Key Note Speakers

  • Craige Roberts (Ohio State University)
    The QUD in Context
    In a formal theory of how context influences linguistic interpretation, one of the things a context must do is guide the retrieval of merely implicit content—the resolution of anaphora and ellipses, and of presupposition more generally; the implicit restriction of quantificational operators and modals; and the calculation of intended implicatures, of standards of precision in the resolution of vagueness, etc.  For all of these crucial tasks in interpretation, the addressee must be able to retrieve the speaker’s intentions.  Hence, context must offer an adequate guide to those intentions. 
        Accordingly, I have proposed a theory of the context of utterance in which the exchange of information in discourse is structured by the interlocutors’ communicative intentions as reflected in the Question Under Discussion (QUD), defined in a technical sense which comports with the treatment of questions in formal semantics. The theory also imposes stringent constraints on how the QUD can evolve and be felicitously addressed which come out of insights into cooperative interaction from Planning Theory in AI.  This hybrid of Gricean pragmatics, formal semantics and Planning Theory offers a natural characterization of the role that intention recognition plays in meaning retrieval, while being naturally integrated into a dynamic semantic theory like the Discourse Representation Theory of Kamp and his colleagues, or dynamic categorial grammars like those of de Groot or Martin.  And it offers significant improvements over the Relevance Theory of Sperber & Wilson, and the SDRT of Asher & Lascarides, while being compatible with the best insights coming out of both those accounts. 
    In this talk I offer a sketch of how the theory bears on the retrieval of implicit content, focusing on anaphora and presupposition resolution.
  • Matthew Stone (Rutgers University)
    Coherence, coordination and context
    In face-to-face conversation, speakers use all the means at their disposal to get their ideas across.  They talk, but they also gesture and carry out practical actions in the world.  In this talk, I illustrate how words, gestures, diagrams and demonstrations can function together as integrated ensembles that contribute to conversation.  These interpretations depend on establishing discourse coherence: recognizing the conventional moves that implicitly connect communicative actions and update the evolving state of the conversation.  Tracking and exploiting coherence in computational dialogue agents allows systems to collaborate more robustly and to learn to adapt to their users.
    Short Biography:
    Matthew Stone completed his Ph.D. in the Computer and Information Science Department at the University of Pennsylvania in 1998.  Since then he has had an appointment in the Computer Science Department and Center for Cognitive Science at Rutgers, the State University of New Jersey, where he is now an Associate Professor.  Stone has also had visiting positions at the University of Edinburgh and the Universitaet Potsdam.  He works on problems of meaning in human-human and human-computer conversation.

Industrial Talk

  • Frédéric Weis & Michele Dominici (IRISA, Université de Rennes 1)
    Context Management in a lightly instrumented Smart Home
    This work focuses on context recognition with the goal of providing functionalities that improve comfort of inhabitants and realize energy saving in future Smart Homes. Differently from health-related research, it targets the group of “standard” consumers, with no particular needs or disabilities and thus less keen on compromising their privacy or their comfort. This drastically reduces the freedom of choice in terms of instrumentation, as the tradeoff between privacy and offered functionalities becomes more difficult to balance. Instrumentation choices impact technical aspects like context modeling and reasoning algorithms. When the instrumentation only includes non-invasive environmental sensors, many of the techniques used in activity recognition literature become less effective.

    Our contribution is the design and development of a context-capture-and-reasoning architecture supporting the provision of adapted functionalities to the inhabitants of a smart home, with the requirements quoted above. In this talk, we present the principles that characterize this architecture: the acceptability-driven design; the awareness of the gap between the reality of human activity and the capabilities of the capture process; the step-by-step abstraction of contextual information; the management of uncertainty imprecision at individual and cross-layer levels.