The annual Boao Forum for Asia kicked off in Boao, Hainan, from April 8th to 11th. As an international exchange event, the forum attracted participants from all over the world. As a major highlight of this forum, AI simultaneous interpretation can be said to have won the limelight. Not only people attending the event can feel the changes brought by science and technology at the event, but those who did not attend also expressed strong concern and wanted to see AI. How tall is the interpretation.
However, just yesterday, a picture sunned by netizens broke all kinds of illusions about AI's simultaneous interpretation and attracted a lot of rants. Both professional translators and machine translation experts sang. Then the question arises. AI translators want to be comparable to or even surpass humans. How far is it to go?
AI simultaneous appearances Boao Forum
This year's Boao Forum for Asia was the first to use AI translation to provide on-site guests with interpretation services. However, just one day later, AI's news about “dropping the chain” spread across the Internet. There were even rumors that the site had to hire a human translator again. "Take the plate."
In addition to inaccurate translation problems, the site was also exposed to the collapse of the translation system and caused garbled situations. I do not know how the guests at the scene saw such translations:
Tencent has also responded today, saying that this is an oolong event. The outsiders mentioned the invitation to hand-transmit, not the "Tencent simultaneous biography" team used for the live AI simultaneous biography of Boao, Hainan, but for services. Tencent news team's professional coverage in Beijing.
Tencent also admitted that in the face of the complex language environment and professional content of the Boao Forum for Asia, "Tenxun simultaneous biography" did make a mistake and answered several questions incorrectly. At the same time, as the AI product for innovation incubation and landing, “Tenxun simultaneous biography” is still learning and growing up, but the deficiency is insufficient, and the deficiency is to continue to strengthen learning.
Man watching a joke
Tencent AI submitted the incident of “dropping the chain” at the Boao Forum. There were mixed reactions from various parties. Basically it was divided into three groups: people who ate melons, people who read jokes, and no one lacked rational minds. As well as technical experts and AI researchers who translate the causes of problems for AI translators.
This reminds us that in March of this year, Microsoft announced that "the machine translation system it developed for the first time in the Chinese translation of general news translation has reached the level of human expertise, beyond the amateur translators, achieving a milestone breakthrough in natural language processing," resulting in a lot of Tucao , analysis and evaluation. At the same time, it raised a debate on whether machine translation can rival or even surpass humanity, and what the impact of machine translation development will have on human translators.
In the same way, this Tencent accident again performed this scene again.
Translation industry
The translation industry has been guilty and self-deprecating as "the rice bowl has finally been saved." For example, the translator, who knows the user "turns a sister," tells the voice of some translators:
At the same time, there are also relatively sensible people who calmly analyze the problems that Tencent AI's interpreters have made on such an important occasion: “Tencent has been betting on AI on such a tall and politically-bold Boao Forum. The reason is blasphemous. This already shows the current view of the government and large companies on machine translation."
Some people also said that although the level of machine translation is far from satisfactory, machine learning has developed rapidly in recent years, and there are fewer and fewer places where translations have to be handled manually. This has become a powerful tool for assistant translators to quickly complete their work tasks.
However, progress can't cover up the problem of over-promotion in the market. Some big companies boast that AI translation will replace human translation. In this regard, Douban users have done a rather calm analysis, and they believe that one day human translators may be replaced by machines like lawyers, doctors, fitness coaches, etc. Now that machine translation is replacing human translation, this area is so active. The reason why capital is favored is because it is different from some traditional industries that have high technical and moral risks, such as the medical industry.
The cost of making mistakes in the translation field is lower, and it also has greater tolerance for mistakes. But everyone needs to be vigilant about over-hyping publicity and have their own judgment. Here quoted by a watercress user's evaluation: "The big enterprises publicize the astonishing implementation of socialized publicity, one point of the product is blown into a very unacceptable, blowing a hundred percent is not an exaggeration."
Experts and AI researchers
It is not once or twice that machine translations make mistakes on major occasions. For example, because the machine translation has caused jokes, such as HKUST, Sogou, etc., large companies are still happy to do so, and they have continued to invest in the development and research of MT.
How far is the AI translation from the artificial level?
Does machine translation really replace humans? May be, probably not.
On March 14th this year, a team of researchers from Microsoft Asia Research Institute and Redmond Research Institute announced that its R&D machine translation system has reached the level of human professional translators in the Chinese-English test set of general news reports. . This is the first translation system that rivals the professional translators of humans in the quality and accuracy of news reports.
For the development of NLP, this is indeed a milestone breakthrough. However, I would like to thank Xiao Bian rude, according to Xiaobian years of experience in the performance of the English translation, Microsoft's translation system is not so powerful:
This is the tweet that Wu Enda set about preparing a new book. After Microsoft's translation, as shown in the above picture, it seems a bit difficult to read right? So what can Google Translate do?
This is only two small experiments, mainly to illustrate: Although the artificial intelligence translation system has achieved very good results, but at the current level, want to replace the professional translators, afraid of some difficulties.
Even if there is a deep learning model, if the semantic understanding is not done properly, the problem of machine translation still cannot be solved.
What is the use of machine translation in the localization industry today? The news media does not translate translations directly by machine translation. Manufacturers do not use machine translation to translate product manuals. At present, we are familiar with the direct use of pure machine translation. It is a search query or a user review of a translation provider or other website. The characteristic of this scene is that it requires a very fast translation, or a very large amount, and artificial translation is impossible to complete.
Whether or not to use machine translation, or no translation, this is back to the first point of view. Machine translation is not perfect, but it is better than nothing. In fact, because some manufacturers will employ labor to proofread or modify these comments on machine translation, or post-assessment of search word translation. There are also outsiders who say that machine translation actually provides more job opportunities for human translators.
It can be concluded that in a given scenario, given sufficient background information, machine translation may actually be better than human translation. At the same time, machine translation may be faster than manual translation. However, this is precisely the limitation of machine translation - it cannot respond quickly to translations in complex situations. If you temporarily switch topics and scenes, machine translation may not be able to achieve particularly accurate translation levels.
What do you think? Perhaps you have a better point of view, welcome to follow the message below 。