UzWordnet and generative AI turn Uzbek learning into a four-game adventure

In a new paper on arXiv (2607.14104), Alessandro Agostini, Saydobid Khusanov, and Mirkamol Mirkamilov present a system architecture that lets learners practice Uzbek by playing games. The system integrates two core lexical resources: UzWordnet (a wordnet for Uzbek) and the largest currently available orthographic dictionary for the language. Generative AI acts as a fundamental component for learning support, providing contextual hints, feedback, and adaptive difficulty. The team designed four educational games that cover vocabulary, sentence construction, and comprehension.

The approach is novel because it serves a dual purpose. While learners improve their Uzbek, the game dynamics produce structured interaction data that can be used to enrich UzWordnet—filling gaps, adding new word senses, and improving lexical coverage. This creates a virtuous cycle: the more people play, the better the lexical resource becomes, which in turn improves the learning experience. The paper outlines a game-based methodology for achieving this without requiring explicit annotation effort from players. For professionals working on low-resource languages or AI-assisted education, this shows a practical path to scale both language learning and resource building simultaneously.