A Statistical Model to Verify Verbal Material

In natural language processing up to now mainly rule-based systems have been used. The rulebased approach has been successful for limited domains, where the manual construction of the rules is feasible. In recent studies, however, a need for automatic generation of the rules has been seen. On the other hand, when children learn their mother tongue no explicit rules seem to be needed. Therefore, in our study we try to examine how far we can get in natural language processing without any rules. Instead we use a statistical approach. It is based on our previous work in the field of associationism (Wettler 1990, Rapp 1990), but also takes up results obtained by Hilberg who uses hierarchically structured networks to significantly reduce the redundancy of texts (Hilberg 1989).