Game Master LLM: Task-Based Role-Playing for Natural Slang Learning

TL;DR

This paper addresses how to improve second-language learners' acquisition and spontaneous use of casual slang with an LLM-driven, task-based RPG. The authors implement a three-phase narrative game guided by a GPT-4o Game Master, paired with implicit and explicit feedback, to foster active spoken practice. In a between-subjects study with 14 participants, the RPG condition produced higher normalized gains in meaning and contextual usage of target phrases and greater active usage than a traditional AI classroom. Qualitative data further indicate enhanced engagement and perceived realism, suggesting narrative-driven LLM interactions can effectively support vocabulary acquisition in informal language contexts.

Abstract

Natural and idiomatic expressions are essential for fluent, everyday communication, yet many second-language learners struggle to acquire and spontaneously use casual slang despite strong formal proficiency. To address this gap, we designed and evaluated an LLM-powered, task-based role-playing game in which a GPT-4o-based Game Master guides learners through an immersive, three-phase spoken narrative. After selecting five unfamiliar slang phrases to practice, participants engage in open-ended dialogue with non-player characters; the Game Master naturally incorporates the target phrases in rich semantic contexts (implicit input enhancement) while a dedicated Practice Box provides real-time explicit tracking and encouragement. Post-session, learners receive multi-level formative feedback analyzing the entire interaction. We evaluated the system in a between-subjects study with 14 international graduate students, randomly assigned to either the RPG condition or a control condition consisting of a traditional AI-led virtual classroom. Results from an immediate post-test show that the RPG group achieved greater gains in both comprehension of the target phrases and their accurate, contextual use in sentences. Quantitative analysis of in-activity word-usage frequency, combined with qualitative survey responses, further indicates that the game-based approach provided more practice opportunities and higher perceived engagement, resulting in a more natural learning experience. These findings highlight the potential of narrative-driven LLM interactions in vocabulary acquisition.

Figures (7)

• Figure 1: Overview of the LLM-powered RPG interaction loop. Spoken user input is transcribed by Whisper and sent to the GPT-4o Game Master along with the current game state, predefined scene assets as well as descriptions of NPCs. The Game Master dynamically advances the narrative, selects and displays the appropriate background and speaking/companion NPC(s) based on the player's actions and location, updates visual elements, and naturally recasts the learner’s target slang phrases in meaningful contexts when situationally appropriate.
• Figure 2: Overview of Game Modules and LLM Narrative Generation: The agent receives the core game materials, including a set of NPC characters with brief descriptions to guide narrative progression, the current game state with possible locations and encounters, the set of target phrases, and the player’s chosen hero with its associated abilities and decision history. Based on these inputs, the model generates the next narrative segment, integrates the target phrases along with contextual explanations, and outputs the next game state and the next speaking NPC. The visual interface is then updated according to the agent’s interpretation.
• Figure 3: Narrative flow structure of the role-playing game. The game is organized into sequential phases, each containing distinct communicative goals and challenges. Within each phase, player responses drive branching interactions (illustrated by multiple paths and NPC portraits). All paths converge at a checkpoint before advancing to the next phase.
• Figure 4: Figure 4: Interface of the Game. Colored circles highlight the main elements: White – Game State (top background), yellow - Narration / Mentor's Dialogue Box (bottom-middle, AI-generated text and prompts), magenta - NPC/mentor portrait (bottom-left, updates with tone), cyan - Recording Button, red - Target Words Tracker (right panel, lists the five slang phrases with meanings; each phrase turns red after one spoken use and green after two).
• Figure 5: The visual interface of the AI english class. The Mentor's Dialogue Box (bottom-middle) displays the AI-generated narrative and prompts in response to user input. Players use the Recording Button (bottom-center) to speak their replies.