Roblox’s AI Translates Chat in 16 Languages Instantly

Roblox breaks language barriers with real-time AI chat translation in 16 languages

Multi-player gaming may let you connect with people and form connections with gamers from around the world who you would never meet otherwise. However, language limitations may frequently prevent people from connecting, and Roblox's new AI chat translator aims to address this.

Roblox launched real-time AI chat translations on Monday, allowing its 70 million daily active users to converse effortlessly. Roblox AI translates chat in 16 languages, including English, Chinese, Spanish, Korean, and more.

"Thanks to our new real-time AI chat translations, we've made possible on Roblox something that isn't even possible in the physical world — enabling people who speak different languages to communicate seamlessly with one another in our immersive 3D experience," said Daniel Sturman, chief technology officer of Roblox. 

Users can use the translation tool in any Roblox experience that includes Roblox's in-game text chat service. The chat window will automatically translate any text into the user's local language without requiring the user to take further action.

According to Roblox, the translations take place in real-time with a lag of around 100 milliseconds. This allows users to converse with the same flow as they normally would.

The user does not have to lose out on the original text. If they still want to see the original message, they may just click the button next to the translated message.

To power the translator, Roblox created its own unified, transformer-based translation large language model (LLM), which can handle the difficult task of translating a combination of 16 languages to each other in a single model, saving resources compared to creating individual models for each pair, which would have resulted in 256 unique models.

Roblox stated that it is committed to enhancing the quality of its model and, as a consequence, will be releasing a platform that will allow users to offer input on translations, flag mistranslations, and even suggest better translations.

The model is already accessible for usage. English, Chinese (simplified), Chinese (traditional), French, German, Indonesian, Italian, Japanese, Korean, Portuguese, Russian, Spanish, Thai, Turkish, Polish, and Vietnamese are among the languages supported.