Evaluation of GermanNet: Problems Using GermaNet for Automatic Word Sense Disambiguation

WordNets such as GermaNet have been frequently used as an inventory of word-senses for word-sense disambiguation tasks. In the work described here we evaluate the adequacy of GermaNet for this task. That is we attempt to determine the degree to which GermaNet provides an adequate inventory of senses for word-sense annotation of running text. Our findings were on the whole very encouraging. GermaNet provides an appropriate sense for 83 % of the content words in our texts. More interestingly, an error analysis showed that simple morphological processing could significantly improve coverage.