Design and evaluation of spoken dialog systems

Abstract

Interactive spoken dialog systems extend the range of automated telecommunication services beyond simple limited-choice form-filling applications to goal-directed tasks covering richer, more complex domains. Creating effective and efficient dialog systems requires not only accurate and robust speech recognition and language modeling, but also iterative, principled design of the user interface (UI), coupled with system evaluation with real users. This paper focusses on the design and evaluation of spoken dialog systems, first discussing how general UI principles that are well-understood for graphical UIs are also relevant for spoken language interfaces, and then describing the PARADISE framework for evaluating spoken dialog systems.