There are various challenges to overcome in order for such agents to succesfully communicate with each other in a dynamic environment. For example, operations for adding arguments and splitting functions are only partially defined. In addition, there are combinatorial explosions in both the object-level inference within the evolving ontology and the meta-level inference required to diagnose and repair that ontology. Such problems can be addresed by developing repair plans. A repair plan is analogous to a proof plan, a generic proof outline that can be used to guide the search for a proof. Repair plans are generic combinations of diagnosis and repair operations that guide the ontology evolution process. By grouping these meta-level operations they tradeoff completeness against a dramatic reduction in search.
To evaluate our work, we want to compare the evolutionary behaviour with the manually executed ontology modifications. For this, we work in the domain of physics as some of the most seminal advances in the development of physics have required profound ontology evolution, e.g., the discovery of latent heat and theories about orbital perturbations.
The following people are working on the project: