The Journal The Authority on Global Business in Japan

CEO of Gengo, Matt Romaine

CEO of Gengo, Matt Romaine

A recent UK publicity stunt for Google Translate, a machine translation service by Google Inc., went drastically wrong.

A sign was emblazoned with “Good Evening, Old Street” on the left, and an Arabic translation on the right. But the Arabic was backward, with breaks in the wrong places.

The mistake highlights today’s big problem with machine translation: It doesn’t do its job properly. Adding to the headache, the “translated” sign in central London, created by an algorithm, is unaware of the cultural sensitivity of its work.

Wherever one lies on the political spectrum, surely it is understood that not all Arabic speakers appreciate Western interventions in and attitudes to the Middle East.

CULTURALLY CHALLENGED
Artificial intelligence that is as capable of translating language as a bilingual human can is still a long way off. Cultural awareness is one big reason.

Peter Durfee, director at the Nippon Communications Foundation, has been translating for decades and believes machines have a lot of improvements to make.

“You need to present the information in the source text; you need to think about effective ways to present that same information in the target text. And this is as far as even the best conceivable machine translation system can go.

“But you also need to consider the motivations of the original author—Am I translating a speech where . . . [Prime Minister Shinzo] Abe is pushing flexible constitutional interpretation, or Haruki Murakami is pushing the merits of playing jazz albums to cats?”

The expectations of readers also have to be taken into account.

Academics see little chance that machines will fully replace translators. Carl Benedikt Frey and Michael A. Osborne in 2013 calculated that a translator would have a 38 percent chance of the work they did being replicated by machines.

“With a growing corpus of human-translated [digitized] text, the success of a machine translator can now be judged by its accuracy in reproducing observed translations,” they wrote in their report, The Future of Employment: How Susceptible are Jobs to Computerization?

The fact that machines now can, to a degree, learn from humans at a faster pace has practical applications. The academics cite United Nations documents, which are translated into six languages, as an example.

Such multilingual texts help Google improve its algorithms and translate more accurately. But algorithms are not humans.

“If you just want to grasp the meaning of a document, get a rough idea of the meaning, then machine translation is fine,” says Kaori Sasaki, the founder of Unicul International Inc., which offers services including translation, interpretation, and conference organizing.

“But if you want to get more from a document, you need a higher-quality service.”

What’s more, many languages, including Japanese, simply do not have a big enough presence on the global cultural landscape to provide the sort of translated documents that the robots are hungry for.

“To be truly successful, a machine translation engine has to have a huge corpus of matching terms thrown into it over the years,” says Durfee.

“For some language pairs, this has progressed much farther than in the Japanese [language] sphere: Everything published by the European Union apparatus, for instance, has to go into a whole raft of languages, and you end up with hundreds of millions of (public domain) words that can be stuck into Google’s and Bing’s analytical tools for parsing.”

Re-translating between two languages exposes the limitations of translation machines.

Re-translating between two languages exposes the limitations of translation machines.

COLLABORATIVE CODE
Gengo, Inc. has found a way to get the most out of machines and offer a human service.

The company, which operates the Gengo.com website, introduces its around 16,000 human translators to clients through an online database. It offers services in 37 languages and ratings for staff.

“We don’t believe it’s possible to completely replace human translation with a machine—not only because of the creativity involved, but also because of the evolving nature of languages,” says company CEO Matt Romaine.

Gengo’s database gives it a key advantage on speed. Most of the jobs are started within two hours and completed within three.

“New publishing and content distribution models dominated by names like Huffington Post and BuzzFeed—a domain we internally call casual media—also rely on Gengo to expand their global reach,” Romaine explains.

He sees machine translation and artificial intelligence as a benefit rather than burden.

“Machines can assist to make the process more productive,” Romaine says. “From searching for terms, to checking how many times certain words have been used, perhaps to not sound redundant, machines will only be tools for assisting human translators.”

But there are limits to how far this assistance will go. “Clients need to provide context that only a human could understand. For example, if someone wanted a translation of ‘eats shoots and leaves’—is that referring to a koala or a criminal?”

Durfee sees a further use for machines. “One type of translation that has increasingly gone to the machines is massive-scale triage in the discovery stage of big legal cases,” he says.

Without the need for anything literary or to appeal to a public, machines can fish through reams of documents in a different language and roughly summarize them.

Key information can then be sent to the human to get the job done properly. Such scouring was never rewarding for people in the first place, and few translators would regret no longer getting to search for the proverbial needles in haystacks.

MACHINE–HUMAN BAZAARS
It is beyond the literary and the promotional that machine translation has its greatest potential. Google’s sign in London still got its message across. Nobody struggled to get the meaning, even if the language was clumsy.

And, with the Tokyo 2020 Olympic and Paralympic Games coming up, Japan is working toward machines that can bear the brunt of the translation burden.

This year, the government will spend around ¥1.38 billion on developing automatic voice translation software for 10 languages. The idea is to cover the needs of the vast majority of visitors to Japan and change the nation’s difficult-to-travel image.

In the private sector, Panasonic Corporation has begun tests of its automatic translation service in collaboration with Mori Building Co., Ltd.

“By offering the devices at information desks, Panasonic will be able to confirm their appeal and functionality for users while also identifying areas for improvement,” the company says.

“For Tokyo 2020, translation is a must as we welcome visitors from around the world,” says Sasaki.

“And machine translation for this is amazingly advanced. Most of the daily necessities for translation, such as informing people with allergies of ingredients, can be done by machines.”

Companies seeing the possible success of products such as Panasonic’s may, in the future, be tempted to clean up their budgets by ditching translators for computers.

Sasaki argues that this means translators today have to do more to show companies why humans are needed: “Clients really need to be educated to choose which kind of services they want.”

It is beyond the literary and the promotional that machine translation has its greatest potential.