In-Car Multi-Domain Spoken Dialogs: A Wizard of Oz Study

Proceedings of the of the EACL 2014 Workshop on Dialogue in Motion (DM), pages 1–9, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics In-Car Multi-Domain Spoken Dialogs: A Wizard of Oz Study Sven Reichel, Ute Ehrlich, Andr´e Berton Speech Dialogue Systems Daimler AG Ulm, Germany {sven.reichel, ute.ehrlich, andre.berton}@daimler.com Michael Weber Institute of Media Informatics Ulm University Germany michael.weber@uni-ulm.de Abstract Mobile Internet access via smartphones puts demands on in-car infotainment sys- tems, as more and more drivers like to ac- cess the Internet while driving. Spoken dialog systems support the user by less distracting interaction than visual/haptic- based dialog systems. To develop an intu- itive and usable spoken dialog system, an extensive analysis of the interaction con- cept is necessary. We conducted a Wizard of Oz study to investigate how users will carry out tasks which involve multiple ap- plications in a speech-only, user-initiative infotainment system while driving. Re- sults show that users are not aware of dif- ferent applications and use anaphoric ex- pressions in task switches. Speaking styles vary and depend on type of task and di- alog state. Users interact efficiently and provide multiple semantic concepts in one utterance. This sets high demands for fu- ture spoken dialog systems. 1 Introduction The acceptance of smartphones is a success story. These devices allow people to access the Internet nearly anywhere at anytime. While driving, using a smartphone is prohibited in many countries as it distracts the driver. Regardless of this prohibition, people use their smartphone and cause severe in- juries (National Highway Traffic Safety Adminis- tration (NHTSA), 2013). In order to reduce driver distraction, it is necessary to integrate the smart- phone’s functionality safely into in-car infotain- ment systems. Since hands and eyes are involved in driving, a natural and intuitive speech-based in- terface increases road safety (Maciej and Vollrath, 2009). There are already infotainment systems with Internet applications like e.g. weather, music streaming, gas prices, news, and restaurant search. However, not all of them can be controlled by nat- ural speech. In systems based on graphic and haptic modal- ity, the functionality is often grouped into various applications. Among other things, this is due to the limited screen size. The user has to start an application and select the desired functionality. A natural speech interface does not require a frag- mentation of functionalities into applications, as people can express complex commands by speech. In single-application tasks, such as calling some- one, natural speech interfaces are established and proven. However, users often encounter complex tasks, which involve more than one application. For example, while hearing the news about a new music album, the driver might like to start listen- ing to this album via Internet radio. Spoken lan- guage allows humans to express a request such as “Play this album” easily, since the meaning is clear. However, will drivers also use this kind of interaction while using an in-car spoken dialog system (SDS)? Or is the mental model of ap- plication interaction schema dominant in human- computer …

Content truncated to 500 words