ABSTRACT
One way in which drivers can send messages (e.g., text messages and emails) is using voice dictation. The technology that enables dictation in a vehicle is challenging given the noisy environment and the need for high accuracy. In this paper, we investigate the impact of input method and driving workload on the structure and complexity of dictated messages. We describe findings from a driving simulator study (Green et al., 2011) in which dictated messages were collected and analyzed. We explore the hypothesis that driving workload or input method affects the message language characteristics. We show that both driving workload and the input method significantly impact the message vocabulary and its perplexity. This may be exploited for enhancing dictation, for example by creating language models that are specific to driving and by adapting them according to workload.
