An iFly Tek robot translator performs interpretation at the opening ceremony of RoboCup, an annual international robotics competition, in July 2015 in Hefei, Anhui Province (XINHUA)

Having recently won 60 consecutive Go games, AlphaGo, the artificial intelligence (AI) program created by Google, has once again raised the question of whether AI will one day surpass human intelligence. An updated version of the program registered as a player on online board game platforms in China and defeated a series of elite players.

Machine translation (MT), an implementation of AI, is gaining momentum in China. At the China Language Service Industry Conference, organized by the Translators Association of China (TAC) in Beijing in December 2016, tech company executives and computer scientists discussed the possibility of MT replacing human translation.

"I don't think MT will replace human translation. This is probably not the future trend and is not what we as researchers hope for. The ideal situation should be to use machines to aid translators and improve the speed of translation," said Xiao Tong, a professor of Northeastern University located in Shenyang, Liaoning Province, who specializes in MT research.

Developments

MT has made strides in China in recent years, with both large tech companies such as Baidu, Tencent and Alibaba, and smaller ones coming up with products.

Baidu, for instance, launched Baidu Translate in 2011 and incorporated neural machine translation (NMT) into the service in 2015. Compared with previous statistical translation methods, which separated sentences into components to be translated, the system consists of a large neural network, which sees each sentence as a whole and is therefore able to derive a more fluent and accurate translation.

Now Baidu Translate supports 28 languages, and the company has also developed an app which translates written and spoken words and produces output in readable and audible forms. Using the app, people lacking a common language can communicate without the help of an interpreter.

E-commerce giant Alibaba, meanwhile, started to delve into MT in 2012 in a bid to facilitate its cross-border e-commerce business.

In 2015, Alibaba acquired 365 Translation, China's largest online human translation platform to help its vendors overcome linguistic barriers in cross-border e-commerce and thereby accelerate their globalization strategy.

Chinese IT company iFly Tek launched the Xiaoyi Translation Machine in November 2016, which is capable of interpreting between Chinese and English, and also around the same time unveiled a product capable of simultaneous interpretation for conferences.

With its MT solutions winning awards at multiple international competitions, the company has taken the lead in MT research in China since 2014.

Wang Shijin, Deputy Director of iFly Tek Research Institute (Beijing), said all of his company's MT products apply NMT technology, and the firm's advanced speech recognition and synthesis technologies provide the foundation for the success of its machine interpretation products.

Baidu CEO Robin Li unveils the company's AI system, Baidu Brain, and demonstrates its translation capabilities on September 1, 2016 (CFP)

Challenges

China's domestic tech companies still lag behind their foreign counterparts, such as Google, in MT.

"Most domestic MT products are younger than Google Translate, and they are not on the same level as Google in terms of algorithms. Therefore, it will be difficult for domestic companies to catch up with Google," said Zhang Xuetao, Director of the Language Service Competence Assessment and Training Center of TAC.

Google launched an NMT system in September 2016, saying it will reduce translation errors in Google Translate by between 55 percent and 85 percent.

"Many translation companies in China use Google Translate to do translation and employ human translators to proofread the results. Their translation speed has greatly improved," Zhang added.

Fu Hebin, a researcher with the Center for International Communication Studies at the media conglomerate China International Publishing Group, said he used MT in his previous job as a Chinese-English translator. He chose Google Translate because Google's MT was cutting-edge even before it embraced NMT.

"At present, MT cannot deliver satisfactory results in complicated contexts or in difficult professional fields. To further improve MT results, breakthroughs need to be made in big data and deep-learning technologies," said Wang.

Deep learning is the fastest-growing field in machine learning. It uses many-layered neural networks to learn levels of representation and abstraction that make sense of data such as images, sound and text.

Wang added that his company will work together with domestic universities and research institutes to overcome technological difficulties in MT research.

A machine translation forum during the China Language Service Industry Conference on December 23, 2016 (COURTESY OF TRANSLATORS ASSOCIATION OF CHINA)

Mutually beneficial

Zhang believes MT will not replace translators but increase demand for such professionals.

"MT requires competent translators to do post-editing. A lot of job opportunities will be created in the process. In the meantime, MT has improved the efficiency of translation and lowered the price of translation services, which will inevitably increase the demand for translators. For instance, previously, cross-border e-commerce platforms wouldn't consider using human translators to translate their websites owing to the high price and slow speed. However, in the future, they might consider using human translators to post-edit machine translation results as the price lowers and translation speed improves," said Zhang.

Fu noted that while competent translators can leverage MT, the technology is likely to mislead people who lack well-developed capabilities in the source and target languages. Furthermore, he believes that MT can sometimes deliver better translations than humans thanks to its big data technologies.

Zhang said MT will also make the job of consecutive interpretation much easier. Previously, interpreters needed to take notes to facilitate their interpreting. Now, with the help of MT speech recognition systems, they can see machine-translated versions of speeches, which they only need to polish.

Wang said both human and machine interpreters have shortcomings. Human interpreters are slower, while machine interpreters are less accurate. He believes a combination of the two approaches will improve the speed and accuracy of interpretation.

Copyedited by Chris Surtees

Comments to jijing@bjreview.com