How Do People Communicate with Conversational Agents? Identifying Communication Styles Through Lexical and Acoustic Features

With the integration of AI technology in voice interactions, personalized adapting conversational agents (CAs) to align with users’ communication styles is crucial for achieving smoother and more natural interactions. However, how humans talk to CAs, especially their lexical and acoustic features, remain unknown. This study aims to compare Human-to-Human and Human-to-CA communication styles and identify their differences. Results indicate that Human-to-Human communication is more social-oriented, with higher use of personal pronouns, social words, and emotional expressions, alongside greater pitch and speech rates. Human-to-CA interactions are predominantly task-oriented, featuring perceptual process words, shorter sentences, and subdued acoustic variations. Moreover, communication styles dynamically adapt to interaction contexts, diminishing the influence of static personality traits. A linear discriminant analysis model achieved 77.8% accuracy in identifying styles based on these features. These findings provide guidance to designers for designing personalized and adaptive dynamic voice interaction systems to enhance users’ satisfaction and engagement. © 2026 Taylor & Francis Group, LLC.