This week Google announced a new cloud service called Translation Hub, which allows companies to translate documents on a self-service basis.
An employee can input their document in the Hub and select all the languages for which they’d like translations. The Hub then quickly spits out the translation, even preserving the formatting of the original document (if it’s a Google Doc or Slide, PDF, or Microsoft Word doc). Google says the Hub also offers management controls so that users can easily scrutinize the translations and give feedback.
“Before machine translation became dominant, customers typically would send their documents out to a translation service and sometimes wait weeks to get a translation,” says June Yang, Google’s VP of cloud AI and industry solutions. “Here they can do a self-service, and in a matter of 30 seconds get a translated document back.”
The service is ideal for multinational companies and those with a global customer base. Materials science company Avery Dennison has been using a trial version of the Hub in collaboration with Google for a few months, says Murali Nathan, who runs the company’s digital innovation and employee experience group. Before transitioning to Translation Hub, the translation work at Avery Dennison was very decentralized.
“Each department had spent their own budget on a translation provider,” Nathan says. “And they build and retain what is called its translation memory,” meaning all the intelligence and experience accumulated from translating Avery documents (e.g., with the right word choice, technical terms, and tone). But all that knowledge was mostly stored in human brains, some of them outside the company.
“If the translation provider had a person exiting [and] they brought in a new person, that’s going to affect the quality of translations,” Nathan says, noting that in the time Avery Dennison has used the Translation Hub (aided by some specialized language models trained on Avery-specific lingo) he’s seen a 90% reduction in the company’s translation costs.
Google Translate launched in 2006 and was a purely consumer-facing product for a good six years. By 2012 Google realized that companies were trying to use the product for business documents, so it created an application programming interface (API) for the service. “Basically all they got from the API was exactly what they would get from Google Translate, just automated and supported,” says Macduff Hughes, who has led the Google Translate engineering team since 2012.
In 2016 machine translation got a major boost from the advent of neural networks. New natural-language models that could translate whole sentences began obliterating previous records for accuracy. Since then, Hughes says, “the translation industry has been moving more and more to (human) post-editing of machine translation being the standard way to do translation.”
Neural networks also opened the door to offering translation as a cloud service. As Hughes says, it turned the Translate service into “something that really worked for professional use cases.”
After Translate debuted in the cloud, Google began adding other features. In 2018 it launched Auto ML translation, which gave enterprise customers a way to customize the translation models to support their own vocabulary and stylistic requirements. This more specialized machine learning model is trained using data provided by the customer, according to Google, which charges its cloud customers a per-page translation fee. Generic translation without any customization costs 15 cents per page. More advanced or customized translation can cost as much as 50 cents per page.
Avery Dennison’s Nathan is pleased that Google’s Translation Hub has saved his company time and money, but he sees something more profound in it, too. He points out that better translation service means the ability to communicate with people in their own language, which can make them feel more valued and included.
“I wanted to make every employee feel they are part of this organization, because that is really key for our D&I [diversity and inclusion],” Nathan says. “D&I talks about diversity and inclusion, about cultural diversity, ethnic, binary, nonbinary . . . but what about language diversity? It’s a dimension. And I thought that that dimension is really important, especially to me being the leader of digital employee experience.”
