Workshop Aims. Traditional approaches to dialogue modeling have focused on the representation and manipulation of symbolic representations of agent and dialogue states as well as communicative actions. More recently, data driven probabilistic approaches have been receiving greater attention within the research community. The aim of this workshop is to explore the prospects for hybrid approaches that combine both paradigms as well as analyses of the types of problems that are more amenable to one approach versus another. Thematic Questions Some of the questions that workshop participants will discuss and attempt to answer include: · Can a logic-based NLU (natural language understanding system) be combined effectively with a classification-based one? · How can we profitably combine logic-based methods for `"speech act" interpretation with probabilistic methods for the same? · What are the differences between these methods when it comes to interpretation versus generation of Natural Language utterances? · Some of the trade-offs in speed and robustness between simpler template-based or canned text output solutions and deeper, grammar based generation of utterances are well known. But can we have a fuller picture of these trade-offs and of their range of application? · Are there any advantages of probabilistic methods when framed in terms of plan based approaches or discourse coherence relations? How can natural language pipelines that incorporate a probabilistic approach at one stage of processing (for example, speech act understanding) be integrated with another module, say, a dialogue module that is strictly logic-based?
Find the program of the workshop here