Position Papers
Did I Stutter? Why Perfect Voice Technology Discriminates Against Speech-Impaired Users
Voice assistants are great tools for offering support to vulnerable user groups. However, their systems carry inherent biases and barriers that obstruct the inclusion of diverse speech varieties, such as those of individuals with neurodegenerative language disorders. We pose that commercial voice technology is built around a narrow, idealized speaker norm that marginalizes atypical speech across the entire pipeline. Focusing on two neurological conditions, Parkinson's and Alzheimer's disease, we illustrate how speech impairments give rise to conversational barriers. Lastly, we examine how the burden of adaptation is inequitably placed on users and suggest that conversational user interfaces need to be designed for imperfect personas instead of an idealized speaker norm.
Learning from Traditional Chatbots: Adapting LLM Chatbot Interfaces to User Context
LLM chatbots combine conversational and graphical elements in increasingly complex interfaces, but often do not adapt them to the user's context. We argue that lessons from traditional chatbots can improve the graphical interface of LLM chatbots. Quick replies and cards can support users with quicker interactions, better overviews, and conversational repair. While free-text provides more human-like interaction, users might benefit in more personal tasks to be reminded of the chatbot as a machine. We therefore propose that LLM chatbot interfaces should be adaptable and select graphical elements based on the context.