Predicate Generalization in and for VerbNet-based SRL
As a Use Case for German

In my presentation, I will summarize work on German Semantic Role Labeling with our VerbNet-style role inventory, introduced in Mujdricza-Maydt et al. (2016). This includes overview on the compiled corpora and comparison of the annotations with similar SRL frameworks, as well as first attempts on learning the semantic role labels with mate+ (Roth and Woodsend, 2014).  I will discuss possible neural models useful for lower-resourced German SRL. The main bottleneck we face is data sparsity, thus different methods for data augmenation as well as efficient data usage via cross-lingual and cross-framework techniques will be presented.