Business Article

East Meets West
/ Categories: News, Science, Technology

East Meets West丨Xu Bin: Can AI Translation Eliminate Cross-Language Communication Barriers?

At present, the rapid development of artificial intelligence technology has profoundly affected the translation and language service industry. Big model platforms represented by SparkDesk, ChatGPT, Ernie Bot, etc. have sprung up and accelerated their iterations and upgrades, which make the traditional translation industry face unprecedented challenges and have a direct impact on cross-cultural exchanges and international information dissemination.

How can AI technology break the cross-language communication dilemma? How to achieve the complementary advantages of AI translation and human translation, and what help will AI translation provide for Chinese culture to go abroad? Focusing on the above issues, Xu Bin, a professor at the School of Foreign Languages of Shandong Normal University and vice president of the World Interpreter and Translator Training Association, recently accepted an interview with East Meets West Question to explain.

 

Xu Bin is a translator and professor at the School of Foreign Languages of Shandong Normal University. He is also the vice president of the World Interpreter and Translator Training Association (WITTA TTES), the director of the Translation Technology Committee of the Shandong Translators' Association, and the secretary-general of the Committee on School-Enterprise Cooperation. He has long been devoted to translation practice, translation teaching and research, and has published more than 60 books with a total of more than 15 million words.

What are the characteristics of popular AI translation tools nowadays? What scenarios are they mainly applied to?

At present, AI translation tools are divided into two types: one is a machine translation engine developed by following the traditional machine translation research path, including Baidu, Youdao, Google and so on. This type of engine has adopted neural machine translation technology one after another since 2016 and has made great progress in the fluency and readability of the translated language.

The other is the AI intelligent translation tool "emerging" from the big language model represented by ChatGPT. Big Language Modelling is a deep learning algorithm that learns to process massive datasets, acquires relevant knowledge, and "hones" its ability to generate text and other content, including translation capabilities.

The OpenAI website describes ChatGPT as a "language model for optimizing conversations". Credit: Screenshot from the OpenAI website.

Compared with human translation, AI translation presents the following characteristics:

First, there are many languages. Whether it is traditional machine translation or large language models, the technical paths of research and development are different, but they all have multi-language capabilities and support the translation of dozens or even hundreds of languages. In contrast, human translators are often only competent in translating between their native language and one foreign language, and very few can translate skilfully between their native language and two or three foreign languages.

Secondly, it is fast. Compared with manual, the biggest advantage of machine translation is fast speed, especially Baidu translation, Google Translate and other traditional machine translation engines, hundreds of thousands of words of the manuscript can often be completed in a few minutes. The translation speed of the large model is slightly slower, but the speed is also much faster than that of human translation.

Third, knowledge accumulation and function iteration are fast. The developers of machine translation will continuously improve and update the knowledge content of the machine translation engine and carry out functional iteration according to the actual application.

Fourthly, there is uncertainty, AI translation has many advantages and disadvantages. One of the main disadvantages is the doubtful "comprehension", which may cause mistranslation due to cognitive blindness.

A young man visits and tries out the GPT-4. Credit: Wang Gang.

In recent years, there have been frequent mistranslations that have successfully passed through human review and found their way into news or other texts. For example, a news article reported that at the scene of a shooting in New Zealand, the gunman had "many books (magazines) strapped to his legs", but in fact, the word "magazines" in the original report did not mean "the book", but "magazines" for semi-automatic rifles.

In addition, AI translation tools cannot actively identify flaws in the text itself, much less do a good job of translating rhetorical patterns such as puns, counterpoints, and poems.

Combined with the above discussion of the advantages and disadvantages of AI translation, it can be found that there are two types of application scenarios for AI translation: referential large-scale text translation, and auxiliary translation of technical documents.

How accurate are AI translators in cross-cultural communication? What role does it play?

From the point of view of the history of machine translation research and development, the accuracy of AI translation has reached an unprecedented level. However, when applied in practice, it is found that it only plays the role of a "good assistant" and cannot independently complete the translation work.

I personally experienced an incident. In early 2023, a publishing house saw that the big model represented by ChatGPT was on trend and had very good translation capabilities, so it formed an AI intelligent translation studio, hoping to adopt the model of ChatGPT as the main model, supplemented by human beings, and gave the books introduced to ChatGPT to be translated completely, and then the translations produced were edited by human beings. A book of about 200,000 words will only be given about 10 days to review and embellish by human beings. During the process, the editors would try not to reference the original text, so that the book can be published and distributed at the fastest speed.

When I heard this, I tried it on my computer and concluded that "this mode of translation is completely unrealistic".

When discussing the translation level of AI, if you don't do a comprehensive and in-depth evaluation, but just take some random sentences and passages to do the test, you may feel that it does a good job. However, the problem is that book translation is a comprehensive and complex project, and ChatGPT's performance will "fluctuate", and even make mistakes in the translation.

There is no pattern to these errors, so the number of errors hidden in the 200,000-word manuscript is frightening. To identify and correct these errors, it is still necessary for high-level or even expert translators to read and understand the original text comprehensively and thoroughly, and to carefully compare and verify the text sentence by sentence.

ChatGPT. Credit: Visual China.

From this point of view, even with the help of intelligent translation tools such as ChatGPT, the translators' labour work has not been eased too much. Because of this, the project is not moving forward smoothly, and publishers still need to recruit translators according to the traditional process, and strictly follow the translation and publishing process of three reviews and three proofreads.

There are many similar examples. Even for simple texts, sometimes ChatGPT cannot cope. Nowadays, many management and inspirational books imported from overseas are not of high quality, and in the process of translation, translators and editors need to do a lot of touch-ups and corrections. In doing so, they also need to take into account the needs of the intended readers, which is a very difficult thing to do.

In general, the current AI translation can give people a "stunning" feeling in limited cases. But for the comprehensive book translation work is still difficult to handle, the accuracy rate can not meet the requirements of book publishing, and it is even more difficult to say "good".

As AI translation technology becomes more commonplace, does it mean that mankind is entering the era of barrier-free communication?

For general communication, AI translators can currently play a very good role in assisting communication, as well as in certain professional fields. In these occasions, even if the AI translation may be wrong, in most cases does not affect the communication between the two sides to understand the general idea. At the same time, by using techniques such as "paraphrasing", people using AI technology can allow AI to avoid mistakes.

But there is still a difference between this and accessible communication. In the foreseeable future, the popularisation of AI translation technology does not mean that humans will immediately and seamlessly enter the era of barrier-free communication either.

I came to this realisation from personal experience. In the summer of 2019, I travelled to German-speaking countries such as Austria and Germany, and despite having studied German at university, I had never really used it in my life, and wanted to try it out for myself that one time. I remember ordering a beer at the bar in an Austrian hostel, just as the owner had a friend over, and sitting and chatting for over half an hour. I couldn't think of a few words at first, but I was able to stumble through the general idea.

Afterwards, I think if I were a tourist with a translator, I'd have no problem ordering a beer and paying for it, but the other person probably wouldn't have any interest in making small talk with someone with a translator - the switching process of machine translation, the lagging sensation caused by the wait, would make the conversation so stilted that it would be completely impossible to have the wonderful experience of human interaction.

Staff members introduced a demonstration AI translation stick to the audience at the China Sound Valley Experience Centre in the Hefei National Hi-Tech Industrial Development Zone on Nov 2, 2018, in Hefei. Credit: Xu Qingyong.

I went to Japan in the summer of 2023 without speaking a word of Japanese, and all my communication in Japan was done with the help of a translation app on my mobile phone, which train to take, how to get to a certain place, etc. I had no chance to make "small talk" with strangers at all.

In short, with AI translation, we can indeed cross the language barrier and engage in some exchanges based on the transfer of necessary information, but such exchanges are mechanical and rigid, and can hardly become the kind of casual conversation between friends. The lack of casual conversation between people makes it difficult to generate real emotions; the lack of emotions makes it difficult to build deeper relationships.

From the perspective of international communication, what kind of assistance will AI translation technology provide for Chinese culture to "go abroad"?

From the perspective of international communication, AI translation can help foreigners who don't know Chinese understand Chinese culture independently, helping them search and target the content they need. Previously, even such work was done manually, which was inefficient and too costly.

Now with AI translation, anyone can use such a tool to proactively explore cultural and technological content in another language, and after broadly identifying what they are interested in, then seek professional translation services for more accurate information. With the help of AI translation tools, more foreigners can take a broader view, observe Chinese culture more actively, and actively choose what they wish to know about Chinese cultural content.

On 7 November 2018, Sogou Translator Pro was made available to guests in need at the venue of the 5th World Internet Conference "Light of the Internet" Expo in Wuzhen, Zhejiang Province. Sogou Translator Pro is the official and only designated translator for the Fifth World Internet Conference, supporting online voice translation in 42 languages and adopting the world's original neural network machine translation technology as well as offline neural machine translation technology developed based on a new version of NMT technology. Credit: Du Yang.

In addition, AI translation technology can assist high-level translators, make human translation more efficient, improve the efficiency of Chinese cultural communication, and solve the translation bottleneck problem in the process of "going abroad".

From the perspective of professional translation, AI translation can also help human translators improve the completion of Chinese cultural foreign translation projects. Because in translation practice, by using AI technology and specialised computer-assisted translation tools, translation efficiency can be improved, while guaranteeing the quality of the translation. If specialised training can be carried out with the help of high-quality corpus resources, AI interpreting will provide more accurate assistance for foreign translation of Chinese culture.

East Meets WestKailun Sui

Other posts by East Meets West
Contact author
blog comments powered by Disqus

Contact author

x

Latest