You are here

Dr. Stefan Schlobach

Assistant Professor Knowledge Representation and Reasoning group, Artificial Intelligence.

Since October 2004, I am an working as a Universitair Docent (Assistant Professor) at the Knowledge Representation and Reasoning group at the Afdeling Artificial Intelligence of the Vrije Universiteit Amsterdam. Before, I was a member of the Information and Language Processing Systemsgroup at the Universteit van Amsterdam, working in the Pioneer project Computation with Meaning.

Before I joined ILPS I was a research associate in the Department of Computer Science at King's College London working in the field of Knowledge Representation using Description Logics. In the Group of Logics and Computation we implemented a Description Logics based Reasoning System, called Wellington.

My own research included the integration of a Knowledge Discovery component into Wellington and the application of different learning and data mining mechanisms in this framework. If you are interested, please do not hesitate contacting me. I did my PhD at the University of London, adapting techniques from rough set theory to Description Logic and modal knowledge bases for automated construction of Terminologies. The name of the thesis is "Knowledge Acquisition in Hybrid Knowledge Representation Systems".

Currently I am working on non-standard reasoning techniques in Description Logics, such as explanation or the support of debugging, and to develop logical criteria for modelling of ontologies or terminiologies.

With my former collegues in the Pionier project I worked on the use of knowledge sources for Question Answering. Please have a look at our Quartz system.

My core research area is Knowledge Representation and Reasoning with a focus on the creation of expressive formal ontologies, and their application to practical problems. It is widely accepted that intelligent autonomous agents have to agree, at least partially, on a common vocabulary for information exchange. My belief, that semantically rich, but computationally well-behaved, ontologies with a clear formal semantics can often play this role in an ideal way, is the driving force for most of my research. It strikes me that there is an enormous discrepancy between the richness of current representation languages, and the actual use of their expressiveness in ontologies or controlled terminologies. Some challenging research questions follow from this observation: what level of representation is best for a particular application, why might inappropriate levels be used, and how can I, as a formal logician, support the use of the appropriate expressive potential?

My main research interest is to understand and to bridge this gap between expressive potential and reality from a task driven perspective. One of these tasks is the construction of controlled medical terminologies, a major undertaking, which is often very expensive, error prone and highly sensitive. To help improve current modeling and reasoning environments I introduced, jointly with Ronald Cornet, formal criteria for good modeling of terminologies. By introducing methods for debugging, explanation and automatic semantical enrichment to support the modeling process, we aim at encouraging others to use more expressive formalisms.

Medical terminologies are often rigid, formal and sensitive, and, thus, not very representative. The ontologies used in Textual Information Access, the second task I have recently been involved with, are very different. Here, we use publicly available ontologies for knowledge intensive Question Answering, which are often very shallow and semantically poor. Similarly, the knowledge sources that we automatically construct from text to answer questions, currently contain almost no structure at all. This makes Textual Information Access an exciting challenge to use structured knowledge of varying representational depth, a challenge offering great opportunities to evaluate the potential and limits of formalisms of different expressiveness.

In both applications the focus has been on increasing the representational expressiveness. Often, however, appropriate ontologies are too expressive to reason with, too complex to construct, or even contradictory. From a logical perspective this induces exciting research challenges, on which I have recently embarked. An example is, whether we can restore efficient reasoning, by partitioning or by simply ``ignoring'' the contradictions, while retaining the relevant formal correctness.

Works with