Automatic taxonomy building within an object formalism

Within an object knowledge representation formalism, class taxonomies are basic structures used to organize and query the knowledge base. It is therefore prootable to provide an object representation formalism with automatic means for taxonomy inference. Methods deriving classes from sets of unlabeled individuals have been developed within the data analysis and the machine learning elds but these methods use, in general, data descriptions which are less expressive than object models. A taxonomy building strategy based on object proximity estimations is presented in this paper which deals successfully with advanced features of object formalisms like inter-object relationships, multiple perspectives and generic typing. One of its basic tools is the topological object proximity model, which is universally applicable on object features and exible, i.e. it includes a range of concrete dissimilarity functions. The output of the topological function are used by various clustering algorithms to build a hierarchy of object classes, characterized both intensionally and extensionally. The strategy has been implemented by the T-tree module within the Tropes object knowledge model.