How to Design the Perfect Prompt: A Linguistic Approach to Prompt Design in Automotive Voice Assistants – An Exploratory Study Anna-Maria Meck Ludwig-Maximilians-Universität München, BMW Group anna-maria.meck@bmw.de Lisa Precht BMW Group lisa.precht@bmw.de ABSTRACT In-vehicle voice user interfaces (VUIs) are becoming increasingly popular while needing to handle more and more complex functions. While many guidelines exist in terms of dialog design, a methodical and encompassing approach to prompt design is absent in the scien- tifc landscape. The present work closes this gap by providing such an approach in form of linguistic-centered research. By extracting syntactical, lexical, and grammatical parameters from a German contemporary grammar, we examine how their respective mani- festations afect users’ perception of a given system output across diferent prompt types. Through exploratory studies with a total of 1,206 participants, we provide concrete best practices to optimize and refne the design of VUI prompts. Based on these best practices, three superordinate user needs regarding prompt design can be identifed: a) a suitable level of (in)formality, b) a suitable level of complexity/simplicity, and c) a suitable level of (im)mediacy. CCS CONCEPTS · Human-centered computing; · Human computer interac- tion (HCI);· HCI design and evaluation methods; KEYWORDS Automotive User Interfaces, Voice User Interfaces, Linguistics, Prompt Design Guidelines ACM Reference Format: Anna-Maria Meck and Lisa Precht. 2021. How to Design the Perfect Prompt: A Linguistic Approach to Prompt Design in Automotive Voice Assistants ś An Exploratory Study. In 13th International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI ’21), September 09ś14, 2021, Leeds, United Kingdom. ACM, New York, NY, USA, 10 pages. https://doi.org/10.1145/3409118.3475144 1 INTRODUCTION The in-car environment provides the optimal framework for speech control. The possibility for drivers to keep their hands on the wheel and their eyes on the road makes maneuvering functions with a Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. AutomotiveUI ’21, September 09ś14, 2021, Leeds, United Kingdom © 2021 Copyright held by the owner/author(s). Publication rights licensed to ACM. ACM ISBN 978-1-4503-8063-8/21/09. . . $15.00 https://doi.org/10.1145/3409118.3475144 Voice Assistant (VA) more efcient, less error-prone, and less dis- tracting than carrying them out manually. Studies fnd less lane deviation and steadier speed for participants executing functions via voice when compared to touch [1]. Furthermore, this operating mode reduces drivers’ cognitive load, not distracting them from their primary driving task [2ś4]. Designed inconsiderately though, a reverse efect can be observed. Studies show an increase in cog- nitive load when prompts (i.e. VA system outputs, e.g. in form of łOkay, I’ll start the navigation right away. Your next destination is Munichž) are designed too complexly, e.g. in terms of an intricate syntactical structure [5, 6]. The same efect can be observed when applying voice for improper use cases involving high cognitive demand. Studies suggest that VA usage can even ‘adversely afect trafc safety’ in these situations [7]. As of today, in-car voice user interfaces (VUIs) are oftentimes designed based on GUI solutions not pursuing a voice frst approach [8]. This adds to the above- mentioned issue as both interfaces difer in many regards. Visual aids in VUIs are reduced or entirely absent in comparison with GUIs, making it harder to convey information. Additionally, the lack of a visible hierarchical structure makes revisions and edits more difcult for users [9]. Regarding the operation mode, a dimin- ished sense of agency for users interacting with a speech interface compared to a keyboard interface can be found. A study by Lim- erick et al. links this to the increased cognitive working memory load accompanying the use of speech [10]. With technical advance- ments supporting more and more complex use cases via voice in the future, this problem will intensify. In addition, the number of users of in-car VAs is still on the rise, increasing the amount of in-car conversations overall [11]. Designers of in-car voice experiences are thus facing the problem of designing for a surging number of users whilst handling the requirements of increased technical complexity without sufcient guidelines. Research has been conducted as to how a conversational user interface needs to be designed in terms of best practices for dia- log guidance and dialog management, covering structural as well as technical aspects of voice design [2, 7, 12ś14]. The design of prompts on a linguistic-centric level has received less attention. To the extent of the authors’ knowledge, the composition of sys- tem outputs has not been studied on a broad linguistic spectrum. Language-dependent syntactical, grammatical, and lexical parame- ters infuence drivers and their driving performance though [15]. It is therefore crucial that the design of VA system outputs is be- ing carried out attentively. Moreover, Stier et al. exposed diferent user preferences for syntactical structures when comparing them across in-vehicle use cases [5]. In addition to these use case efects, the type of prompt also has a potential impact on its evaluation. In order to investigate this, we propose a three-part cluster for 237