensions we can achieve much in this line).
What is OWL (and Description Logics, or similar formalisms) good for? Well, they are fairly good at representing and reasoning with partial information. Configuration is a classic application, and anything that can be reduced to a configuration problem. SAT like problems, since OWL includes propositional logic. For representing and reasoning about other (less expressive) formalisms. For example, we were able to reduce WS-Policy to OWL (since it’s just propositional logic). The advantage here wasn’t just clarifying the semantics or generating a WS-Policy checker. There are a whole suite of services that had not occured to policy folks that were obvious on the OWL side: policy containment, equivalence, disjointness and incoherence. Just being able to state more general policies is helpful, especially when the reasoner can find which other policies specialize these more general ones. (Other examples of such reductions are UML, ER diagrams, relational schemas and queries, XPath queries, and XML (though XML required some extensions).
One reason that DLs are popular in bioinformatics is that 1) the information is often structural (configuration!) and 2) one can categorize structures based on partial information. So, if you have this sort of problem, OWL might be a big win for you.
(I mention torture again! Be upset about it!) I was just reading about executions of homosexuals in Iran. (I am of Iranian descent.) Amnesty has a map of the worldwide status of sexual minority rights. It’s sad to have to hope and look forward to no part of this map being black, and to know that it is unlikely to be so in the near term.
March 20th, 2006 at 8:39 pm
Hi,
I posted the following message to the Semantic Mediawiki list, about something I called “Triple+”: http://sourceforge.net/mailarchive/message.php?msg_id=15160740
I’ve seen a RDF proposal called “Quads” (http://robustai.net/sailor/grammar/Quads.html), but that still seems too structured for common use.
I’m no semweb expert, but understand most of the underlying concepts of it, but still feel the cost/benefit ratio of implementing and maintaining a semweb application is too high currently. My feeling is things need to be simplified, and be made more sloppy (the “be liberal in what you accept” adage).
Whats your opinion on this?
Thanks.
March 21st, 2006 at 2:48 pm
Some comments and a question: Why do you say we don’t want an answer derived from network effect? It seems like an answer that aims for something intrinsic about OWL is bound to end up somewhere about OWL’s expressive power (like your examples of it being suitable for configuration, bioinformatics, etc.) but I have two concerns about this: (i) expressive power doesn’t answer what are good (let alone killer) applications are—lots of stuff can be said in OWL, but what is it good for once it has been said? I don’t think “representing and reasoning with partial information” is an application, and (ii) there are more expressive and still tractable (well, for values of ‘tractable’ that somebody who is willing to accept OWL as ‘tractable’ will endorse) logics than OWL – but people in semweb still seem to think that this does not hurt the value of OWL, so it seems to be something other than its expressive power that makes people endorse it. So about “representing and reasoning with partial information” – even if OWL is very good at that, it doesn’t seem enough to distinguish it from other formalisms.
About network effect—would you reject the following:
(NE) “OWL was designed to be expressive, which is a plus. Its real advantage comes from reuse of concepts on the scale of the web and linking. [insert other usual network effect-y things]. ”
I think the above will fail—i.e. network effect will not lead to these happy things—but I think that distinguishing OWL by picking on the fact that it’s a “web logic” (and hence bound for network effects) is more accurate than trying to distinguish OWL by some of its logical features (expressivity, syntax [Oh god, not syntax!], etc.) which are not in any sense very new or very interesting, imo.
March 21st, 2006 at 6:12 pm
In relpy to Yarden: I say I don’t want an answer derived from the network effect because we don’t (largely) have a suitable network effect, and it’s not clear that without first order benefits that we’re right to encourage adoption based on second order benefits. At least, it’s ungrounded, especially in the presence of other things that already have the second order benefits, nigh overwhelmingly.
(Of course, there are relative arguments. If someone wanted to work with SHIF ontologies, I would not recommend that they use KRSS syntax instead of OWL RDF/XML syntax…the latter has advantages solely from it being a standard even though, in many ways it’s a technical PITA.)
I don’t think I said that “representing and reasoning with partial information” is the application, just that (for certain ranges) that’s what OWL does well, so any app (e.g., configuration) which requires that is a good candidate for OWL.
People regularly ask for more expressive power (e.g., OWL 1.1)...so I don’t see your point. Obviously, if you can’t represent something in OWL due to some other limitation (e.g., you need role value maps) then OWL doesn’t help you, but I don’t think I suggested otehrwise.
Also, I’m not trying to distinguish OWL from other similar logics, but OWL from e.g., “webized Datalog” or XML. Where does the particularly expressivity help? What is it good for, in general? Of course, I appeal to traditional DL applications, because those are known. We can see how something like OWL would help with them.
I’m not trying to speculate…I’m trying to help people acquire a concrete understanding of what you could sensibly use OWL for.
March 25th, 2006 at 3:02 pm
[...] In a prior post, I tried to articulate some of the parameters of when OWL is a good fit for an application. Browsing around, I re-stumbled upon this old paper Living with CLASSIC: When and How to Use a KL-ONE-Like Language. OWL is, indeed, a KL-ONE-like language and a lot of the points in this paper are still applicable. [...]