We intend on developing electronic dictionaries and Finite State Transducers for the automatic processing of the Albanian Language. We describe some peculiarities of this language and we explain how FST and generally speaking NooJ’s graphs enable to treat them. We point out agglutinated words, mixed words or ‘XY’ words that are not (or cannot be) listed into dictionaries and we use FST for their dynamic treatment. We take into consideration the problem of unknown words in a lately reformed language and the evolving of features in the dictionaries.