Flitto's language data makes AI translation more accurate

Simon Lee, CEO and founder of Flitto, speaks during an interview with The Korea Times at the firm's office in Seoul, Nov. 6. / Courtesy of Flitto

Startup diversifying income sources through video translation in YouTube

By Jun Ji-hye

Artificial intelligence-powered (AI) translation has become one of the major interests of IT companies here and abroad as they seek to offer their customers more advanced services.

Google, Microsoft and other tech giants have been striving to create more elaborate translation services, but machine-generated results were sometimes unsatisfactory and even ridiculous, reducing the level of customer satisfaction.

This market trend offered an important opportunity for Korean startup Flitto to increase its sales and clients rapidly as it has collected vast amounts of language data using accurate human translations since its 2012 foundation as a translation crowdsourcing platform.

Simon Lee, CEO and founder of the Seoul-based startup, said in an interview with The Korea Times that his firm has been able to sell its language data to a number of companies as they needed the data to train their machine translation systems.

"An increasing number of companies have been going into AI-powered translation services, and they purchased data to advance their services," he said. "The number of our new partners continues to increase."

The firm's sales have jumped from 400 million won ($358,000) in 2015 to 1.4 billion won in 2016, 2.5 billion won in 2017 and 5 billion won as of August, with Lee expecting this year's sales to reach 7 billion won.

Most of its revenue comes from the sale of language data, called "corpus," Lee said, noting that the total corpus sales volume is expected to reach 30 million this year from 2.1 million in 2015.

He said machines need to learn a lot to improve their performance, noting that unsatisfactory results produced by machines resulted from a lack of accumulated data.

"It is almost impossible for machines to produce language data on their own as there are an enormous range of languages, grammars and pronunciations," he said.

There are two methods of collecting language data _ one is collecting it through translations from a group of experts, and the other is using collective intelligence of many random people, Lee explained.

The first method could create more accurate results but is expensive and time-consuming, while the other can be done more quickly at lower prices but issues regarding quality could arise.

Flitto has collected language data through the collective intelligence of people online.

"We have made efforts to resolve problems related to the collective intelligence method," he said. "We ask people to repeatedly correct errors to find the answers eventually. Our quality control system works satisfactorily."

This has enabled the startup to sharpen its competitiveness and offer high-quality language data at one-10th the price of other companies, he said.

Like many other startups, Flitto also experienced the "Death Valley," a term referring to financial difficulty startups face three to four years after their foundation.

The Death Valley comes when startups fail to reach a goal they set at their establishment.

When experiencing harsh times, he and his employees were doubtful whether they were headed in the right direction.

"The number of employees had increased to 80 but fell to 40 during the Death Valley," he said. "At the time, I even made an apology in front of everyone."

The only way of overcoming the harsh times was unwaveringly moving toward the goal, he said, noting that the firm's sales began to pick up in 2015 amid the development of AI technology and growing demand for data.

One of the startup's clients is Japan's NTT DoCoMo, which is developing its translation system ahead of the Tokyo Olympics. Other clients include Baidu, Tencent, Microsoft and Expedia.

The firm's service offering 24 languages is also popular among individual users as it has secured 9.5 million users in 173 countries.

Creators on YouTube comprise a large proportion of individual users as they actively utilize Flitto's translation service in producing subtitles on their video clips to captivate overseas viewers.

Lee said he is giving a lot of attention to YouTube, saying his firm can diversify sources of income on that borderless platform.

"Not only companies such as entertainment management agencies but also individuals have created their YouTube channels, uploading video clips," Lee said. "This is raising demand for video translation services as they want to provide subtitles in other languages to attract more viewers around the world."

There have not been many companies offering video translation services, he said, noting Flitto is able to offer almost 100 percent accurate video translation service through its quality control system.

"I expect this market to grow rapidly," he said.

'Machines cannot beat human translators'

Lee said work to collect language data using human translation will need to continue as AI-powered translation will still not "perfect."

"AI-powered translation services have been improving but nobody can say they are 100 percent accurate as languages are infinite, with newly coined words continuing to emerge," he said. "The goal is to continue to collect language data using human translation and use such data to train AI so both sides can evolve together."

He said his firm plans to offer a wider range of languages. Arabic, Vietnamese and Hindi will be added this year.

The startup is also working to collect voice data to meet growing demand amid the emergence of a number of AI voice recognition services.

"We are keeping abreast of trends," he said.

The firm currently has 50 employees in Korea and 20 in its branches in China and Japan.

Lee said the startup is considering opening another branch in the United States where it has a number of partners.

"The Middle East is another place I want to open a branch as we have many users there," he said.

Lee, who studied business administration at Korea University, worked at SK Telecom, the nation's top mobile carrier, before setting up Flitto.

Site Map E-paper Subscribe Register LogIn
Site Map E-paper Subscribe Register LogIn