A team of Microsoft researchers has managed to create an Artificial Intelligence (AI)-powered machine which can translate sentences of news articles from Chinese to English with the same quality and preciseness as humans.
Researchers from the company's Asia and US labs said their system was able to achieve as well as humans when translating on a set of news stories, which were commonly used for tests.
The team’s aim was to prove that its systems could perform as good as a person when it used languages such as Chinese and English, for which there is a lot of data, Arul Menezes, Partner Research Manager of Microsoft's machine translation team, said.
"Given the best-case situation as far as data and availability of resources goes, we wanted to find out if we could actually match the performance of a professional human translator," Menezes said.
Hitting human parity
In order to make sure the results were accurate and similar to what people would have done, the team also hired external bilingual human evaluators who compared Microsoft's results to two human translations.
"Hitting human parity in a machine translation task is a dream that all of us have had. We just did not realize we would be able to hit it so soon.”
"Hitting human parity in a machine translation task is a dream that all of us have had. We just did not realize we would be able to hit it so soon,” Xuedong Huang, Technical Fellow in charge of Microsoft's speech, natural language and machine translation efforts, said.
Three research teams in Microsoft's Beijing and Redmond, Washington, research labs worked together so that they could make the system more precise and produce results more similar to humans.
"Much of our research is really inspired by how we humans do things," Tie-Yan Liu, Principal Research Manager with Microsoft Research Asia in Beijing, said.
The team used a dual-learning method. This meant that every time they sent a sentence through the system to be translated from Chinese to English, they also translated it from English to Chinese.
This is similar to what people may choose to do to make sure that their automated translations are accurate. This also allowed the system to refine and learn from its own mistakes.
The translation milestone was especially satisfying because of the possibilities it has for helping people understand each other better, Huang, who also led the group which recently achieved human parity in a conversational speech recognition task, said.
“The pursuit of removing language barriers to help people communicate better is fantastic,” he said.
Machine translation problems not yet solved
Machine translation is an issue researchers have worked on for many years. Many people believed machines could never be as good as humans during much of that time.
However, the researchers have said that this discovery does not mean that the problem with machine translations has been solved. Menezes said the research team can apply the technical discoveries they made in this test to Microsoft’s commercially available translation products in multiple languages.
This will open up new paths for more accurate and natural-sounding translations in other languages as well as texts with more complex or special vocabulary.