Abstract :
Ontology mapping (or: ontology alignment, or integration) is one of the most active areas the semantic Web area. An increasing amount of ontologies are becoming available in recent years, and if the semantic Web is to be taken seriously, the problem of ontology mapping must be solved. Numerous approaches are being proposed, a yearly competition is being organized, and a number of survey papers have appeared. Nevertheless, with only a few exceptions, two obvious intuitions on ontology mapping have been overlooked: if humans perform "ontology mapping" in their daily life (a task we all solve every day), they do not do this in a vacuum. Instead, they exploit a rich body of background knowledge already shared by both agents involved in the mapping process. Similarly, humans do not expect that their daily-life ontology mapping is perfect. We can very well cope with approximate translations between concepts used by different agents (in fact, we are so good at it that we barely notice that we do this). In this talk I will discuss recent work where we have quantitatively shown that indeed, ontology mapping can benefit from background knowledge, and that, somewhat surprisingly, more background knowledge leads to continuously improving results. We also discuss how the use of such background knowledge can be exploited to find approximate mappings when perfect mappings cannot be found.
