Classifying Semantic Relations in German Nominal Compounds using a Hybrid Annotation Scheme

This paper reports on novel results for the automatic classification of semantic relations that hold between the constituents of nominal compounds in German. It utilizes a hybrid annotation scheme that models semantic relations using a combination of prepositional paraphrases and semantic properties. The machine learning (ML) experiments use the support vector machine (SVM) implementation in Weka for single-label prediction tasks and Weka SVMs in conjunction with the Mulan library for multi-label prediction.