The Latest in Machine Translation Tech

The boom in machine translation (MT) is continuing, with the market predicted to exceed $1.5 billion by 2024. With global companies looking to provide more personal, localized experiences for customers at higher volumes, improvements in AI and technology mean they are increasingly turning to MT for cost-effective translations. And with key players releasing new technology in the next few months, this trend is showing absolutely no sign of slowing down. To keep you in the loop, here’s a breakdown of recent technology developments in the world of machine translation, along with some cautionary tales for companies thinking of taking the leap for themselves.

Google Assistant introduces ‘Interpreter mode’

By 2022, 55% of households are set to own smart speaker devices, and 50% of searches this year are expected to be done by voice. One of the latest advances is Google Assistant’s new Interpreter mode, which translates conversations in real time in 29 different languages, including Thai, Slovak and Hindi. The Interpreter mode was announced at CES last year and is now available on smartphones, smart displays and Google Home speakers. Running on the Google Nest Hub, the technology is already being snapped up by the leisure, travel and tourism industries to help improve the experience of international customers. American Airlines has announced that its Admirals Club lounges at LAX airport will offer the service when a multilingual team member is not available, while hoteliers are also keen on offering it to international guests following successful trials at several hotels in 2019.

Waverly Labs launches the Ambassador

In November 2019, Waverly Labs announced its latest translation earpiece, the Ambassador: an over-ear device which can be used to either hold a conversation in real-time with a partner or automatically detect and translate any speech within a radius of 8 feet. The device works when paired with the user’s smartphone’s text-to-speech app, which transcribes the speech, translates it into a new language and reads it back to the listener. The Ambassador can interpret 20 languages and 42 dialects, including historically MT-challenging languages like Korean and Japanese. Up to four earpieces can be connected to a single smartphone, and it can also connect to a speaker to broadcast to an audience. The Ambassador can be reserved now but won’t available until March 2020.

Pixel Buds to offer access to instant translations

Also set for release this year will be Google’s Pixel Buds. A direct competitor to both the Ambassador and Apple’s Airpods, the in-ear phones will offer hands-free interactivity with Google Assistant. It means users will be able to access live language translation on-the-go in combination with the new Interpreter mode. They also offer a number of nice little touches, including the ability to detect the noise level of the environment and adjust the phone’s volume accordingly, making life easier when trying to communicate in a crowded room.

Machine translation to blame for Facebook’s blushes

However, things aren’t all smooth sailing. In a warning for any organization that uses automated MT software, Facebook’s internal machine translation hit the spotlight for the wrong reasons during Chinese leader Xi Jinping’s visit to Burma in January. After its software mistakenly translated the president’s name from Burmese into an offensive English term, Facebook put the error on its system not having Mr. Jinping’s name in its Burmese database and guessing at the translation. This caused severe embarrassment for the company, since China represents its second most lucrative market after the US. It is not the first time Facebook’s Burmese translations have caused issues: in 2018, it was translating anti-Rohingya posts that advocated killing Muslims into English sentences such as “I shouldn’t have a rainbow in Myanmar.”

Skype translation issues spark privacy concerns

Another concerning aspect of the MT boom is the risk to user data. For example, Microsoft workers in China were able to listen to conversations made using Skype’s experimental live text translation feature with no cyber-security protocols in place, according to the Guardian. A whistle-blower claims that the lack of security measures meant recordings were accessed through accounts with identical passwords, with nothing to guard against criminal or state interference and almost no employee vetting. Last August, Vice Magazine reported that Skype audio clips were being transcribed by humans to help improve the app’s algorithm, with no warnings in either Skype’s or Microsoft’s privacy policy that conversations might be used in this way. In response to the Guardian outlining these data risks, Microsoft said it had ended the relevant grading programs and moved any remaining grading operations into ‘secure facilities’ outside of China.