Dialogue models for conversations in Slovak language

Proposed paper describes prepared corpora and the process of training and testing of statistical models of the human-human dialogue interaction. Such dialogue models can be used to model and manage human-human and human-machine dialogue interactions. Two databases of dialogues in Slovak language were prepared to provide data for training and testing. Statistical dialogue models were trained in the similar procedure as statistical language models (n-grams) used to be prepared. Several types of n-grams were trained for both - DiaCoSk and CallConv corpora. Because developed databases contain dialogues from different domains and conditions, their comparison were performed and observed differences were analyzed and discussed.