Google employs machine learning to boost translation capabilities to near-human level

google ai translate
Malarie Gokey/Digital Trends
No one would accuse Google Translate, the favored tool of unscholarly high school language students everywhere, of being an inaccurate interpreter. The 10-year-old internet interpreter can fluently translate more than 100 tongues, recognize foreign restaurant menus and signage, and differentiate between dialects in real time. But there’s always room for improvement, and in Translate’s case, it’s occurring through machine learning.

The project is called Google Neural Machine Translation, or GNMT, and it isn’t strictly speaking new. It was first employed to improve the efficiency of single-sentence translations, explained Google engineers Quoc V. Le and Mike Schuster, and did so ingesting individual words and phrases before spitting out a translation. But the team discovered that the algorithm was just as effective at handling entire sentences — even reducing errors by as much as 60 percent. And better still, it was able to fine-tune the accuracy over time. “You don’t have to make design choices,” Schuster said. “The system can entirely focus on translation.”

In a whitepaper published on Monday, the Google Brain team detailed the ins and outs of GNMT. Under the hood is long short-term memory, or LSTM, a neural networking technique that works a bit like human memory. Conventional translation algorithms divides a sentence into individual words which are matched to a dictionary, but LSTM-powered systems like Google’s new translation algorithm are able to “remember,” in effect, the beginning of a sentence when they reach the end. Translation is thus tackled bilaterally: GNMT breaks down sequences of words into their syntactical components, and then translates the result into another language.

GNMT’s approach is a boon for translation accuracy, but historically, such methods haven’t been particularly swift. Google, however, has employed a few techniques that dramatically boost interpretation speed.

As Wired explains, neural networks usually involve layered calculations — the results of one feeds into the next — a speed bump which Google’s model mitigates by completing what calculations it can ahead of time. Simultaneously, GNMT leverages the processing boost provided by Google’s specialized, AI-optimized computer chips it began fabricating in May. The end result? The same sentence that once took ten seconds to translate via this LSTM model now takes 300 milliseconds.

And the improvements in translation quality are tangible. In a test of linguistic precision, Google Translate’s old model achieved a score of 3.6 on a scale of 6. GNMT, meanwhile ranked 5.0 — or just below the average human’s score of 5.1.

It’s far from perfect, Schuster wrote. “GNMT can still make significant errors that a human translator would never make, like dropping words … mistranslating proper names or rare terms … and translating sentences in isolation rather than considering the context of the paragraph or page.” And prepping it required a good deal of legwork. Google engineers trained GNMT for about a week on 100 graphics processing units — chips optimized for the sort of parallel computing involved in language translation. But Google is confident the model will improve. “None of this is solved,” Schuster said. “But there is a constant upward tick.”

Google isn’t rolling out GNMT-powered translation broadly, yet — for now, the method will remain relegated to Mandarin Chinese. But the search giant said it’ll begin AI-powered translations of new languages in the coming months.

GNMT may be the newest product of Google’s machine learning experiments, but it’s hardly the first. Earlier this year, AlphaGo, software produced by the company’s DeepMind division, became the first AI in history to beat a human grand master at the ancient game of Go. Earlier this summer, Google debuted DeepDream, a neural network with an uncanny ability to detect faces and patterns in images. And in August, Google partnered with England’s National Health Service and the University College London Hospital to improve treatment techniques for head and neck cancer.

Not all of Google’s artificial intelligence efforts are as high-minded. Google Drive uses machine learning to anticipate the files you’re most likely to need at a given time. Calendar ‘s AI-powered Smart Scheduling suggests meeting times and room preferences based on the calendars of all parties involved. And Docs Explore shows text, images, and other content Google thinks is relevant to the document on which you’re working.

Movies & TV

Apple’s first original feature film reunites Bill Murray, Sofia Coppola

The Lost in Translation team of Sofia Coppola and Bill Murray will reunite with their A Very Murray Christmas collaborator Rashida Jones for On the Rocks, Apple and studio A24's first official feature together.

Chromebook 13 vs. Google Pixelbook: Acer model takes on the king

Acer's Chromebook 13 is throwing tons of speed at the Chrome OS market, to go with a midrange build and traditional clamshell design. Is that enough to challenge the Google Pixelbook?

Benchmark scores surface for Google's midrange Pixel 3 XL Lite

The Google Pixel 3 and Pixel 3 XL are considered to be two of the best Android smartphones, but it looks like Google could be prepping a midrange line. Say hello to the Pixel 3 Lite and Pixel 3 Lite XL.
Emerging Tech

Face-scanning A.I. can help doctors spot unusual genetic disorders

Facial recognition can unlock your phone. Could it also be used to identify whether a person has a rare genetic disorder, based on their facial features? New research suggests it can.

Swatch and Mastercard team up for on-the-wrist, on-the-go mobile payments

Swatch has announced its Swatchpay technology is now available in Switzerland, enabling mobile payments from your Swatch watch. It works in a similar way to Apple Pay and Google Pay. Here's everything about it.

Sony is showing something off at MWC -- will it be the Xperia XZ4?

Sony may have released the Xperia XZ3 in the past few months, but already it's preparing to release a follow-up, the Xperia XZ4. We're learning plenty about the phone now some details have started to leak out, and it's getting exciting.

Do these Geekbench results accurately represent the Moto G7?

The Moto G6 range is still relatively new to the market, but rumors have already started about the Moto G7, which is expected some time in 2019. Apparently, a G7 Power version will be joining the G7, G7 Play, and G7 Plus.
Home Theater

Set your ears free with the best completely wireless earbuds

If you can't stand the tangle of cords, or you're just excited about completely wireless earbuds, you're going to need some help separating the wheat from the chaff. Our list serves up the best true wireless earbuds around.

Is this the first image of a Galaxy S10 being used in real life?

It won't be long now; With 2019 underway, the Samsung Galaxy S10 is almost here. Before it arrives, here's absolutely everything you need to know about all three of Samsung's next flagships.

Omron HeartGuide brings blood pressure monitoring to your wrist

High blood pressure leads to heart attacks, strokes, and many other health problems, so it's important to keep an eye on. Omron's HeartGuide is a fitness tracking watch that can also monitor your blood pressure from your wrist.

Google is buying mysterious smartwatch tech from The Fossil Group for $40 million

Google is about to step up its smartwatch game. The company has agreed to buy an unnamed smartwatch technology from The Fossil Group for a hefty $40 million. Considering the acquisition, it's clear Google is serious about smartwatches.

Learn how to play YouTube in the background on iOS and Android

We show you how to play YouTube in the background with apps such as Opera, Chrome, and Firefox -- along with the premium offerings like YouTube Premium -- whether you have an Android or iOS device.

Android vs. iOS: Which smartphone platform is the best?

If you’re trying to choose a new phone and you’re not sure about the merits and pitfalls of the leading smartphone operating systems, then come on in for a detailed breakdown as we pit Android vs. iOS in various categories.

Verizon’s deal could get you a free iPhone XR — but there’s some fine print

Verizon launched a new deal for its smartphones aimed at encouraging customers to open a new line. If you're willing and you want two new phones, you could get a free Samsung Galaxy S9, iPhone XR, or Pixel 3.