Using Synthetic Compounds for Word Sense Discrimination

In many morphologically rich languages, conceptually independent morphemes are glued together to form a new word (a compound) with a meaning that is often at least in part predictable from the meanings of the contributing morphemes. Assuming that most compounds express a subconcept of exactly one sense of its nominal head, we use compounds as a higher-quality alternative to simply using general second-order collocate terms in the task of word sense discrimination. We evaluate our approach using lexical entries from the German wordnet GermaNet (Henrich and Hinrichs, 2010).