Roblox Unveils a Groundbreaking AI-based Real-Time Translation Model for Gamers
Summary:
Roblox has newly introduced an AI-based "unified translation model" that allows real-time text communication among players speaking different languages. The system, which supports 16 languages, was built in-house and translates messages with a base latency of 100 milliseconds. The company handled less frequently used language pairs with "back translation" and trained the system to understand slang via human evaluators. The new system showed improved user engagement and session quality in testing.
Roblox, an online gaming platform, has devised an innovative artificial intelligence (AI)-fueled “unified translation model,” enabling players to engage in real-time text communication irrespective of their language differences. Roblox CTO Dan Sturman detailed in a Feb. 5 bulletin how the company had to construct its proprietary Large Language Model (LLM) for translating textual messages within a base latency of 100 milliseconds, ensuring a smooth conversational experience for users. He further visualized a scenario where players could interact seamlessly with friends from different parts of the world, like Korea, oblivious of their respective languages. The LLM would ensure real-time translation of text exchanges across 16 languages supported on the platform.
While acknowledging that Roblox already facilitates in-experience content translations, Sturman emphasized the intent of transcending the realm of static content translation. The primary challenges in developing the translator were designing a system that supported independent translation across all 16 languages, and ensuring its rapid functioning for enabling real-time conversation. To overcome these hurdles, they adopted a unique approach to build a unified, transformer-based translation LLM that could accommodate all language pairs within a single framework.
The AI translator's development at Roblox initiated with the fabrication of a transformer-based LLM harnessing public and private datasets. Subsequently, Roblox integrated “expert” translation applications into the LLM to facilitate each language's training. For less frequently used language pairs, like French to Thai, Roblox had to resort to “back translation” due to paucity of superior quality data. The model parses translations through a quality estimation model prioritising the understandability of the translated content.
Roblox's model has also been programmed to decode human slang as part of the ongoing learning process. Human evaluators are responsible for updating the system with new and trending phrases in all languages. In the final phase of testing, the translator exhibited an enhanced user engagement and session quality. Roblox has a staggering daily user count exceeding 70 million across 180 countries worldwide, with a daily exchange of over 2.4 billion messages. Last November, Roblox CEO David Baszucki expressed his desire for interoperability, imagining an environment where metaverse users could seamlessly transfer nonfungible tokens (NFTs) and diverse digital assets across multiple autonomous platforms.
Published At
2/6/2024 4:33:53 AM
Disclaimer: Algoine does not endorse any content or product on this page. Readers should conduct their own research before taking any actions related to the asset, company, or any information in this article and assume full responsibility for their decisions. This article should not be considered as investment advice. Our news is prepared with AI support.
Do you suspect this content may be misleading, incomplete, or inappropriate in any way, requiring modification or removal?
We appreciate your report.