The detection of hypernymy in texts has a long history in the community of natural language processing. Many quite efficient algorithms have been developed to recognize them in large corpora. The question arises whether these algorithms can be exploited to detect other semantic concepts such as grammaticalization and semantic variance and change.

This thesis tests some of those algorithms on large corpora by transforming the latter one into a distributional semantic model which represents the proximity and similarity of the words inside the corpora. This allows a detailed discussion and analysis of the precision of the algorithms later in this thesis.