Roblox has developed an artificial intelligence model that it claims translates text chats so quickly that users don’t even realize it’s translating other players’ messages. It works with 16 languages, including English, French, Japanese, Thai, Polish, and Vietnamese.
Dan Sturman, CTO of Roblox, said in an interview with The Verge that the goal is to make Roblox users feel more comfortable interacting with each other by allowing them to understand what each other is saying.
The translator automatically translates chats, but users can click on an icon to see the original message. It also automatically translates image assets, such as words on buildings, into the user’s default language.
“We know that engagement increases when users are talking or interacting with others in their own language,” Sturman said. “We took that concept and removed the language barrier with automatic translation.”
Roblox started by building a large language model (LLM) based on transformers, which was trained on public and internal data. It placed the LLM within a mixture of experts (MoE) architecture, an environment that ran multiple translation applications, each of which was an expert in one language. This allowed Roblox to save resources without the need to build a separate LLM for each language. Sturman said that, considering the scale of their project, his team believed it was easier to build their own model than to modify a commercially available LLM.
Sturman said that Roblox monitors chats (it does so for trust and safety reasons) and can get feedback if the translations are not exactly perfect. He added that the addition of the chat translation AI “doesn’t change anything in our privacy and safety process, and prohibited words will still be blocked.”
Roblox, in its attempt to attract older audiences in recent years, has been working with generative AI models to improve the user experience. For example, last year it launched an AI chatbot assistant for developers.
Sturman said that the hope is that, eventually, the translator model will go beyond simply translating text chats. “In the future, we could use AI to translate non-conforming [prohibited] words to conforming words or apply it to voice chats for real-time voice translation. There are many possibilities,” Sturman said.
Other companies are also developing AI translation models. Meta launched its text-to-text and speech-to-text translator, SeamlessM4T, which handles almost 100 languages. Google’s Universal Speech Model also translates around 100 languages and is already deployed on YouTube to translate subtitles.