Speaker: Volha Petukhova, post-doctoral researcher at Universität des Saarlandes, Saarbrücken, Germany

Abstract: The design of dialogue systems that exhibit interactive behaviour which is natural to its users may be expected to benefit from a good understanding of human dialogue behaviour and from the incorporation of mechanisms that are important in human dialogue. The recognition of the intentions encoded in speakers' utterances is one of the most important aspects of language understanding for a dialogue system. In computational dialogue modelling, speakers' intentions are often encoded as dialogue acts. I will discuss the ISO 24617-2 dialogue act annotation standard: basic concepts, tagset of communicative functions, relations and qualifiers. I will show in detail how dialogues can be segmented in multiple ISO dimensions and annotated. Further, I will present the results of a series of machine learning experiments carried out in order to assess the automatic joint segmentation and classification of dialogue acts. Finally, I will briefly discuss how the dialogue act information can be used by the dialogue manager to update the system's information state in order to generate adequate system's future actions.