Enterprise

Amazon is planning to rival Google with a service that translates languages

Key Points
  • Amazon has built translation technology to use internally, and now it's looking to let third parties access it.
  • Alphabet released its first translation service for developers in 2008.
Andy Jassy, CEO of Amazon Web Services and Infrastructure.
Source: Amazon Web Services

Amazon's cloud computing division has been working on a translation service that developers could use to make their websites and apps available in multiple languages, CNBC has learned.

Amazon already has machine-translation technology that it uses across the company to do things like provide product information in multiple languages. Now, the company is preparing to make it available through Amazon Web Services, said a source familiar with the matter. Amazon could announce the service before its annual re:Invent conference in Las Vegas in November.

The imminent launch comes almost two years after Amazon acquired a translation start-up called Safaba. A co-founder of Safaba, Alon Lavie, leads Amazon's machine translation research and development group in Pittsburgh.

The other major cloud infrastructure providers, Alphabet and Microsoft, sell translation services. Google first released a language translation service for developers in 2008. Facebook has developed machine translation technology for use in its own main app.

AWS in recent months has released services that draw on artificial intelligence in various ways — there are tools for recognizing objects in images and for turning text into speech, for example. A translation service would fit in well with that strategic push, which could help AWS further diversify its revenue away from the raw computing and storage resources that other companies provide.

The field of machine translation has been exploding with activity in the past two years as researchers have found gains by adopting deep learning. Deep learning is a type of AI that involves training software systems, called neural networks, on lots of data — like text snippets labeled with translations — and then getting the networks to make predictions about new data.

When Google Translate switched from a phrase-based system to a neural machine translation system for Japanese a few months ago, people immediately picked up on the higher quality of translations. Google has also added the neural system to its service for developers.

AWS, which generated $3.66 billion in revenue in the first quarter, offers more than 70 services, including applications for email and video conferencing. Companies that already extensively depend on AWS might well be willing to experiment with a new application programming interface for translation. Amazon also operates the Mechanical Turk web service that companies can use to farm out small tasks such as translating sentences to a number of people.

Independent companies like Gengo, Smartling and Unbabel rely on AWS infrastructure to operate human-powered translation services for companies.

Amazon declined to comment.