Analysis of the sentence writing test is conducted in this paper to support diagnostics of the Parkinsons disease. Drawing and writing tests digitization has become a trend where synergy of machine learning techniques on the one side and knowledge base of the neurology and psychiatry on the other side leading sophisticated result in computer aided diagnostics. Such rapid progress has a drawback. In many cases, decisions made by machine learning algorithm are difficult to explain in a language human practitioner familiar with. The method proposed in this paper employs unsupervised learning techniques to segment the sentence into the individual characters. Then, feature engineering process is applied to describe writing of each letter using a set of kinematic and pressure parameters. Following feature selection process applicability of different machine learning classifiers is evaluated. To guarantee that achieved results may be interpreted by human, two major guidelines are established. The first one is to keep dimensionality of the feature set low. The second one is clear physical meaning of the features describing the writing process. Features describing amount and smoothness of the motion observed during the writing alongside with letter size are considered. Resulting algorithm does not take into account any semantic information or language particularities and therefore may be easily adopted to any language based on Latin or Cyrillic alphabets.