I care about the world, you care about me.
Last week, the eighth session of the “Huke Frontier Knowledge Lecture Series”
We invited top AI expert Min Kerui to discuss the developments in machine translation over the years and to live demo the current level of machine translation.
Tracing the history of machine translation and contemplating the future of human translation as a profession.
This time, we suggest everyone click to watch the replay video of the live demonstration of machine translation.
After the live stream ended, we compiled nearly 15,000 words of the lecture transcript, selected as follows.
Guest Introduction
Min Kerui, CEO of Meta Technology. He holds a Bachelor’s degree in Computer Science from Fudan University and a Master’s degree from the Department of Mathematics at the University of Oxford in the UK, later pursuing a Ph.D. in Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign. He has served as CTO and co-founder of Boson Data, Chief Scientist at Cheetah Mobile, and an intern researcher at Microsoft Research Asia, participating in the Google AdSense content-based advertising modeling group’s click-through rate prediction project. His research results in text analysis and machine vision have been published in relevant international conferences and journals such as CIKM, CVPR, ACL.
Lecture Content
|
1 |
About MetaSOTA |
|
2 |
Introduction to Machine Translation |
|
3 |
Status of Meta Translation System |
|
4 |
Looking to the Future |
Lecture Highlights
1. How to View Machine Translation?
“Machine translation may be the technology application with the least innovation in business models in the world.”
Why is it said to have the least innovation in business models?
Looking back at the entire history, as early as 30 or 40 years ago, humans were already thinking about how to use machines or mechanical means to translate articles and languages.
2. What Level Can Machine Translation Achieve in Recent Years?
From our observations in this field, for major language translations like Chinese-English, machines have often accumulated a very large parallel corpus, based on this data, machines can achieve very good results in translating general types of texts such as news articles.
About a year ago, Microsoft published a paper discussing how they were able to achieve human-level translation in news translation by incorporating many detailed optimizations into neural machine translation. When a news article was given to both a professional translator and Microsoft’s machine, the translation results were basically indistinguishable.
Swipe up and down to see the full image
3. Can Machine Translation Replace Humans?
Can machine translation replace humans? We believe that for a long time, it may not be a complete either-or replacement relationship, but it could significantly speed up the language understanding and translation work that was originally thought to be only done by humans, with machines taking on 95% of the workload, while humans take a secondary role as assistants.
Swipe up and down to see the full image
4. What Does the Future of Machine Translation Look Like?
We believe that with technological development, the level of translation is likely to rise from sentences to entire texts. In this process, issues such as translation style and consistency can gradually be modeled into machine translation systems, enabling the translation of thousands or tens of thousands of words and understanding of entire texts.
Min demonstrated the legal text translation machine with the highest “score” on the market, showing the content produced in real-time. Interested friends can read it carefully.
Machine Translation of Sentences
Machine Translation of Paragraphs
Machine Translation Preserving the Format of the Entire Article
We believe that in specialized fields such as legal translation, there will be a huge shift in models within five years. For example, the first reaction of some top law firms in the industry regarding the translation of many articles and legal documents is still that this needs a lot of human intervention, still primarily done by humans.
But we believe that in no more than five years, the human-centered translation model will shift to a machine-centered one. Meanwhile, overall efficiency will also improve, at least a tenfold increase in efficiency; for example, a task that might have taken two weeks could be reduced to less than a day.
Moreover, machine translation may not exist as a completely independent application. For instance, in the office process, machine translation might be the last step in writing an article, integrated with intelligent document processing as part of the workflow.
[This lecture transcript totals 14,679 words. Scan the QR code on the poster below to permanently watch the video replay and access the complete transcript (fee required).VIP users free.]
– END –
Huke Frontier Knowledge Lecture Series
▸When Can Machine Translation Replace Humans?
▸Listen to MIT architects talk about the future of cities
▸We invited aviation experts from Seattle to talk about the Boeing behind air disasters
“I am Lin Bohu” Official Backup
If I go missing one day, remember to come here to find me
If you like it, please click on the “Like” button at the bottom right ↘