Gábor Borbély's talk: We investigate a theoretic approach to model the length of natural sentences. The valency of a word is in the center of our interest. We present our results and experiences of fitting our model parameters to corpus data.