From the Description Logics Workshop 2010, Waterloo

The 23rd International Workshop on Description Logics was held from 4-7 May at the University of Waterloo, in Canada. The full proceedings are online as one large pdf and as individual files for each paper, which contain the papers of the 29 oral presentations (including mine) and 14 posters. Unsurprisingly, the following brief report contains only a selection of the very latest research outcomes in the DL arena that passed the revue in the past 3 days.

Keynotes

Ian Horrocks’ keynote was about his quest to search for the “holy grail” and the lessons learned along the way. That is, he started his research with the problems of the GRAIL language and the too slow classification of the GALEN terminology. With much persistence and desire to solve the problems, eventually his FaCT reasoner managed to get the classification of GALEN core down from 24 hours to 400 seconds. The next steps were to extend the language and introduce optimizations to improve the performance (whereby careful study of typical inputs were crucial for successful optimization)—in an ongoing virtuous spiral. Moving on in the time line, the Semantic Web is, according to Horrocks, alike a “grand challenge” and “killer app” for DLs. Closing the presentation, OWL 2 DL finally contains all the features that GRAIL has (in particular role chaining), but the reasoners were still unable to classify GALEN (until Kazakov’s recent approach with consequence-driven reasoning that reduced it to < 10 seconds). So, while most papers that Horrocks wrote are not particularly written for (nor particularly readable according to) bio- and biomedical ontologists, they might find it nice to know that the base motivation comes from trying to solve the problems they brought in.

The keynote by Phokion Kolaitis was purely database-oriented and focused on schema mappings in the context of database integration (comprising the data federation and translation approaches) and schema evolution, which concerned a line of research originally motivated by the experiences obtained with the CLIO project. During the talk, the emphasis was on the composition and inverse operators and for the former the consequences of chaining different kinds of mappings (e.g., GAV + GAV, GAV + GLAV).
Unfortunately, I missed the keynote by Roberto Sebastiani due to the fuzzy notion of “nearby within walking distance” between the accommodation and the conference venue on the rather large and spacious campus.

Papers

The papers were grouped into sessions about theory, extensions, ontology, reasoning, EL, systems, querying, DL-Lite, OWL, and modules.

Extensions included, among others, complexity of temporal description logics in relation to temporal conceptual modelling and tractable reasoning (i.e., temporal extensions to the DL-Lite family that are the basis for the OWL 2 QL profile) [1], presented by Alessandro Artale. Other extensions, such as fuzzy, rough, and probabilistic, passed the revue in other sessions. For instance, using a probabilistic DL (that is, the option to represent defaults) for repairing TBoxes that was presented by Thomas Scharrenbach [2], approximate least common subsumer [3] by Anni-Yasmin Turham, and my paper in the ontologies section. My paper was about the feasibility of DL knowledge bases with rough concept or vague instances [4]—yes, or and not and, because there are both theoretical and practical limitations to have rough DL knowledge bases in their full glory even when we take into account only the basic aspects of rough sets. The upside is that several research lines on DL languages & tools on the interaction between ontologies and data (and the interest shown by reasoner developers, such as Volker Haarslev of RacerPro, in the experimentation) as well as other avenues, such as semantic scientific workflows, will be very useful to improve the situation so that the combination of ontologies and data can be used better for hypothesis testing to advance science at a faster pace.

Mariano Rodriguez presented a new case study of Ontology-Based Data Access in industry [5], which considers additional features of the system, such as dealing with incompleteness of the data and integrity constraints, and addressing performance issues by assessing the query structure better. Performance optimization was also a motivation for the query answering for expressive DLs by creating “islands” in the ABox [6] presented by Ralf Moeller, and for developing a scalable reasoner for OWL 2 EL and RL using Java and database technologies (MySQL), called OREL [7], presented by Sebastian Rudolph.

Two papers dealt with the topic of (ultimately) helping the modeller to figure out in the case when there is an inconsistency, why this is so. One paper dealt with the complexity of pinpointing (which is not great, as many a modeller who used Protégé 4.0-alpha) in the tractable DL-Lite [8], which was presented by Rafael Peñaloza, and the other one (presented by Matthew Horridge) was about masking the “irrelevant” parts of the justification so as to keep the explanation as short as possible [9]. Another requested feature is dealing with updates of the ontology, for which several strategies are possible, and one such approach for DL-lite ontologies [10] was presented by Dmitriy Zheleznyakov. Also modularization and extraction of sections of an ontology is a well-known request, and an empirical study was presented jointly by Chiara del Vescovo and Thomas Schneider discussing how well the algorithms work: full automated modularization does not look good from a practical perspective, and computing only some modules will be more feasible [11]. This is still fine, I think, because, generally, full modularization is not what the modelers are after anyway, but they only would want to have one or a few subsections extracted from the larger ontology. (In addition, one could use granularity to modularise a large ontology aside from letting one be guided solely by the syntactical features of the ontology.)

That’s it for this year’s DL workshop. DL’11 will be held in Barcelona (colocated with IJCAI’11).

References

[1] Alessandro Artale, Roman Kontchakov, Vladislav Ryzhikov and Michael Zakharyaschev. Temporal Conceptual Modelling with DL-Lite. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp9-19.
[2] Thomas Scharrenbach, Rolf Grütter, Bettina Waldvogel and Abraham Bernstein. Structure preserving TBox repair using defaults. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp384-395.
[3] Anni-Yasmin Turhan and Rafael Penaloza. Role-depth Bounded Least Common Subsumers by Completion for EL- and prob-EL-TBoxes. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp255-266.
[4] C. Maria Keet. On the feasibility of Description Logic knowledge bases with rough concepts and vague instances. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp314-324.
[5] Domenico Fabio Savo, Domenico Lembo, Maurizio Lenzerini, Antonella Poggi, Mariano Rodriguez-Muro, Vittorio Romagnoli, Marco Ruzzi and Gabriele Stella. Mastro at Work: Experiences on Ontology-Based Data Access. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp20-31.
[6] Sebastian Wandelt and Ralf Moeller. Distributed Island-based Query Answering for Expressive Ontologies. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp185-196.
[7] Markus Krotzsch, Anees Mehdi and Sebastian Rudolph. Orel: Database-Driven Reasoning for OWL 2 Profiles. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp114-124.
[8] Rafael Peñaloza and Baris Sertkaya. Complexity of Axiom Pinpointing in the DL-Lite Family. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp173-184.
[9] Matthew Horridge, Bijan Parsia and Ulrike Sattler. Justification Masking in OWL. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp32-42.
[10] Dmitriy Zheleznyakov, Diego Calvanese, Evgeny Kharlamov and Werner Nutt. Updating TBoxes in DL-Lite. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp102-113.
[11] Chiara Del Vescovo, Bijan Parsia, Ulrike Sattler and Thomas Schneider. The modular structure of an ontology: an empirical study. Proc. of DL’10, 4-7 May 2010, Waterloo, Canada. pp232-243.

Advertisement