A04 aims at uncovering and modeling aspects of register variation in German morphosyntax. In the first part, the project will utilize existing formal and empirical methods like Biber’s Multidimensional Analysis and develop them further by (i) using a probabilistic approach to registers, (ii) using a more sophisticated method for automatically inferring registers from distributions of features in texts, namely Latent Dirichlet Allocation, (iii) and validating the cognitive reality of the automatic identification of registers using psycholinguistic methods. In the second part, it will extend the PI’s CoreGram implementation in HPSG to model speakers’ probabilistic knowledge of register-related effects in morphosyntax. This is innovative in asmuch as no similar formal implementation of register effects has been attempted before.