A hierarchical intention recognition model for situated dialogue system

Analysis of the speaker´s intention plays a crucial role in implementation of high-quality dialogue systems. Whether or not the utterance is situated in the environment is a very important factor for the precise intention analysis in situated dialogue systems. For our teaching system we propose a hierarchical intention taxonomy. First, we separate the intention into two categories by checking whether it is situated, after that we classify the intentions into more fine-grained categories. The experimental results show that, compared with the flat classifier, the hierarchical classifier gives a better performance.