Focus tree: Modeling attentional information in task-oriented human-machine interaction

Abstract

This paper introduces a new model of attentional state in task-oriented human-machine interaction. It integrates three lines of research: (i) neurocognitive understanding of the focus of attention in working memory, (ii) the notion of attention related to the theory of discourse structure in the field of computational linguistics, and (iii) investigation of a corpus that comprises recordings of spontaneous speech-based human-machine interaction. The underlying idea was to make a computationally appropriate representation of attentional information that imitates the function of a focus of attention in human perception. The introduced model addresses both the research questions of storage and processing of attentional information. Finally, the paper illustrates the model for concrete interaction domains, and discusses its implementation within a prototype spoken dialogue system.