International evaluation of NLU benefits in the domain of in-vehicle speech dialog systems

Abstract

An in-vehicle speech dialog system (SDS) can support visualhaptic interfaces and reduce eyes-off-road time while driving. This work evaluates two SDS that varied according to the degree of natural language understanding afforded by the speech dialog system. In a Wizard of Oz simulation, two alternative SDSs were tested in a driving simulator. The Lane Change Test was used to compare a command and control system with a system supporting natural language input. This driving simulator study was conducted using the same setup in Germany, USA, and China. 40 participants per country were instructed to perform interaction tasks from contexts like media, telephone, and navigation. The results show that natural language SDS could lead to a faster and more intuitive way of interacting with in-vehicle SDS. US and Chinese users especially preferred the natural language enabled system over the command and control system.