Go to Contents Go to Navigation

Humans beat AI in language translation

All News 17:23 February 21, 2017

By Kim Han-joo

SEOUL, Feb. 21 (Yonhap) -- Humans beat artificial intelligence (AI) language software in translation at a high-profile battle held in South Korea on Tuesday, though experts forecast the cutting-edge technology is improving at fast rate and may reach human-level accuracy soon.

Hosted by the International Interpretation Translation Association, a group of four professional translators competed against three AI-powered programs provided by U.S. Internet giant Google Inc., South Korea's top Internet provider Naver Inc., and leading automated interpretation company Systran International.

Professional translators in a competition against "AI translators" in Seoul on Feb. 21, 2017. (Yonhap)

Both sides were tasked with translating random English articles -- literature and non-literature -- into Korean and other Korean articles into English. A total of 50 minutes were given to translate the texts and the translated works were evaluated by two professional translators.

Without revealing the identities, the organizers said the four professional translators scored an average of 25 out of 30 in translating Korean into English, while the AI software scored between 10 and 15.

Automatic translation programs using machines have improved recently thanks to the introduction of the technology called the neural machine translation (NMT). Major tech giants including Google and Naver rolled out various NMT services last year.

"NMT machines competed against professional translators who do this for a living," said Kang Dae-young, director of the association, adding that the final works of the professional translators were far better than the NMT machines.

The NMT system is based on a deep learning framework that learns from millions of examples from over 100 different languages. Unlike previous machine translation that was adopted 10 years ago, the new system considers an entire sentence as one unit. Previous systems independently translated words and phrases within a sentence.

An AI-based translation software in a competition against professional human translators in Seoul on Feb. 21, 2017. (Yonhap)

"The problem of NMT translation was that it looked that the machines were unable to understand context," the organizers said, adding that 90 percent of NMT-translated texts were grammatically awkward.

Systran, however, said the gap between the two will soon be shortened as the technology provides computers with the ability to learn without being explicitly programmed.

"I would say the technology is at the stage of elementary school at the moment but it will improve to the level of high school and college just in one or two years," Kim Yoo-seok, director of Systran, said during a forum held by the association.

Google Inc. is the leading provider of the AI-based translation platform by becoming the first to introduce its NMT system last year, which significantly improves translation quality and reduces errors.

A battle between professional translators and artificial intelligence (AI) language software was held in Seoul on Feb. 21, 2017. (Yonhap)

With the aim to catch up with global industry players, Naver also launched the country's first automated translation app called Papago that translates between four languages -- Korean, Japanese, Chinese and English.

Local software firm Hancom Interfree Inc. has also rolled out a translation app called "Genie Talk" that is capable of translating between Korean, Japanese, English, Chinese, Spanish, French, German, Russian and Arabic.

The battle comes amid heightened interest in AI following the high-profile matches between Google's AI program AlphaGo and South Korean Go champ Lee Se-dol last year.

There have been growing calls that South Korea should revamp its R&D strategy to catch up with other global powerhouses in the AI field. The Seoul government said it plans to raise the competitiveness of AI technology to the same level as an advanced country in 10 years.

Humans beat AI in language translation - 4


Send Feedback
How can we improve?
Thanks for your feedback!