According to Bloomberg, thousands of translators walked into offices across the Chinese mainland on December 6, 2016, translating countless foreign language pamphlets, letters and technical manuals into Chinese. The marathon lasted for a whole month, with 15 hours of translation work a day. The client providing the materials can obtain the expert-level Chinese translation of the original at the contract price. But Baidu, the company that organized the mass translation effort, got something more valuable: millions of English-Mandarin word pairs to train its online translation engine.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include information robot, editing robot, writing robot and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
In artificial intelligence (AI), the Chinese company is facing competition from tech giants Alphabet, Facebook, IBM and Microsoft. In February 2017, China’s National Development and Reform Commission appointed Baidu to lead a new AI lab, a sign that the Chinese government believes the company has the qualities of a “national champion” in the field. Robin Li, co-founder and CEO of Baidu, said on the day of the AI Lab’s launch in February that Baidu has invested more than 20 billion yuan in research and development over the past two and a half years, most of it in AI.
But national interest is not the main motivation for Baidu to invest in AI: Baidu’s revenue growth fell about 6 percent in 2016, compared with an average of more than 30 percent in the previous three years. Baidu’s total revenue was 70.5 billion yuan in the fiscal year ended December 31, 2016, of which search advertising accounted for the majority, but it is under siege from local rivals. In September 2016, EMarketer released a report showing that Alibaba has surpassed Baidu to become the new leader of China’s digital advertising market. Baidu hopes AI will help it regain its top spot in search and ensure its new investments succeed.
That matters because baidu, now 17 years old, has tried to diversify, with mixed results. Daily visitors to Baidu’s group-buying site Nuomi fell 59 percent in the 12 months to February 2017, and its food delivery service Baidu Waimai fell to third place, said Natalie Wu, an analyst at China International Capital Corp. Moreover, iQiyi, a Netflx-like streaming video service, needs to spend 12 billion yuan this year to keep its content rich, despite its huge popularity, according to Analysts at China Renaissance Capital.
These faltering efforts mean that Baidu’s push for AI is becoming increasingly important. “The mobile Internet era has come to an end, and we will actively invest in AI,” Li said in an interview on March 10. I think it will benefit a lot of people and support industrial transformation.” In January, Baidu named Qi Lu, a Microsoft executive, as chief operating officer to reshape the company around technologies such as deep learning, augmented reality and image recognition. In addition, Andrew Ng, a Stanford academic who worked on Alphabet’s deep learning team, joined Baidu in 2014 and is now chief scientist.
Under Ng, Baidu’s AI team — spread out in Beijing, Shenzhen, Shanghai and Sunnyvale, Calif. — has expanded to 1,300 people in its research LABS, with hundreds more expected this year. “A lot of things were invented in China and a lot of things were invented in the US,” says Mr Ng. Through the people of these two countries, we can see the latest trends.”
When the Sunnyvale Research Center opened one day in May 2014, Ng and his senior associate, Adam Coates, sat in front of a blank whiteboard and identified their first project. After sketching out the possibilities and challenges, they used language recognition as a basis for building a range of other products. By mid-2015, the 50-person team had launched a product called Deep Speech, which could decipher many conversations in English. Instead of separating phrases word by word, the software parses large amounts of language data and then extrapolates patterns, a process known as deep learning. The system can transcribe speech more accurately than traditional translation engines, which rely on vocabularies and phonetic symbols, because it takes into account the context of words to determine their meaning.
One problem that has plagued the research team, however, is the transfer of English words and names from other languages over time. “If you want to say, ‘Play Tchaikovsky,’ the software will give you a meaningless answer,” Coates says. For this reason, we call this phenomenon the ‘Tchaikovsky problem.'” Rather than simply adding “Tchaikovsky” to the system’s vocabulary, Baidu’s programmers helped Deep Speech learn to understand the words itself. This involves extracting more data to help the system understand better in context.
Dr Zhao Shiqi, deputy head of Baidu’s natural language processing division, recalls that when he was a computer science student at Harbin University of Science and Technology, only 2m English-Chinese word pairs were available for computer translation. Now, Baidu word pairs have reached 100 million. Still, that’s a far cry from Alphabet’s 500 million word pairs. To close the gap, Baidu has resorted to ancient tactics, even if it is employing human tactics to solve the problem. The company now regularly hosts year-round and staged manual translation marathons, like the one in December 2016. Baidu offers prizes such as smartphones and water purifiers. The data collected will help enhance the performance of Baidu’s translation engine and help further development of Deep Speech.
The software developed by baidu’s Sunnyvale team made its debut in July 2016, when Baidu quietly launched TalkType, a voice-centric Android keyboard app. The technology has already been integrated into other products, including siri-like personal assistants. The machine learning technology that Baidu has already infused into Deep Speech is helping it bring intelligence to other products. Baidu, for example, showed off its voice-controlled robot xiaoyu Jiayin at the CES in Las Vegas in January.
Baidu’s online portfolio gives it one of the largest and most detailed sets of consumer data in China, which, in theory at least, could give it a huge advantage in developing AI-powered products and services. With sticky rice and take-out, the company knows what Chinese families buy and eat. With the help of Ctrip, the world’s second-largest online travel site, Baidu also knows where people like to go on vacation. Every month 665 million smartphone users surf baidu’s mobile portals and apps, and 341 million use Baidu Maps to find destinations. “It is a mistake to think of AI as a product,” said analysts at HSBC Holdings Plc. “It underpins and gives the product more power.”
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
The new AI products contributed little to Baidu’s results, but the company’s nascent expertise in this area could help it gain dominance elsewhere. These are areas where Baidu already has a foothold and is pushing it to new levels, such as cloud computing and driverless cars. “In the next three to five years, another Baidu could emerge in these areas,” zhang Yaqin, president, said of Baidu’s $60.2 billion market capitalisation. Now, it’s time to place your bets!”