当前位置：网站首页>With an average daily call volume of more than 1.3 billion times, aridamo Institute developed the world's first real-time translation live broadcast
With an average daily call volume of more than 1.3 billion times, aridamo Institute developed the world's first real-time translation live broadcast
2020-12-07 19:22:10 【Aliyun yunqi】
brief introduction ： in recent years , Live e-commerce is blooming everywhere , But most of them are live Chinese in China . If you want to buy a product recommended by a foreign e-commerce anchor , What to do if you don't know the language ？ This problem has been solved by Alibaba （ Hereinafter referred to as “ Ali ”） Conquered , Aliexpress is an e-commerce platform for overseas consumers all over the world , It's the same as Taobao , Also for all end consumers , But the difference is that express is global 200 E-commerce platforms in many countries and regions .
in recent years , Live e-commerce is blooming everywhere , But most of them are live Chinese in China . If you want to buy a product recommended by a foreign e-commerce anchor , What to do if you don't know the language ？
This problem has been solved by Alibaba （ Hereinafter referred to as “ Ali ”） Conquered , Aliexpress is an e-commerce platform for overseas consumers all over the world , It's the same as Taobao , Also for all end consumers , But the difference is that express is global 200 E-commerce platforms in many countries and regions .
Live e-commerce often “ Hundreds of millions of yuan with goods ” We've made great strides , The first mock exam is also made by cross-border sellers . However , Because the languages of different countries and regions are different 、 Different cultures 、 Consumption habits are also different , So if you want to do live broadcast in express , Language 、 logistics 、 Payment and technology are very complex systems , Among them, language is the most urgent problem to be solved .
In recent days , Ali Damo Institute jointly launched Express AI Live translation , Chinese to English can be carried out at the same time 、 Russia 、 Translation of three Western Languages . It is reported that , This is the world's first launch AI E-commerce live broadcast of real-time translation , Chinese businesses only need one button to broadcast in Chinese , It can cover English all over the world at the same time 、 Spanish 、 More than a billion people in Russian speaking areas .
“ Forced to ” Translation engine developed
Two years ago. , Fastcom begins to explore live broadcasting , However, it is different from Taobao live broadcast , During the peak period, the users who rush to the live broadcasting room come from 19 A country , And speak different languages . at that time , There is no multiple real-time translation , Or all live in English , Or it can be broadcast in a certain language for users in a certain language region , Like French or Spanish .
According to Lu Chen, who is in charge of the official live broadcast of fastcom , Most powerful businesses invite foreign anchors , And small and medium-sized businesses have to give up , Either live in silence or in Chinglish （Chinglish） On the spot . data display , Nearly 90% of the merchants on Fastlink come from China , and 82% More than 80% of Chinese businesses give up cross-border live broadcasting because of language difficulties .
Compared with Taobao live broadcast , Instant messenger is more like a baby , She needs to be strong “ nutrition ” To grow fast . This nutrition is a real-time translation tool that can realize multi translation and multi-function .
Before the development of multi translation , Express has tried out a ready-made translation engine , But for the translation output script of the live studio , Even the best quality translation engine , The translation result is also unqualified . at that time , There are also Russian and Spanish users invited to participate in the evaluation , Their evaluation is also completely incomprehensible .
After the research of Express link, we found that , The reason why these translation engines can't use live overseas , One is the conventional simultaneous interpreting with professional radio equipment to guarantee the accuracy of translation . But the live environment of e-commerce is very noisy , Many businesses start broadcasting directly at the gate , In this way, it is difficult to guarantee a good radio .
Secondly, the conventional simultaneous interpreting thesaurus is fixed. , Translation tools can be trained effectively to ensure accuracy . But the number of live e-commerce products is over 100 million , And it changes every day , There is no way for translation tools to learn quickly . And every anchor is from a different place , Some speak with dialect . And some of the anchors speak very fast , This will greatly increase the difficulty of translation . The above reasons make express have to develop their own engine .
There is no training set , Businesses themselves “ build ”
And then , Express time consuming 7 Develop a Chinese English translation engine in six months , And in 2020 year 5 Month on line . After the launch of the business, the broadcast rate has been 65%, This year, “828 Great promotion ” Period line 90% This function is used in live broadcasting rooms of .
But there are still deficiencies , Even if the ability to translate Chinese into English reaches the limit , Only in the studio 30% 's users can understand what the anchor says , because AE It's a global platform , There are all kinds of users in the studio , What is the highest proportion of English users 30%, Therefore, there is an urgent need to introduce the multi function of Chinese translation .
Two months later , Multi functional completion of Chinese Translation , After online, the user experience has been greatly improved , Sales volume of some stores increased accordingly 3 times . at that time , The purchase and conversion of many businesses' live broadcasting rooms , The conversion rate is higher than that of live broadcasting . Some live broadcast celebrities and overseas businesses found that , It also requires the use of multi-functional Chinese translation , But many businesses don't speak Chinese , For this reason, Speedpost has added multi translation and multi-function , And this year 7 Monthly online .
There are also many difficulties in the R & D process . The first is cross team collaboration , The R & D team is located in the United States 、 China 、 Three countries in Singapore . in addition ,60 Many R & D personnel have hardly met , How can they build trust quickly 、 And pushing it down quickly is a big problem .
The second is the collection of training sets . Cross border live broadcasting is rare in the world , There is no existing training set on the market that can be reused directly . Fortunately, Ali has a strong business ecology , Many businesses are actively helping to collect training sets .
The third is concurrent bearer point . If you only do one translation , There will be no pressure at all . But there are thousands of live broadcasts every day , Translation of several languages , This requires thousands of translators , Therefore, the concurrency pressure of the server is very high . later , Alibaba cloud has transformed this , It makes the server bearable .
The fourth problem is automated presentation . Multi translation and multi-function can automatically identify the type of users , And show the corresponding language , For example, it can identify you as Chinese , To show what the anchor said in Chinese . Before you see the simultaneous interpreting effect is all in PC End up , But the mobile end is very complicated , For this reason, R & D personnel have done a lot of adaptive work .
Self developed more efficient algorithm model , can “ understand ” Anchor accent
The unified multi language and multi domain fusion model developed by Dharma Institute , Hundreds of languages , And many fields of each language are compressed into a neural network model for translation , And the translation quality remains unchanged . According to Luo Weihua of Dharma Academy , Hands on the court AI Translation technology has become the first in many industries , Daily average adjustment amount exceeds 13 100 million times , Cumulative 3000 Million words , Can provide 214 High quality translation of different languages .
According to Zhao Wenqian, the person in charge of the real-time translation products of fastcom live broadcast , There are hundreds of millions of products on Express link 、 Tens of billions of terms related to commodities , The requirement of live real-time translation technology is very high , Need to overcome accent irregularities 、 The live broadcast environment is complex 、 There are many professional terms for commodities 、 New products and new words change quickly .
Fan Kai, a researcher at the natural language processing laboratory of the Ali Damo academy, is mainly responsible for the problem , He said it was a solution AI deafness 、 Questions you can't understand , Dharma Institute has developed a more efficient speech algorithm model , It can not only be used in noisy environments “ Hear clearly ” Live broadcast content , Can also “ understand ” The anchor's accent . By integrating multi domain knowledge into the translation model , The algorithm model can draw inferences from other examples , Without retraining, you can quickly learn the changing terms in different scenes .
Now? , Express Platform Support 18 Languages , It can guarantee the designed channel and product positioning , Be able to conform to the local culture , such “ When in Rome, do as the Romans do ” Mainly reflected in the following three aspects .
Appear to be . There are a lot of diversified products on Express link , Besides classifying them , It has to be easy to find . So the platform has to rely on the words the user enters , We'll find out what he's looking for right now . In the context of multilingualism , The challenge will be even greater . With “Case” For example , It has the dual meaning of box and case , When the user enters the word , What do you want to buy , It needs to be combined with specific scenarios . Besides , The same words have different meanings in different languages , for example “Basket” There are in English “ Basket ” It means , And in French “ Gym shoes ” It means .
Facing hundreds of millions of product search , It will be very slow to solve the whole process manually . So , SmartCom uses artificial intelligence to solve this problem , Of course, the initial training of artificial intelligence is solved by Algorithm Engineers . After training , Speedpost will eventually be able to do whatever the user enters , Can be based on the words he searched before and after , To know their purchase intention .
like the palm of one 's own hand . We hope that users can know the products like the palm of one's hand , Therefore, the product title needs to carry a lot of content . such as , What is this product ？ How to find ？ Why buy ？ But some business titles , Some have no punctuation , There are no special rules on the title technique . This problem can also be solved by artificial intelligence , It punctuates the title , And can be organized into local users familiar with the expression .
believe firmly . Express is a cross-border platform , Many businesses are located overseas , Goods need to travel across the sea to reach users . therefore , Users' credit to the platform 、 Business credit and product quality have high requirements . And before you buy, you want to know whether the product is good or not ？ Only through user evaluation to understand . But there are a lot of user reviews , There are also many kinds of languages involved . in addition , Users will use a variety of non-standard language to evaluate , such as “Very gooooood” Such phrases , Some evaluations also use emoticons . And use “ To many ” Translation engine , These problems can also be solved .
Because of the epidemic ,2020 Many businesses that had never set foot in live broadcasting before , For example, fruit farmers in mountainous areas of China have also begun to sell their products live . With express , Whether it is Gannan navel orange 、 Or the mink from Zaoqiang in Hebei Province can go all over the world . Again , Chinese users can also buy products that were previously available only on behalf of customers . Technology has never only benefited users in one country , It's the world .
Link to the original text
This article is the original content of Alibaba cloud , No reprint without permission .
- C++ 数字、string和char*的转换
- Won the CKA + CKS certificate with the highest gold content in kubernetes in 31 days!
- C + + number, string and char * conversion
- C + + Learning -- capacity() and resize() in C + +
- C + + Learning -- about code performance optimization
C + + programming experience (6): using C + + style type conversion
Latest party and government work report ppt - Park ppt
Online ID number extraction birthday tool
Field pointer? Dangling pointer? This article will help you understand!
GVRP of hcna Routing & Switching
- LeetCode 91. 解码方法
- Seq2seq implements chat robot
- [chat robot] principle of seq2seq model
- Leetcode 91. Decoding method
- HCNA Routing＆Switching之GVRP
- GVRP of hcna Routing & Switching
- HDU7016 Random Walk 2
- [Code+＃1]Yazid 的新生舞会
- CF1548C The Three Little Pigs
- HDU7033 Typing Contest
- HDU7016 Random Walk 2
- [code + 1] Yazid's freshman ball
- CF1548C The Three Little Pigs
- HDU7033 Typing Contest
- Qt Creator 自动补齐变慢的解决
- HALCON 20.11：如何处理标定助手品质问题
- HALCON 20.11：标定助手使用注意事项
- Solution of QT creator's automatic replenishment slowing down
- Halcon 20.11: how to deal with the quality problem of calibration assistant
- Halcon 20.11: precautions for use of calibration assistant
- "Top ten scientific and technological issues" announced| Young scientists 50 ² forum
- Reverse linked list
- JS data type
- Remember the bug encountered in reading and writing a file
- Singleton mode
- 在这个 N 多编程语言争霸的世界，C++ 究竟还有没有未来？
- In this world of N programming languages, is there a future for C + +?
- js Promise
- js 数组方法 回顾
- ES6 template characters
- js Promise
- JS array method review
- 【Golang】️走进 Go 语言️ 第一课 Hello World
- [golang] go into go language lesson 1 Hello World