In this paper I describe work for my Ph.D. dissertation which is currently in progress. The overarching goal of the work is to develop a methodology for empirically evaluating the effects of different interface design decisions in spoken dialogue systems. The methodology I will use is the dual-task method, borrowed from cognitive psychology, which is advantageous because it provides fine-grained information about the cognitive load of the user while he/she is engaged in interacting with the system. For my dissertation I will focus specifically on the use of definite referring expressions and the question of whether “natural” or “fully-specified” definite referring expressions are easier for users to generate and/or understand. The answers are important because both strategies are used in systems on the market today. More importantly, I hope my work will provide a tool for software developers, and encourage them to carefully weigh the empirically observed costs and benefits of various design decisions.