People in everyday conversation do not speak fluently. We often include hesitations, repetitions and restarts in our ongoing speech. Computational and formal models of dialogue do not currently directly incorporate this key aspect of interaction, treating it orthogonally to the processing of ‘clean’ utterances. Disfluencies, however, are often meaningful rather than accidental, and exclamations such as ‘ooh’ and ‘ah’ can also have meaning, sometimes conventionalized, in dialogue contexts. The aim of the 3-year DUEL project between Université Paris Diderot (Paris 7) and Bielefeld University is to model the human capacity for speaking and understanding disfluent utterances, and to create formal models and computational systems capable of this processing. The other challenge in this enterprise is to model this interaction incrementally, that is, online as it happens word-by-word in real dialogue.
Another heretofore under-researched area of computational and formal linguistics is laughter. It has been known by interaction scientists for some time that laughter in dialogue often has very little to do with humour. As with disfluency, precise interactional effects of laughter can also be explored through formal and computational modelling of the state of speakers in dialogue. Consequently, the other aim of this project is to incorporate laughter into our speaker models and working dialogue systems.