Implicit Knowledge in Argumentative Texts In argumentative texts the connections between sentences often rely on implicit knowledge. Such knowledge is important for automated argument analysis, e.g. to judge how solid an argument is or to construct concise arguments. We design a process for obtaining high-quality implied knowledge annotations for argumentative texts. This process involves several steps to promote agreement and monitors its evolution using textual similarity computation. Analysis of the added knowledge shows that (i) it is similar to sentences found in Wikipedia, (ii) a majority of it can be mapped to common sense knowledge relations, and (iii) it is characterized by a high proportion of generic sentences. Based on the latter observation we then build a language independent semantic clause type classifier which among others distinguishes between generic sentences, generalizing sentences, events and states. We explore this task in a deep learning framework, where we introduce an attention mechanism that pinpoints relevant context within the classified clause and from earlier clauses. Furthermore, we present a novel take on modeling and exploiting genre information which helps improving classification performance. We present experiments for English and German that achieve competitive performance.