False memories: subjective and objective responses to an “accented” voice assistant

The default accent of voice assistants tends to be a standard variety of a language, and commentary around creating voice assistants (VAs) suggests that more regionally marked dialects are avoided for their more marked social associations and their (assumed) reduced intelligibility. In our study, participants "user-tested" two VAs, one with a Southern US English accent, and one with a Mainstream US English accent. Northern US participants evaluated the Southern VA more negatively and commented that she was harder to understand. On the other hand, Southern participants had mixed valence affective reactions to the Southern VA and International participants demonstrated no strong social associations with either accent, and both groups rarely mentioned intelligibility of either VA. Additionally, in a listening game (a short version of the DRM false memory paradigm), there were no significant effects of VA accent on real or false memories for any of the three participant groups. Our results add to work suggesting that while linguistic variation can lead to challenges in communication, intelligibility is often invoked too easily as a reason for exclusion. The potential benefits of and problems around expanding variation in the voices of voice assistants are also discussed.