pio3 - Fotolia

Microsoft’s Chinese-English translation system achieves human parity

A China-US research team at software powerhouse Microsoft has developed an AI-powered system that can translate Chinese news articles into English as well as humans do

A team of Microsoft researchers from China and the US have developed an artificial intelligence (AI) powered translation system that can translate Chinese language news articles into English with human accuracy.

The researchers said they were able to achieve human parity on a test set of news stories released by an industry and research group at a machine translation conference in 2017.

Huang Xuedong, Microsoft’s technical fellow for speech, natural language and machine translation, said the achievement was a major milestone in one of the most challenging natural language processing (NLP) tasks.

“Hitting human parity in a machine translation task is a dream that all of us have had,” Huang said. “We just didn’t realise we’d be able to hit it so soon.”

To improve accuracy, the researchers used a combination of methods to train their system. Besides deep learning techniques, they also used dual-learning methods that mimic the way people correct machine-translated text, as well as joint-training where English and Chinese text is translated back and forth to improve overall performance.

But Microsoft has said its translation system has not been proven on real-time news stories, even as it is looking to apply its technical breakthrough to its translation products down the road.

Machine translation has been a difficult problem to solve in NLP, given that there are usually different ways of saying the same thing or describing a concept due to language and cultural nuances.

“Machine translation is much more complex than a pure pattern recognition task,” Zhou Ming, assistant managing director of Microsoft Research Asia and head of natural language processing, said in a company blog post. “People can use different words to express the exact same thing, but you cannot necessarily say which one is better.”

Read more about AI in APAC

  • Malaysia’s RHB Bank launches chatbot to help consumers apply for personal loans, making it the latest bank in Southeast Asia to make use of chatbots to improve customer service.
  • New Zealand’s Auckland Airport is testing an AI-powered avatar at the arrivals area to answer biosecurity questions from travellers.
  • Despite the touted merits of human-machine collaboration, business leaders in APAC are divided on the impact of AI-powered machines on the future of work.
  • Forward-looking organisations in ASEAN are embracing AI, but uneven access to connectivity and a lack of skills and understanding of the technology are holding back wider adoption.
  • Alibaba has set up a joint research facility at Singapore’s Nanyang Technological University to develop artificial intelligence-based technologies in retail, transportation and healthcare.

Chinese technology bigwigs have also achieved notable breakthroughs in NLP. In September 2015, Tencent’s Dreamwriter, an AI-powered writing robot that automatically writes news articles from public data sources and press releases, published its first financial report in Chinese.

Since then, the robot has been able to learn on its own and develop article templates based on reporting routines and rules for a wider variety of stories related to topics such as motoring, film and even social issues.  

In July 2017, China outlined its vision of becoming a world leader in AI by 2025. Under its national AI development blueprint, the Chinese government plans to grow the size of China’s AI industry to RMB150 billion ($23.8bn) over the next two years, and RMB400 billion by 2025.

A $2.1bn AI technology park in Beijing is also expected to play host to more than 400 companies specialising in cloud services, 5G services, biometrics and high-performance computing when it opens in 2023.

Read more on Artificial intelligence, automation and robotics

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.

-ADS BY GOOGLE

SearchCIO

SearchSecurity

SearchNetworking

SearchDataCenter

SearchDataManagement

Close