In this paper we propose a new approach to categorization where the category naming is fully automatic. It is performed by using occurrences of Japanese nouns, verbs and particles which have specific ability for combining grammatical and semantic information. We explain the importance of automatic categorization for our commonsense research and Artificial Intelligence. Then we will introduce the technical side of the idea and the results of the preliminary trials on randomly chosen sentences from the WWW corpus. In the last part we will propose an idea of using automatically labeled categories in our Schankian script retrieval project. By developing non-human, full automatic methods for machines we expect to achieve applications which could work in open domain with as little user's intervention as possible.