COBRA is a tool for performing context-based reasoning in OWL ontologies, with an emphasis on its applications. COBRA is built on top of standard OWL reasoning tools, and uses the OWL-API, which makes it easy to use, maintain, and keep up-to-date with reasoning technologies. To understand the behaviour of COBRA, we must rst explain the idea of context-based reasoning. We consider as input for the system an annotated OWL ontology. Intuitively, this annotated ontology is a compact representation of a class of ontologies, which we call contexts. More precisely, every axiom in an OWL ontology is annonated, using the owl:annotation tag, with a list of labels that express the contexts that this axiom belongs to. For example, if we have two contexts C1 and C2 containing axioms 1 and 2; and 2 and 3, respectively, then the annotated input ontology O will contain three axioms 1; 2, and 3 annotated with fc1g, fc1; c2g, and fc2g, respectively. One of the main goals of context-based reasoning is to identify the class of contexts that entail a given logical consequence. An obvious method for solving this reasoning problem would be to rst extract all the di erent contexts from the annotated ontology, and perform standard reasoning in all of them. Given the speed of state-of-the-art OWL reasoners, this approach, in the following called the nave approach produces a very e cient reasoning procedure for realistic OWL ontologies, assuming that the number of contexts is small. However, as the number of contexts grows, so does the total execution time of the nave method, by a linear factor. To handle cases where the number of contexts is high (e.g., above a thousand) more e ective methods based on axiom-pinpointing, for detecting large clusters of contexts entailing the consequence, through only one call to the reasoner, and a variant of Reiter's Hitting Set Tree (HST) algorithm [3] for ensuring that all possible contexts have been detected through a systematic search. Speci cally, we use the HST approach for computing the so-called boundary of a consequence described in [1]. Notice that in this case, the HST approach can be further optimized, since the background lattice required in [1] corresponds to the class of all sets of labels, with union and intersection as operators, which are easier to compute. COBRA implements the nave and the HST methods via a sequence of blackbox calls to standard OWL reasoners. For the HST method, it must call a rea? Partially supported by DFG within the Cluster of Excellence `cfAED' ?? Supported by the International MSc Program in Computational Logic (MCL)
soner capable of explanation services. Speci cally, the reasoner must not only
answer whether a consequence holds from an ontology or not, but in the a
rmative case, additionally provide a justi cation; that is, a minimal sub-ontology
still entailing this consequence. Additionally, since the nave and the HST
algorithms both require reasoning over several, slightly di erent, ontologies,
incremental reasoning tools have a strong positive impact in the performance of the
overall system. The current version of COBRA uses the explanation capabilities
of HermiT [
A typical application scenario for context-based reasoning is in the area of
error-tolerant reasoning, where the di erent contexts correspond to the repairs of
an unwanted consequence (see [
The demo will showcase the behaviour of the di erent components of the tool and their use for practical applications, with special emphasis on error-tolerant reasoning.