This book makes theoretical, methodological, and empirical contributions to the design of effective and usable spoken interfaces. Specifically it investigates design implications of two general approaches to system design - the natural approach and the standardized approach - as they relate to generation of referring expressions. Both discourse reference and the generation of scalar adjectives are explored. The method of this exploration is adapted from cognitive psychology, and it is called the dual-task paradigm. Participants follow instructions generated by either a natural or a standardized system, while doing another simple task at the same time. Performance on both tasks, across participants, is used to compare the cognitive load of the two systems. Both the method and the findings will be of interest to spoken system developers and researchers. The book also provides new findings about the obligatoriness (and lack thereof) of pragmatic inference in human language comprehension, which will be of interest to psycholinguists, cognitive psychologists, and cognitive scientists.