We present the last version of our system Adresy dedicated to the recognition of words written on a digitizing tablet. Adresy is designed to be used with a big (given) vocabulary. In this context, it achieves a very good performance because it is able to learn automatically the writing style of any specific user, directly from a set of a few samples of words. Moreover, Adresy improves continuously its performance in a user-transparent way, thanks to a second, faster, learning process called adaptation. This paper describes the main aspects of Adresy. Moreover, the power of our system is proven through four experiments performed on a database of ten thousand handwritten words.