What we can learn from Dialogue Systems that don't work: On Dialogue Systems as Cognitive Models
David Schlangen · PUB – Publications at Bielefeld University (Bielefeld University) · 2009
In the ‘real world’, dialogue systems typically are made to work long days in callcentres of airlines and banks, fielding customer queries (and often inviting customer rage). In academia, a strong line of research is aimed at making such systems better at such tasks (in the hope of reducing customer annoyance). Here, I want to explore potential uses of spoken dialogue systems not as members of the workforce but in the lab, as a tool for the cognitive sciences. I argue that dialogue systems can be employed as situated, implemented computational models of language-capable agents; models whose predictions can be evaluated in real-time in ecologically valid settings, by human conversant. I sketch a methodology for building such models, propose areas where they can best be employed, and discuss the relations between research in this direction and more applied research. 1