Online gaming giant Roblox has introduced its latest innovation: an AI-driven translation model that enables seamless communication among players, regardless of language differences. Developed by Roblox’s chief technology officer, Dan Sturman, the system facilitates real-time text conversations across 16 languages, with a remarkable base latency of just 100 milliseconds.
Sturman emphasizes the transformative potential of the technology, envisioning scenarios where players connect with newfound friends from distant corners of the globe without language barriers hindering their interactions. While Roblox already offers automatic translation for in-game content, the ambition was to enable dynamic, real-time conversations.
The journey towards this feat was not without challenges. Building a robust translator necessitated overcoming hurdles related to language diversity and real-time responsiveness. Roblox’s approach involved crafting a bespoke large language model (LLM) trained on a blend of public and proprietary data, subsequently fine-tuned by specialized translation apps for each language pair.
Addressing nuances like slang and colloquialisms posed additional complexities. Human evaluators were enlisted to decipher and incorporate trending terms, ensuring the translator remains culturally relevant and up-to-date.
The impact of this innovation extends beyond mere convenience, as evidenced by final-stage testing results showing heightened user engagement and session quality. With 70 million daily active users spanning 180 countries, Roblox is poised to redefine the gaming landscape by fostering global connections in an increasingly interconnected world.
Looking ahead, Roblox CEO David Baszucki envisions a future of metaverse interoperability, where digital assets seamlessly traverse multiple platforms—a vision that aligns seamlessly with the company’s commitment to innovation and inclusivity.
