Improving parsing German by addressing the unknown word problem In recent years, neural network parsing models have achieved very high accuracies across different languages, but have we explored their full power in parsing languages like German? In my talk, I will present one of the biggest problems in parsing German, which is the high ratio of unknown words. I will first introduce the head selection parsing model (Zhang et al., 2017), that has been reported to have very good result in German, and then report our work in progress where we try to augment the word representation layer by incorporating more information from characters and compound splitting. Our first results show that Label Attachment Score can be improved by 0.5% using denser, additional character-based word embeddings.