Intent Classification and Slot Filling for Turkish Dialogue Systems

The rapid development of systems with artificial intelligence plays significant role in human-machine interaction. The artificial intelligence algorithms that take instructions via natural language, can provide appropriate answers to the commands they receive. The systems with artificial intelligence which are in an interaction with humans are named as dialogue systems. In other foreign languages, especially in English, there are lots of data sets that can be used in dialogue systems. The most popular and preferred data set among these data sets is named ATIS (Airline Travel Information System). Currently, there is no Turkish data set with characteristics required by dialogue systems. With this study, the ATIS data set is adapted to implement Turkish dialogue systems. After converting the data set into Turkish, intent classification and slot-filling tasks are implemented by BERT multilingual model which is one of state of the art natural language processing methods. Micro averaged F1- score values are 89 percent and 88 percent for intent classification and slot filling correspondingly.