Virtual Reality and BERT-Powered Games Redefine English Teaching in Universities

The study from the North China Institute of Aerospace Engineering shows that a VR-based English educational game, enhanced with realistic virtual agents and a BERT-powered Q&A system, significantly improves university students’ language skills, engagement, and learning outcomes. It outperforms traditional and mobile learning methods across listening, speaking, reading, writing, and motivation, demonstrating VR’s strong potential to transform English education.

A research team from the School of Foreign Languages at the North China Institute of Aerospace Engineering introduces a bold transformation of university English teaching by merging virtual reality and artificial intelligence. The study argues that traditional methods, heavy on grammar drills and memorization, no longer meet the demands of practical communication. Their VR-based English educational game seeks to simulate authentic environments, embed cultural learning, and enable dynamic interaction through intelligent virtual agents and a BERT-driven question–answering system.

A Curriculum Built Inside a Virtual World

At the heart of the research lies a three-dimensional teaching framework: strengthening language skills, enhancing cognitive learning processes, and nurturing cultural and emotional awareness. The VR game maps these goals into task-based scenarios set in everyday environments, airport security checks, hotel lobbies, supermarkets, and business negotiations. These scenes, built through Unity 3D, align closely with CET-4 vocabulary and business English textbooks, ensuring that immersion does not drift from academic requirements. Students navigate these virtual spaces, interact with objects, and complete language tasks that mimic real-life communication demands.

Virtual Agents That See, Hear, and Think

To make the experience coherent and human-like, the researchers constructed a dual-layer virtual-agent system that combines geometric modeling with behavioral kinematic modeling. The visual perception module uses a 60° field of view based on ISO visual ergonomics, while the auditory module interprets sound based on intensity and distance thresholds. These perceptual inputs feed into a finite state machine that governs decision-making, allowing virtual characters to respond to environmental changes in real time. This sensory–cognitive loop produces smooth, natural interactions, significantly strengthening the game’s sense of presence and reducing the mechanical feel typical of earlier educational simulations.

AI-Powered Quiz Interaction With Near-Perfect Accuracy

Language interaction inside the game is driven by a BERT-based textual question answering module. The model parses user commands, spoken or typed, and performs semantic matching against a curated database of 600 English question–answer pairs. To improve precision, the system blends cosine similarity and Euclidean distance with optimized weights of 0.6 and 0.4, outperforming either method alone. Expert annotation achieved a Cohen’s κ of 0.91, underscoring data reliability. The result is an exceptionally accurate language engine: up to 99.1% correctness, 99.6% message acceptance, and response times as low as 135 milliseconds. These speeds allow the AI to function conversationally without disrupting immersion.

Stronger Learning Outcomes and Higher Motivation

The study’s experimental phase involved 120 students over 80 days, comparing VR learning with mobile-assisted and traditional classroom methods. Across listening, speaking, reading, and writing, the VR group delivered the best performance, with improvements between 13% and 21% over conventional teaching. Learning interest surged by 45%, far surpassing the other groups. When benchmarked against the three educational dimensions, VR achieved completion rates of 95.8% for skills, 94.6% for learning processes, and 92.6% for emotional and cultural development. Post-test evaluations reaffirmed its advantage: highest overall scores, fastest task completion, and lowest error rates.

Although the authors acknowledge the absence of long-term retention studies, cross-device performance analysis, and standardized international exam validation, they argue that the current findings already demonstrate VR’s transformative potential. By merging immersive environments with AI-driven semantic understanding, the VR educational game presents a credible pathway for modernizing language instruction and raising student engagement and achievement.