Big Data Digest produced
Author: Liu Junhuan
see Fan Zhou more too slow!
I want to read the original comics, but I don’t know Japanese, so I’m so sad!
Although there are now popular translations on all major social media, the comics I want to read are too small for anyone to translate!
Don’t worry, don’t worry, the savior is here.
Last year, a brother named Shidu Xiangzhiyou developed a AI system Mantra, which combines comic-specific image recognition technology, machine translation and foreign language typesetting automation technology , In other words, can translate directly on the original comics.
You should know that in Japan, the publication of manga accounts for 40% of the overall publication, and the cost of manual translation is as high as 200,000 to 300,000 yen per copy .
This AI technology not only improves the speed of comic translation, but also reduces the cost of translation by 60-70% .
Speaking of this, Digest Bacteria thinks that AI has also helped China Shuangwen successfully go overseas. It seems that AI still has a very important contribution to the cultural industry.
This project received support from the FoundX project of the University of Tokyo, IPA, and Dongda IPC project at the early stage of development. Recently, DEEPCORE, DMM VENTURES, Legend Co, Angel Investment, etc. have invested 80 million yen.
It can be seen that everyone has very strong confidence and hope for comics to go overseas.
and tap the blackboard to draw the key points. Mantra service expects that will be officially opened to the public in July and will be the spiritual pillar of this year.
Sword refers to overseas release, Mantra implements more than simple text translation
Mantra mainly implements two functions, the first is 机翻译 In this regard, the researchers have performed mantra-specific technical processing on Mantra. On this basis, Mantra can achieve multilingual barrier-free translation.
What's the matter with the machine translation technology for comics? Brother Shidu introduced that, in simple terms, Mantra automates the traditional manual translation.
Human translation of comics mainly has the following three steps:
To text transcription ;
Fill in the text part of the language bubble and translate ;
Correctly typeset the translated text to .
It can be seen that machine translation of comics is not only a language conversion, but also includes post-translation typesetting.
To achieve these two functions, firstly is accurately positioned to the language bubble , and then is used to identify and translate the text .
sounds easy, but you should know that the bubble text typesetting in the comics has no rules that can be summarized , and there are differences in font size and style , which will make the machine translation of comics better than Traditional commercial translation is more difficult.
To solve this problem, Shidu Xiaoge developed an OCR engine dedicated to comic character recognition. With the help of the OCR engine, the system can accurately detect and recognizeDon't forget the text in the bubble, even if the comic image is very crowded.
On the left is the original Japanese version, and on the right is the machine translated version of Mantra. You can see that the accuracy of translation and typesetting is quite high.
But in general, the difficulty of translation is greater than recognition Difficulty, because comics often show oral behaviors such as stuttering or segmentation , and the relationship between text and expression must be handled well.
Currently, the team is also developing a recognition engine for comic expressions.
To fully realize comic translation must be time-consuming and costly. Because of this, the team’s goal is to reduce the cost of manual translation . The current translation still needs manual correction, in order to better For editing, the team is also developing a new tool that better combines machine translation with human translation.
Regarding Mantra’s potential development goals, Shidu Xiaoge said that he hopes that can use this technology in the publishing industry, expand the overseas market of comics , and use for personal use, which will help comics The author freely publishes his work on social media.
To create a future in which comic text will instantly go to the world
Shidu Xiaoge introduced, in fact, the development of Mantra is purely accidental.
At the doctoral level, he submitted a machine translation paper to the ACL conference. At the same time, one of his Nichia was studying for a doctorate in the field of image recognition, and he had also voted for some related topics. paper.
So in the summer vacation of his Ph.D., the two of them hit it off and decided to do a side job together, using a machine to translate a comic book. This is also the direction they decided after doing some user interviews.
"If we can make good use of the technology related to machine translation comics, maybe we can create a future where text comics can go to the world in an instant."
Later, we became more and more interested in this project After graduating from his Ph.D., he decided to conduct formal research in an entrepreneurial way.
During the development of Mantra, they also encountered a lot of difficulties. For example, Brother Shidu shared that, first of all, is missing in the training data. At that time, there was no bilingual comic data .
Based on the belief that everything is difficult at the beginning, they decided to select a small number of comic works that have been translated into foreign languages, and extract bilingual text data from them. In this way, created the first comic translation at a very low cost. The bilingual database in the field.
Nowadays, in comic translation, has many problems to be solved , for example, sometimes the character dialogue will be interrupted suddenly, a text bubble appears in the middle, and the subject and object are often omitted in the comic dialogue .
Recently, they are also addressing these problems one by one. The current translation accuracy rate of is not very high. , in the absence of human correction, they can barely understand 30% of the content. The goal is to achieve 70%-80% readability.
Mantra’s goal is to ensure that all comic works can be instantly transmitted to fans around the world 上, if a Japanese manga artist’s work is published on the platform It will be instantly translated by into 5-10 different languages . Such a platform can be realized. I believe can also technically eliminate pirated .
In order to make available to more people, they are also planning to expand the languages supported by the platform to beyond the current English and Chinese.
Chinese can not be less! Station B teaches you how to install, keep chasing the drama
Don’t think that translation is only the technical staff of the exporting country in anxiousness, and the fans of the importing country will have to see through. Among them, China will definitely want to. Take a seat.
So, the Chinese group may cry, because there is an UP master on the B station and open source an OCR translation software .
According to the introduction, when sets the translation box on the screen, the foreign language that appears in the box will be automatically recognized and quickly translated into Chinese. not only supports comics, It also supports games, fan dramas, PDF documents , etc.
On GitHub, the official staff summarized the six advantages and disadvantages of this translation software:
has a comprehensive scope of application, almost everything that appears on the screen can be translated;
has many translation interfaces, currently There are 12 translation interfaces;
simple and beautiful interface;
and its simple and foolish operation method;
is equipped with automatic translation mode compared with other OCR translators;
needs to be connected to the Internet, The translation speed may differ depending on the internet speed.
GitHub link:
https://github.com/PantsuDango/Dango-Translator
But in general, it is still very fragrant, and netizens have also said that is too fragrant too It smells, it smells better than YouTube!
In the use of , some people said that although it needs to be paid, the free amount of is enough to play games. . But can seeing such a fragrant translation software really control the urge to make money?
On station B, the developer also uploaded a video on how to install 手手的教:
Video link:
https://www.bilibili.com/video /BV1gp4y1Q7Ts
just right, the Dragon Boat Festival is here, and those who can't get out will wave up!
Related reports:
https://ainow.ai/2019/12/09/181620/
https://ainow.ai/2020/06/08 /223964/
People who click "Watching" will look better!
