Generating Training Data for Semantic Role Labeling in German The difficulty of identifying semantic roles lies in the fact that there is no one-to-one correspondence between the syntactic structure of a sentence and its semantic representation. An additional difficulty is the lack of annotated corpora for languages other than English. Our aim is to develop methods for automatically creating rich annotated SRL data for German. To do this, we propose the use of existing NMT architectures to jointly translate the available labeled sentences from English to German and to project the role annotations to the German target sentence. As a first step, we explore a sequence-to-sequence formulation of SRL in a classical monolingual setting on English PropBank data. We analyze the suitability of such architecture by benchmarking it against existing SRL models on well known labeled evaluation data.