Meta Wants to Use AI to Improve Language Translation

A new effort to translate every language could democratize the internet, experts say. 

Meta has announced a research project to create translation software that works for "everyone in the world." It's part of a growing effort to serve the approximately 20 percent of the world's population who do not speak languages covered by current translation systems. 

"Communicating properly speaking the same language is hard enough; trying to capture and understand the universal nuances of different languages is a whole other ballgame," Scott Mann, the co-CEO of Flawless AI,  a neural net film lab, told Lifewire in an email interview. "To date, the only way to do translation has been for 'human translators' to learn multiple languages and try to interpret and bridge language barriers for different uses."

Meta Translation?

Meta is planning a long-term effort to build language and MT tools that will include most of the world's languages. The company is building a new advanced AI model called No Language Left Behind. It says it will learn from languages with fewer examples to train from and use it to enable expert-quality translations in hundreds of languages, ranging from Asturian to Luganda to Urdu. 

Another project is Universal Speech Translator, where Meta is designing novel approaches to translating from speech in one language to another in real-time to support languages without a standard writing system and those that are both written and spoken.

"Eliminating language barriers would be profound, making it possible for billions of people to access information online in their native or preferred languages," the company wrote in the blog post announcing the project. "Advances in MT (machine translation) won't just help those people who don't speak one of the languages that dominate the internet today; they'll also fundamentally change the way people in the world connect and share ideas."

But some experts say that Meta's translation project faces enormous hurdles. "The industry is definitely still in research mode, and we're excited to be a part of that, but it's not about to be deployed in a product that eight billion people will start using tomorrow," Jesse Shemen, the CEO of translation software company Papercup said in an email interview. 

The current issue is translation quality, Shemen said. His company uses the human-in-the-loop type models for both transcription and translation. "You can reap the vast speed benefits of machine learning while achieving the last mile of quality that people have come to expect with a human touch," he added. 

Better Understanding

Work on improving translation software has been underway for decades. In the film tech industry,  Flawless has developed a technology called TrueSync, that captures and translates the nuance of language and deep expression, enabling authentic, accurate translations of film performances from one language to another. 

The TrueSynch tech allows for 3D translation of monocular images, meaning it can make controlled changes to original footage with a photorealistic output—preserving all the emotions and nuances of the intended performance,  Mann said. "While this is not running in real-time (which is needed for day to day translation such as Meta), it demonstrates the massive potential of AI and Neural networks in the translation domain."

Better translation software also makes good business sense. When companies go global, staffing up native-speaking support agents for every language is tough to do. 

"This is where having accurate and reliable translation software becomes so important: if companies can empower their monolingual (i.e., English-only-speaking), already-trained customer support agents to chat and email with their customers through a translation technology layer, that immediately improves efficiency," Heather Shoemaker, the CEO of the translation software company Language I/O, said in an email interview. 

Shoemaker says her company's software allows companies to communicate with customers in any language through proprietary machine learning technology. The AI-enabled technology can generate real-time, company-specific translations of all user-generated content (UGC), including jargon, slang, abbreviations, and misspellings, into over 100 languages via chat, email, article, and social support channels. 

"Breaking the language barrier enables us to communicate and understand effectively what is being communicated without the possibility of misinterpretation," Mann said. "The world needs to communicate better, and language is the biggest barrier to understanding one another."