- Baidu unveiled an artificial intelligence-powered tool that can translate languages in real time on Wednesday.
- The simultaneous translation feature has been trained on two million pairs of English and Chinese sentences.
- The Chinese tech giant is posing a challenge to Google which last year unveiled its real-time translation tool.
Internet giant Baidu unveiled an artificial intelligence-powered tool on Wednesday that can translate English into Chinese and German in real time, in move to challenge a rival product from Google.
So far, many translation services online allow users to say or write a sentence, but there is a lag before it proceeds to serve up a translation. In contrast, Baidu's so-called simultaneous translation tool allows sentences to be translated almost instantly — much like a digital version of what interpreters do.
The product is a breakthrough for Baidu which has been investing heavily in AI for the past few years, hoping to build the future of its business on the technology.
The so-called natural language processing — the ability for machines to understand human speech — has been developing at a rapid pace. It is seen as a key technology that can be integrated across a number of products, especially voice assistants.
For now, Baidu's translation tool is only able to translate English to Chinese and vice versa, or English and German.
AI tools require huge amounts of data sets to learn.
Baidu trained its AI on two million pairs of English and Chinese sentences, Liang Huang, principal scientist at Baidu, told CNBC by phone on Wednesday. This has allowed Baidu's digital interpreter to do near real-time translation from two languages that have very different sentence structures.
Baidu used the example of this sentence in its blog post on Wednesday: "President Bush meets with Russian President Putin in Moscow."
In Chinese, the verb "meets" is at the end of the sentence. But when translated into English, it becomes the third word in the sentence, as is appropriate in that language. Thanks to the training with the data sets, Baidu's tool is able to predict the word that comes in the English sentence, even before the word is spoken.
"We train our system to predict the English side given the Chinese side prefix," Huang told CNBC.
"You learn from that data that if Bush or any U.S. president is ever in Moscow, he is likely meeting with somebody," the scientist added.
This would not be a problem for non-simultaneous translation as the time lag allows for assessing the sentence structure before making the translation. But with real-time interpretation, there cannot be a delay.
The translation tool can also be adjusted for latency, which means a user can set how much lag there is between a word being spoken and its translation. The higher the lag, the better the translation for some languages.
It currently supports Chinese to English, but Huang said it is "language neutral" and will eventually be able to translate other language pairs too.
Baidu is China's largest search engine and for that reason has often been compared to Google. Its latest product comes over a year after Google unveiled the Pixel Buds, a set of wireless headphones that it claims can do live translation.
Huang said Baidu is looking to integrate the AI interpreter into its Wi-Fi translator, a product it unveiled earlier this year which is both a portable internet hub and translator. The company will also use this technology to translate speeches at its annual Baidu World Conference on November 1 in Beijing, China.
But Huang said the company does not have plans to bring it to other products such as its voice assistant.
"We are still working hard on making available to general public but we don't know when that will happen," Huang said.
Baidu acknowledged the limitations of its translation tool however, and said that it would likely not be ready to replace human interpreters or take the place of a real-life translation at a major political event soon.
"We hope this AI technology will reduce the burden on human interpreters. But we are not tying to replace human interpreters, who we will continue to be depend on their services for many years to come especially for high stake occasions which require consistent and more precise interpretation," Huang told CNBC.