This paper is about a virtual reality system for language learning. The system includes a virtual agent that users can speak to. Understanding how people feel when speaking in a foreign language is important for adapting such systems. The aim was to study how speaking in a first language and a foreign language affects participants differently. Before the study, some educators were consulted about the design of the system. A study was conducted with some participants who spoke in two languages. Various data were collected during the study using sensors. Participants found the system useful and responded positively to speaking with the virtual agent. The results have potential implications for language education.
Conversational practice is essential for language fluency, yet in many contexts access to speaking partners remains limited. An embodied virtual agent (EVA) in virtual reality (VR) offers an on-demand alternative. However, adapting these systems to learner affect requires understanding the biosignal signatures of foreign language anxiety (FLA). The aim was to study how speaking in a first language and a foreign language affects participants differently. Before the study, some educators were consulted about the design of the system. A study was conducted with some participants who spoke in two languages. Various data were collected during the study using sensors. Participants found the system useful and responded positively to speaking with the virtual agent. The results have potential implications for language education.
Conversational practice is essential for language fluency, yet in many contexts access to speaking partners remains limited. An embodied virtual agent (EVA) in virtual reality (VR) offers an on-demand alternative. However, adapting these systems to learner affect requires understanding the biosignal signatures of foreign language anxiety (FLA). We present a VR system in which participants converse with an EVA and examine how native (L1) versus foreign (L2) conversation differentially affects self-reported anxiety, sense of presence, and physiological arousal measured via photoplethysmography (PPG) and eye tracking. Before the study, some educators were consulted about the design of the system. A study was conducted with some participants who spoke in two languages. Various data were collected during the study using sensors. Participants found the system useful and responded positively to speaking with the virtual agent. The results have potential implications for language education.
Conversational practice is essential for language fluency, yet in many contexts access to speaking partners remains limited. An embodied virtual agent (EVA) in virtual reality (VR) offers an on-demand alternative. However, adapting these systems to learner affect requires understanding the biosignal signatures of foreign language anxiety (FLA). We present a VR system in which participants converse with an EVA and examine how native (L1) versus foreign (L2) conversation differentially affects self-reported anxiety, sense of presence, and physiological arousal measured via photoplethysmography (PPG) and eye tracking. A pre-study focus group with three experienced language educators informed EVA design and task selection. In a within-subjects study (N = 40), participants completed conversational tasks while physiological data were recorded via PPG and eye-tracking sensors. Various data were collected during the study using sensors. Participants found the system useful and responded positively to speaking with the virtual agent. The results have potential implications for language education.
Conversational practice is essential for language fluency, yet in many contexts access to speaking partners remains limited. An embodied virtual agent (EVA) in virtual reality (VR) offers an on-demand alternative. However, adapting these systems to learner affect requires understanding the biosignal signatures of foreign language anxiety (FLA). We present a VR system in which participants converse with an EVA and examine how native (L1) versus foreign (L2) conversation differentially affects self-reported anxiety, sense of presence, and physiological arousal measured via photoplethysmography (PPG) and eye tracking. A pre-study focus group with three experienced language educators informed EVA design and task selection. In a within-subjects study (N = 40), participants completed conversational tasks while physiological data were recorded via PPG and eye-tracking sensors. L2 conversation elicited significantly higher state anxiety than L1 (p < .001), while sense of presence remained comparable across conditions. Across windowed HRV and eye tracking, six physiological measures showed significant L1–L2 differences in directions consistent with heightened L2 stress (BPM, MeanNN, LF, LF/HF, SampEn, and Saccade Amplitude). Participants found the system useful and responded positively to speaking with the virtual agent. The results have potential implications for language education.
Conversational practice is essential for language fluency, yet in many contexts access to speaking partners remains limited. An embodied virtual agent (EVA) in virtual reality (VR) offers an on-demand alternative. However, adapting these systems to learner affect requires understanding the biosignal signatures of foreign language anxiety (FLA). We present a VR system in which participants converse with an EVA and examine how native (L1) versus foreign (L2) conversation differentially affects self-reported anxiety, sense of presence, and physiological arousal measured via photoplethysmography (PPG) and eye tracking. A pre-study focus group with three experienced language educators informed EVA design and task selection. In a within-subjects study (N = 40), participants completed conversational tasks while physiological data were recorded via PPG and eye-tracking sensors. L2 conversation elicited significantly higher state anxiety than L1 (p < .001), while sense of presence remained comparable across conditions. Across windowed HRV and eye tracking, six physiological measures showed significant L1–L2 differences in directions consistent with heightened L2 stress (BPM, MeanNN, LF, LF/HF, SampEn, and Saccade Amplitude). Participants valued the EVA's conversational initiative and perceived L2 conversation as authentic practice, though speech-recognition accuracy for L2 speech and text-to-speech naturalness in L1 remain areas for improvement. These findings establish: 1) EVA conversation in VR offers a viable paradigm for studying FLA and 2) consumer-grade PPG and eye tracking sensors provide usable proxies for cognitive load and stress.