The Effects of Embodiment and Personality Expression on Learning in LLM-based Educational Agents

TL;DR

The paper tackles how embodiment and personality expression in LLM-enabled educational agents influence learning and perceived personality. It extends a personality-driven dialogue framework by incorporating GPT-3.5 Turbo and Unity-based 3D humanoids, evaluating three modalities (dialogue-only, dialogue with animation, and fully embodied with motion) across two personality styles (high vs low extroversion-agreeableness). A 3×2 independent-subjects study (n≈210) combines quantitative LOES-S/BFI-2-XS measures with qualitative open-ended feedback, revealing that high-trait agents are more engaging and that embodiment modestly enhances engagement and perceived realism, while learning outcomes show limited model-type effects. Correlations indicate positive links between perceived positive personality traits and learning-related outcomes, with conscientiousness often most strongly associated with learning; findings highlight design trade-offs between realism, cognitive load, and engagement. The work provides practical guidance for building adaptable, expressive educational agents and offers open-source tooling to support future research.

Abstract

This work investigates how personality expression and embodiment affect personality perception and learning in educational conversational agents. We extend an existing personality-driven conversational agent framework by integrating LLM-based conversation support tailored to an educational application. We describe a user study built on this system to evaluate two distinct personality styles: high extroversion and agreeableness and low extroversion and agreeableness. For each personality style, we assess three models: (1) a dialogue-only model that conveys personality through dialogue, (2) an animated human model that expresses personality solely through dialogue, and (3) an animated human model that expresses personality through both dialogue and body and facial animations. The results indicate that all models are positively perceived regarding both personality and learning outcomes. Models with high personality traits are perceived as more engaging than those with low personality traits. We provide a comprehensive quantitative and qualitative analysis of perceived personality traits, learning parameters, and user experiences based on participant ratings of the model types and personality styles, as well as users' responses to open-ended questions.

Figures (4)

• Figure 1: Different 3D agent models used in the study expressing high (left group) and low (right group) traits.
• Figure 2: Sample screenshot from different models and their variations.
• Figure 3: Box plots of each variation's BFI-2-XS and LOES-S scores.
• Figure 4: Box plots of statistically significant effects.