Categories and Classifications in Eurowordnet

In EuroWordNet we develop wordnets in 8 European languages, which are structured along the same lines as the Princeton WordNet. The wordnets are inter-linked in a multilingual database, where they can be compared. This comparison reveals many different lexicalizations of classes across the languages that also lead to important differences in the hierarchical structure of the wordnets. It is not feasible to include all these classes (the superset) in each languagespecific wordnet and to reach consensus on the implicational effects across all the languages. Each wordnet is therefore limited to the lexicalized words and expressions of a language. The wordnets are thus autonomous languagespecific structures that capture valuable information about the lexicalization of each language, which is important for information retrieval, machine translation and language generation. By connecting the wordnets to a separate ontology, semantic inferencing can still be guaranteed. Still, different types of classification schemes can be distinguished among the lexicalized classes. In this paper we will further describe the properties of these different classes and discuss the advantages and effects of distinguishing them in wordnetlike structures.