The use of class assertions and hypernyms to induce and disambiguate word senses

Abstract

With the spread of semantic technologies more and more companies manage their own knowledge graphs (KG), applying them, among other tasks, to text analysis. However, the proprietary KGs are by design domain specific and do not include all the different possible meanings of the words used in a corpus. In order to enable the usage of these KGs for automatic text annotations, we introduce a robust method for discriminating word senses using sense indicators found in the KG: types, synonyms and/or hypernyms. The method uses collocations to induce word senses and to discriminate the sense included in the KG from the other senses, without the need for information about the latter, or the need for manual effort. On the two datasets created specially for this task the method outperforms the baseline and shows accuracy above 80%.