One of the most fundamental research questions in the field of human-machine interaction is how to enable dialogue systems to capture the meaning of spontaneously produced linguistic inputs. This paper introduces an approach to this research question inspired by recent neuroimaging studies of working memory operations. It is widely acknowledged that working memory plays an important role in language understanding. This paper proposes a computationally appropriate method for meaning representation in dialogue systems based on three separate but interrelated operations that are inherent to working memory: mnemonic selection, updating the focus of attention, and updating the contents of working memory. We suggest that this method provides a framework for more robust natural language understanding and designing attention-based dialogue strategies.
展开▼