Automatic recognition of dialogue acts in complex typology

Experiments on automatic recognition of dialogue acts in the Estonian Dialogue Corpus are described. A complex typology consisting of 126 dialogue acts is used. Two different probabilistic suffix tree models have been implemented. The first model uses substrings of utterances in order to classify utterances into several dialogue acts. The second model uses substrings together with discourse information in the form of preceding and following dialogue act. Average recall and precision of 45%-56% have been achieved in the experiments. Patterns for recognition of main information acts are analysed.