Situationally Aware In-Car Information Presentation Using Incremental Speech Generation: Safer, and More Effective

Proceedings of the of the EACL 2014 Workshop on Dialogue in Motion (DM), pages 68–72, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Situationally Aware In-Car Information Presentation Using Incremental Speech Generation: Safer, and More Effective Spyros Kousidis1, Casey Kennington1,2, Timo Baumann4, Hendrik Buschmeier2,3, Stefan Kopp2,3, and David Schlangen1 1Dialogue Systems Group, 2CITEC, 3Sociable Agents Group – Bielefeld University 4Department of Informatics, Natural Language Systems Division – University of Hamburg spyros.kousidis@uni-bielefeld.de Abstract Holding non-co-located conversations while driving is dangerous (Horrey and Wickens, 2006; Strayer et al., 2006), much more so than conversations with physically present, “situated” interlocutors (Drews et al., 2004). In-car dialogue systems typically resemble non-co-located conversations more, and share their negative impact (Strayer et al., 2013). We implemented and tested a simple strategy for making in-car dialogue systems aware of the driving situation, by giving them the capability to interrupt themselves when a dangerous situation is detected, and resume when over. We show that this improves both driving performance and recall of system-presented information, compared to a non-adaptive strategy. 1 Introduction Imagine you are driving on a relatively free high- way at a constant speed and you are talking with the person next to you. Suddenly, you need to overtake another car. This requires more attention from you; you check the mirrors before you change lanes, and again before you change back. Plausibly, an attent- ive passenger would have noticed your attention being focused more on the driving, and reacted to this by interrupting their conversational contribu- tion, resuming when back on the original lane. Using a driving simulation setup, we implemen- ted a dialogue system that realises this strategy. By employing incremental output generation, the sys- tem can interrupt and flexibly resume its output. We tested the system using a variation of a stand- ard driving task, and found that it improved both driving performance and recall, as compared to a non-adaptive baseline system. Figure 1: Overview of our system setup: human controls actions of a virtual car; events are sent to DM, which controls the speech output. 2 The Setup 2.1 The Situated In-Car System Figure 1 shows an overview of our system setup, with its main components: a) the driving simulator that presents via computer graphics the driving task to the user; b) the dialogue system, that presents, via voice output, information to the user (here, cal- endar entries). Driving Simulation For the driving simulator, we used the OpenDS Toolkit,1 connected to a steer- ing wheel and a board with an acceleration and brake pedal, using standard video game hardware. We developed our own simple driving scenarios (derived from the “ReactionTest” task, which is dis- tributed together with OpenDS) that specified the driving task and timing of the concurrent speech, as described below. We modified OpenDS to pass real-time data (e.g. car position/velocity/events in the simulation, such as a gate becoming visible or a lane change) using the mint.tools architec- ture (Kousidis et al., 2013). In addition, we have bridged INPROTK (Baumann and …

Content truncated to 500 words