Machines cannot take away translators’ job completely, but it plays its role in providing translation service by “matching” the right words.
How big is translation market? Zhou Feng, CEO of NetEase Youdao （网易有道）, estimated it is a 40-billion RMB market. According to the 2016 China’s Language Service Industry Development Report, the value that China’s language service sector created in 2015 was about 282.2 billion yuan, an increase of 79% over 157.6 billion yuan in 2011. The average annual growth was about 19.7%. By the end of 2015, there were about 72,495 companies providing language services or language-related services in the country. Translation is indeed a big market, but not until recently did it begin to embrace the Internet.
NetEase Youdao has rolled out human translation services in 2012, tapping into the translation market by offering language service tool. Traffic is the advantage of tool products, but how to cash in on the traffic must be considered after the product is mature. Making money from ads is the most direct method, but the problem is that the ceiling is not high. Naturally, language and service products are connected. In addition to the word search and machine translation, there is also a need for more accurate translation services. Therefore, it is not surprising to see Youdao expanded its business from online dictionary to translation.
When it comes to development logic, Youdao Translation follows a path of “C to B”, that is, after a complete set of service system is established by serving consumers, Youdao will then serve the medium and small businesses and plan to cooperate with large businesses in the next stage. Liu Ren, Vice President of Youdao’s marketing department, mentioned that traditional translation companies focused mainly on B end and ignored C end users because of low demand and unit price, failing to offer specific solutions. The C end to B mode can be seen as prioritizing low-frequency translation demand over high-frequency. Low as the C end users’ demand is, their demand and willingness to pay is strong. Generally, they, looking for professional translation solution, can only seek translation service on platforms like Taobao only to find that the translation quality varies each time. Of course, the traffic that YoudaoDict has accumulated is a precondition for this logic.
According to the needs and content, there are two types of services currently available. One is fast translation, another document translation. With 0.02 yuan/word as the charge, the former applies to users who set a tight deadline and everyday content like business correspondence. The latter applies to professional content or relatively large volume, whose charges depend on the content. According to the data provided by Youdao, from 2012 to date, the total number of Youdao Translation orders has exceeded 1 million, with 1000 orders per day and a nearly 100% increase of human translation orders per year.
Similar to the traditional translation company’s business model, Youdao Translation outsources orders to over 3000 translators. For fast translation content, translators can “grab orders in the background”. For document translation, project managers will “assign translation task” in accordance with the translators’ professional background. And project managers are responsible for the quality control of translation results and customers’ feedback. For B end, Youdao has launched the human translation API service. Specifically, Youdao cooperates with B end customers and sells its translation service by offering API interface and embedded SDR model.
Overall, on the premise of similar service model, compared with the traditional translation services, the advantage of Youdao is mainly at the technical level. That is, its human translation is not “pure human labor”. Rather, based on NMT (Neural Machine Translation) technology, Youdao adopts an approach of “man-machine translation”, that is, after the machine finishes the first draft, translators do follow-up examination and make improvements, bringing down the overall price by 50% compared with human translation.
The other reason why the human translation business is a rigid demand is that much of the free machine translation does not make sense. So, with the emergence of neural network machine translation, will human translation be replaced? At least the Google case indicates there is still a long way to go. Some translation scenes emphasize “faithfulness, expressiveness, and elegance”, but at present NMT technology can only meet the requirement of “faithfulness”. Although it does improve the accuracy of content translation, human engagement is needed if translation involves “expressiveness and elegance”. Therefore, at this level, machine cannot replace human entirely, instead it plays its role in providing translation service by “matching” the right words.
On top of Youdao, Baidu, Jinshan (金山）, and the voice-recognition technology company— IFLYTEK（科大讯飞） all tapped into the translation market by offering language service tool. In addition, 36 Kr wrote that Wormhold （虫洞）Translation, modeled on Taobao, built a platform to serve translators and customers. As a whole, the translation market has become white hot due to low threshold for standardization. Of course, compared with the interpretation , a greater demand, sufficient supply of translators, and the lower cost is also add to the fierce competition.
This article by Qi Si originally appeared in 36kr and was translated by Pandaily.
Click here to read the original Chinese article.