TLDR: GenQuest is a text adventure game that uses Large Language Models (LLMs) to create dynamic, personalized narratives for second language learners. It helps English as a Foreign Language (EFL) students improve vocabulary and reading comprehension through interactive storytelling. The game adapts content to a learner’s proficiency level and provides in-context vocabulary explanations. A pilot study showed promising vocabulary gains and positive user feedback, though participants suggested improvements like L1 translations, multimodal content, and finer difficulty control.
A new text adventure game called GenQuest is leveraging the power of Large Language Models (LLMs) to transform how second language learners engage with English. Developed by researchers from Hosei University, King’s College London, Kindai University, Tokyo University of Science, and the University of Sheffield, GenQuest offers an immersive and interactive storytelling experience designed to boost vocabulary acquisition and reading comprehension.
Traditional narrative games, while engaging, often suffer from static storylines and a lack of personalization, which can limit their educational impact. GenQuest addresses these limitations by using advanced LLMs to dynamically generate narratives that adapt to each learner’s choices and proficiency level. This means players embark on a unique “choose-your-own-adventure” journey, where the story unfolds based on their decisions, making the learning process more motivating and contextualized.
How GenQuest Works
GenQuest integrates two commercial LLMs: Claude Sonnet 3.7 for story generation and GPT-4o for providing clear, context-sensitive language explanations. The game’s architecture is divided into two main modules: a story module and a language module.
The story module handles the narrative flow. When a player starts a game, they select a genre (like fantasy or mystery) and their English proficiency level (from A1 to C2 on the CEFR scale). Claude 3.7 then generates a narrative outline with key story milestones and decision points, ensuring a coherent yet branching plot. As players make choices, the LLM dynamically generates story segments, and previous actions are summarized to maintain continuity. This structure, inspired by commercial interactive titles, allows player agency while guiding them through core story events.
The language module is a key pedagogical feature. During gameplay, learners can highlight any word, phrase, or sentence they don’t understand. A query icon appears, and clicking it activates the GPT-4o-powered language assistant. This assistant provides an in-context explanation tailored to the player’s proficiency level. All queried items are also saved in a personal list for later review, aiding vocabulary retention.
Pilot Study Findings
To evaluate its effectiveness, a pilot study was conducted with nine university EFL students in China over five days. Participants completed daily sessions, followed by a vocabulary test and a survey.
The results were encouraging: participants showed promising vocabulary gains, with an average score of 13.44 out of 20 on the vocabulary test, suggesting they learned or retained about two-thirds of the words they looked up. The survey also revealed positive user perceptions, with high ratings for both the game’s usefulness (average 5.33 out of 7) and ease of use (average 5.85 out of 7). Learners particularly appreciated the game’s novelty, its potential for improving English vocabulary, and its intuitive interface.
Also Read:
- AutoContext: Learning Environment Facts for Smarter AI Agents
- How Well Do LLMs Tutor? A New Benchmark Reveals Strengths and Weaknesses
Learner Suggestions and Future Directions
While the pilot study highlighted GenQuest’s potential, participants also offered valuable suggestions for improvement. Some learners found English-only explanations challenging and requested the option for first language (L1) translations (e.g., Chinese). Others suggested expanding explanations to include collocations, multiple meanings, or usage examples, and recommended recycling queried words in later stories for better reinforcement.
Participants also expressed a desire for finer-grained control over text difficulty and noted occasional issues with narrative coherence, such as logical inconsistencies or overly lengthy passages. The all-text interface was also a point of feedback, with requests for multi-modal content like illustrations to enhance engagement and accessibility. More interactive mechanics, such as fill-in-the-blank options or direct text input, were also proposed.
The researchers acknowledge these points, discussing challenges like accurately aligning LLM-generated texts with specific CEFR levels and the technical constraints of integrating image-generation models. Despite these, GenQuest represents a significant step forward in using AI for language education, offering a flexible and engaging platform that can be adapted to other languages in the future. You can learn more about this research in the full paper: GenQuest: An LLM-based Text Adventure Game for Language Learners.
