CLASSITALL: Incremental and Unsupervised Learning in the DIA-MOLE Framework

The learning algorithm CLASSITALL is a module of DIA-MOLE, a tool that supports an engineeringoriented approach towards dialogue modelling for a spoken-language interface. CLASSITALL is a descendant of the COBWEB conceptual clustering algorithm, and in this paper we especially focus on extensions that are necessary for processing data within a spoken language system environment. While most learning algorithm handle simple data, like e.g. attribute-value pairs, we introduce an approach to use uncomplete and uncertain as well as highly structured knowledge within a learning task. CLASSITALL clustering of structured values is based on sets of least general generalisations. For reasons of efficiency we developed an algorithm with an lazy evaluation strategy for constructing local hierarchies and for propagating information. Some improvements for using classification hierarchies are also introduced.