Gender and accent stereotypes in communication with an intelligent virtual assistant

• Published in: International journal of human-computer studies 2025-01, Vol.195, p.103407, Article 103407
• Main Authors: Piercy, Cameron W., Montgomery-Vestecka, Gretchen, Lee, Sun Kyong
• Format: Article
• Language: English
• Subjects: Accents; Experiment; Gender; Intelligent virtual assistant; Stereotype; Trust
• ISSN: 1071-5819
• DOI: 10.1016/j.ijhcs.2024.103407

•The study explores how Siri's voice gender and accent affect users' trust.•A fully-crossed experiment with 270 participants was conducted, examining Siri's voice gender (male or female), accent (American or Indian), and task type (social or functional), with an additional condition for gender match between participants and Siri's voice.•Functional tasks received higher ratings in reliability, understandability, and faith dimensions of trust.•The study revealed nuanced effects regarding gender match and varying across accent types.•The findings have implications for human-machine communication, highlighting differences between human-human and human-machine interaction scripts. People are using intelligent virtual assistants (IVAs) more than ever before. Today's IVAs can be customized with unique voices including both gender and accent cues. Following evidence that people treat others differently based on their gender and accent, we ask: How do gender and accent of Siri, an IVA, affect users' trust? Students from two institutions (N= 270) participated in a two (Siri's voice gender: male or female) by two (Siri's voice accent: American or Indian) by two (task type: social or functional) fully crossed experiment, including a supplemental quasi-experimental condition for gender match between participants’ and Siri's voice. Results show little effect for gender or accent alone, but the functional tasks condition received higher ratings in reliability, understandability, and faith dimensions of trust. Interactions reveal nuanced effects regarding gender match and varying across accent types. Implications for human-machine communication, in particular differences between human-human and human-machine interaction scripts are presented.