Linguistic and gender factors in user engagement with Arabic LLM-based virtual agents for rehabilitation

Large Language Model (LLM) based virtual agents are transforming robotic rehabilitation by providing personalized, adaptive support. However, when using LLM for rehabilitation, we need to ensure user acceptance and understand how AI-generated language and/or virtual appearance can affect users' responses and engagement. This study investigates the impact of word selections, voice and virtual agent gender on Arabic-speaking users' emotions and engagement in rehabilitation using LLM-based virtual agents. A multi-factorial design will be employed, manipulating command politeness (soft vs. harsh), dialects, and voice/agent gender. Arabic-speaking participants will be exposed to commands presented by LLM-based virtual agents with the variables systematically varied. Emotions will be quantified using a face emotion recognition system, while the engagement will be assessed through user feedback. Our findings showed variance in users' emotional responses to different stimuli, highlighting the importance of considering language style and gender when intervening with users. The results provide insights into cultural and linguistic factors influencing user interactions with LLM-based virtual agents in rehabilitation contexts. The study also contributes to the broader field of human-computer interaction, emphasizing the importance of considering linguistic and cultural factors in designing interactive rehabilitation systems.