Betydningsinventarer – i ordbøger og i løbende tekst

We examine a set of highly polysemous nouns in Danish in order to un-derstand how well word senses in Den Danske Ordbog (DDO) and the Danish wordnet, DanNet, correlate with the meanings found in examples of running text. The overall goal of the task is to provide adequate train-ing data for automatic word sense disambiguation. To this end, we anno-tate a corpus with a full sense inventory from DDO and compare with the annotations provided on the basis of a clustered inventory derived auto-matically from DDO and DanNet’s ontological classes. The results show that our hypothesis of clustered senses being more manageable and providing higher annotation agreement (apart from what can be expected from a higher chance agreement when there are less annotation options), holds for a vast majority of the studied cases. Further, the investigations provide a valuable assessment of the organization of senses in the studied dictionaries.