The AI model is so slick users may not even realize it is translating text from other players, from a range of 16 languages including English, French, Japanese, and Vietnamese.
The vast online game community platform with 70 million daily users aims to improve communication between players, making the overall experience more comfortable, as stated by its CTO Dan Sturman.
“We know engagement goes up when users speak or interact with others in their own language,” he said. “We took that concept and removed the language barrier with the automatic translation.”
This is a significant feature not available in the real world, imagine being on holiday in a foreign country and being able to instantly converse with another person or like-minded individual. Fantastic.
Roblox started the venture by building a transformer-based large language model (LLM), training it on public data as well as internal information.
The LLM was then immersed within a mixture of expert (MoE) architecture, running several translation apps without the need for the company to build an individual LLM for each language.
The official blog release for the text translation feature detailed information on the requirements, “we need a context-aware model that recognizes Roblox-specific language, including slang and abbreviations (think obby, afk, or lol). Beyond all of that, our model needs to support any combination of the 16 languages Roblox currently supports.”
Sturman was keen to stress Roblox will continue to monitor communications between users, both for safety reasons and also to report and fix any issues with incorrect translations.
The model introduced by the game network will become more common in line with further AI developments but it is not the only translator currently in use. Facebook owner Meta has released its speech-to-text and text-to-text translator, SeamlessM4T, whilst YouTube already hosts Google’s Universal Speech Model to provide captions in over 100 languages.