Determining the Degree of Compositionality of German Particle Verbs by Clustering Approaches

This work determines the degree of compositionality of German particle verbs by two soft clustering approaches. We assume that the more compositional a particle verb is, the more often it appears in the same cluster with its base verb, after applying a probability threshold to establish cluster membership. As German particle verbs are difficult to approach automatically at the syntax-semantics interface, because they typically change the subcategorisation behaviour in comparison to their base verbs, we explore the clustering approaches not only with respect to technical parameters such as the number of clusters, the number of iterations, etc. but in addition focus on the choice of features to describe the particle verbs.