Handwritten words look differently for different scribes. The idea is to find a descriptor that describes subtle features for scribal attribution of handwritten text. Furthermore, descriptors are essential to recognise and classify unknown words. In this project we will investigate different kinds of handcrafted feature descriptors for recognition where learning data is too small for deep learning methods. We will also investigate different kinds of shallow neural networks that require only a small amount of labeled data to achieve good results. We will investigate smart clustering methods, which we have devised and will now analyse more extensively. In this addition we will build upon all the things aforementioned, which are almost done. We have published papers during this project and aim to do that for the coming year as well.