A probabilistic parser as a model of global processing difficulty

We present a model of global processing difficulty in human parsing. This model is based on a probabilistic context-free grammar and is trained on a realistic cor-pus sample. It achieves broad coverage and good pars-ing accuracy on unseen text, and its predictions are sig-nificantly correlated with experimental data on word or-der preferences in German. The model makes predictions about the differential behavior of verb final and verb ini-tial sentences and provides evidence for the importance of lexical information in sentence processing