Live from ISWC 2008 in Karlsruhe

It is already the last day of ISWC’08, which had some really good papers, comments from the attendees during the sessions, and ample ambience for networking. I will discuss the keynote speeches first, then mention a few research papers, and close with a few general remarks.

Ramesh Jain gave a good keynote speech on semantic multimedia searches—or: the lack thereof and how to bridge the semantic gap between mere images and the meaning we attribute to them so that we can find the right multimedia in the sea of images, video, etc., perhaps by what he denoted as the “Event Web” as multimedia items are ‘snapshots’ of larger events that give context, and meaning, to those multimedia items. In addition to the extant ontologies, such as LSCOM, he is developing an ontology for events so as to better annotate the items and, consequently, obtain better search results. John Giannandrea’s keynote on Freebase on the other hand, can indeed be summarized by the quote from Babbage he gave: “errors using inadequate data are much less than those using no data at all”. While obviously the wisdom of the crowds and domain expert input for building knowledge bases is a laudable idea and has achieved remarkable successes toward the proverbial “80%”—but it is the remaining “20%” that is the hard part to take it from a ‘web 2’ version to a `web 3’ version of semantic searches (cf. string matching) to retrieve the right set of answers instead of a sea of links, software agent collaboration to plan your trip based on your requirements, and so forth. To take an entertaining example from another knowledge base, SNOMED CT, which is adopted in several countries: while Stefan Schulz and I were searching for suspended concepts and relations (suspended sensu [1]), we came across a congenital absence of one tooth that is a subtype (is a in SNOMED CT) of congenital absence of mouth, of jaw, and of alimentary tract… never mind that acquired absence is a body structure, and the concoction of previous known suicide attempt that throws together temporal, epistemic, and intentional notions into one concept.
The third keynote speech was by Stefan Decker from DERI, and rather provokingly about “how to save the Semantic Web?”. Based on an analysis of the successes of physics, he identified five points: (i) appealing unified message, (ii) credibility, (iii) concerted lobbying efforts, (iv) potentially transformational power, and (v) doable agenda for successes. His answers for AI in general and Semantic Web in particular are, respectively: yes-?-yes-yes-no and no-?-yes-yes-yes. In addition, his vision for the Semantic Web is to aim for a network of knowledge and collaborative problem solving and recollecting that the Semantic Web is, ultimately, for humans. However, part of the latter point was that he dismissed (well, ridiculed in a not so entertaining way) the required theoretical foundations, which annoyed quite a few people in the audience. During the break afterwards, one put forward that it is precisely because of theoretical foundations that physics continues to do well. After all, building tools on quicksand—compared to fundaments on solid ground—is not sustainable in the long run. Surely, the human and engineering components should, will, and gradually already are receiving more attention as the topics of the papers attest, be it here or ESWC and emerging workshops about them; e.g. there was a session on user interfaces and one semantic social networks. On the other hand, is the “semantic desktop” that Decker proposes really a sexy “appealing unified message”? Surely we can—and do—do more, be it to, from a end-user perspective, facilitate bioscientists in their research or focus on goals to streamline public administration and open up and enhance e-learning, to name just three sub-areas.

Of the presented papers, several were more detailed or improved versions of earlier works, such as the one about testing with probabilistic reasoner pronto using P-mathcal{SHIQ}(D) (see here), approximating RCC in OWL [4] and details about how IBM managed to have the SHER scalable reasoner for expressive ontologies (represented in the mathcal{SHIN} DL language) SHER [3] of which earlier work had been presented or discussed during OWLED’07 and this year’s introduction of Anatomylens as real application. SHER achieves scalability via summarization of the ABox and filtering. The RCC & OWL paper [4] seeks to solve the problem of performing spatio-thematic queries by approximating RCC8 (the full-blown version cannot be fully represented in OWL) and use that for consistency checking w.r.t. assertions in the ABox.

Putting data types in an ontology is from a formal ontology (and, eventually database and ontology interoperation) perspective problematic, but many developers seem to want to have them (treating an ontology as if it were a formal conceptual data model) and better than currently possible in OWL. For those who want more of it: your requests have been heard, and with data types in OWL 2, you will be able to state, e.g. geq_5 land leq_{10} , name data ranges, it redefines XSD numeric data types, rdf:text is added as well as date/time, and there will be a data type checker [6].
A nice feature that even I have used during development of the ADOLENA ontology, are the semantic explanations for the deductions (originally in SWOOP, and later also in, e.g., in Protégé 4 where after classifying, one clicks on the “?” that appears with the inconsistent and inferred classes). More precisely, Matthew Horridge presented the work on laconic and precise justifications [2], which has been nominated for the best paper award. Their work enhances the way how explanations are computed and what information about it the justifications is needed so as to give only the minimal required information for repair; put differently: toward minimizing the haystack where to find the needle to fix your ontology.

Several presentations are, or will be, made available on Video lectures.

Last, some indication of where the semantic web still has to go to, just a tiny practical example: the conference site called for tagging blog posts with iswc2008 or ISWC 2008, and if you click their link to do a Google blog search you are supposed to get a long list. But it does not. In fact, their defined Google blog search searches on iswc2008 or “iswc 2008” that does not work when I had it in the text of two posts—well, I had used ISWC’08 and that particular permutation of semantically the same thing was not in the pre-defined search term. Even after changing it on 28-10-2008, it still has not been recognized. A non-blog web search does return lots of hits. Not that I want to insist having my two-seconds fame on the ISWC website as one of the results, but something like that simply should work by now, or ought to… I will add both their desired tags this time, and let’s see what happens. UPDATE: the tagging worked, so there are just few bloggers who bother with the manual tagging, it seems…

Overall, it was an entertaining and very interesting conference, with—from a research perspective—both encouraging results and plenty of topics for further research.

[1] Artale, A., Guarino, N., and Keet, C.M. Formalising temporal constraints on part-whole relations. 11th International Conference on Principles of Knowledge Representation and Reasoning (KR’08). Gerhard Brewka, Jerome Lang (Eds.) AAAI Press. Sydney, Australia, September 16-19, 2008.
[2] M. Horridge, B. Parsia, U. Sattler. Laconic and Precise Justifications in OWL. Proc. Of ISWC’08, 28-30 Oct. 2008, Karlsruhe, Germany.
[3] Julian Dolby, Achille Fokoue, Aditya Kalyanpur, Li Ma, Edith Schonberg, Kavitha Srinivas, and Xingzhi Sun. Scalable Conjunctive Query Evaluation Over Large and Expressive Knowledge Bases. Proc. Of ISWC’08, 28-30 Oct. 2008, Karlsruhe, Germany.
[4] Rolf Grütter, Thomas Scharrenbach, and Bettina Bauer-Messmer. Improving an RCC-Derived Geospatial Approximation by OWL Axioms.
[5] Boris Motik and Ian Horrocks. OWL datatypes: design and implementation. Proc. Of ISWC’08, 28-30 Oct. 2008, Karlsruhe, Germany.