This thesis investigates whether an immersive, mixed-reality drawing game can enhance Japanese vocabulary learning and learner affect. Ten university learners of Japanese (Elementary I–II) used Meta Quest 3S headsets inside a private Horizon Worlds space to play “Paint in 3D,” guessing partner drawings either with single nouns or full Japanese sentences while avatars interacted in real time. The experimental design compared short-term retention and language output across the two formats and analysed post-session survey data on enjoyment, motivation and anxiety. Both groups demonstrated excellent short-term recall: the single-word group averaged 100% and the sentence-level group 92% (4/5 perfect scores) on a picture-based quiz, indicating that the embodied, dual-coded context provided by virtual reality (VR) supported rapid vocabulary binding. As expected, the sentence condition resulted in substantially more Japanese speech, supplying added practice with grammar structures, such as particles, while the single-word task offered quicker low-pressure naming practice. Survey results showed very high enjoyment (M = 4.5/5) and willingness to reuse VR; the sole learner who rated the session only “neutral” was also the only one scoring below 80 % on the quiz, suggesting a positive affect–achievement link. Learners established avatars and game-like flow with lowering anxiety and supporting engagement, aligning with dual-coding and affective filter theories. Overall, both guessing strategies proved equally effective for short-term retention, but they shaped the nature of language practice differently. The findings confirm that a brief, carefully designed mixed-reality activity can deliver strong vocabulary gains while increasing motivation and confidence, highlighting augmented reality (AR) as a possible bridge between physical and digital language spaces and recommending future work on longer-term retention and adaptive task complexity.