Lithuanian language has great inflexion, free word order and other features which distinguish it from other languages. This raises a problem of testing for the Lithuanian language validity of findings established for other languages. In the paper, an empirical study of a collection of Lithuanian texts is performed. It is supposed that authors of texts are basic elements of the population under study and its heterogeneity stems out of the heterogeneity of preferences and choices of the authors. An attempt to estimate structural distributions of words in a collection of texts of different authors is made by making use of a simple statistical model and empirical Bayes approach.