intelligence artificielle

AI: discover how artificial intelligence is revolutionizing translation

Translation is entering a new era with the advent of Artificial Intelligence. AI based on Deep Learning makes it possible to speed up work by pre-cutting content to be translated into another language. Equipped with a multitude of features, translator programs also have the advantage of adapting to specific needs. Find out more through this article.

Artificial intelligence makes it possible to automate translation

It was from 2018 that artificial intelligence dedicated to translation experienced a real boom. Since this year, available technologies have moved up a gear thanks to automation. According to Theo Hoffenberg, founder of Reverso, applications are becoming much more autonomous to the point of revolutionizing the translator’s profession. Many sites now offer to immediately translate publications into the user’s language. This is particularly the case of Facebook, the social network with a billion members.

Although machine translation is common, it is still the tip of the iceberg. Other challenges remain for specialists in the field. Developers need to tweak the tools currently available for more satisfactory results. The quality gaps are obvious from the first words. Literal translations transform the meaning of sentences to the point of misleading the user. Experts believe that English-French translation is on a pedestal compared to those of other languages. This is notably due to French artificial intelligence.

Professional applications for specific jobs

Artificial intelligence is based on machine learning. The computer program carries out the work taking into account reference documents. This is the reason why most automatic translators are developed for the specific needs of a profession. Law firms, travel agencies and financial institutions have their own expectations. The choice of terms used and the level of language depends on the activity. Thus, some programs become translators dedicated to the legal field. Others focus more on health.

Developers mainly work on national translation tools. For example, to dissect documents in Goethe’s language, you will need to turn to a German translator. This is the same case for Chinese and Russian. An emerging country may have several AI-based translators who have their respective areas of expertise. Communication agencies responsible for sector monitoring juggle several tools. Like the algorithms developed by Six Continents, the final reader’s requirements remain the minimum. Unfortunately, the proliferation of automatic translators tends to lower quality.

A hybridization between the human translator and AI

To date, the flesh-and-blood translator is not ready to give way to artificial intelligence. Whatever the computing power of the computer program, humans and their 38 billion neurons remain essential. The person ensures that the results are relevant when it comes to translating content for professional use. In certain areas, it takes two individuals to proofread the text after the translation. This precaution is inevitable in the world of science and technology where English remains the most used language. Currently, the pharmaceutical industries as well as the world of finance also have similar practices. In all cases, the content is automatically translated by artificial intelligence before being verified. Experts talk about hybrid work.

Artificial intelligence is bringing change by creating augmented translation. This is a translation where the terms are chosen according to the context. This evolution of automation saves time. It is 30 to 50% faster and provides fairly relevant work. The Reverso Documents tool is among the most cited examples in this direction. Statistically, online software dominates the market thanks to its accessibility, but above all because of its free nature. These programs ask the user to verify so that the resulting texts are reliable, that is to say faithful to the meaning of the original.

Bestsellers pre-translated by algorithms

Many book publishers rely on artificial intelligence to translate requested works into several languages. For these professionals, it is more humans who are responsible for the translation. However, the help of a computer program remains welcome. Machine learning makes it relatively easy to find legal, medical or other references. The algorithms draw from memories or similar content that has been validated. Considered “transcreation”, it is a way of remaining faithful to the author’s ideas. Naturally, there is a double reading during the post-editing phase.

A lire également  One more step for NVIDIA in AI with the acquisition of ARM for $40 billion

Transcreation is a reality in the field of multilingual publishing. However, many companies in other sectors are also taking advantage of it. This is particularly the case for commercial brands. Blog posts or editorials dedicated to marketing are thus quickly translated into several languages. Artificial intelligence also makes the task of newspapers and international media easier. The latter broaden their targets thanks to pretranslation into several dialects. At the end of the process, it is always the editor-in-chief or publication director who reads the content in his or her native language. He ensures that the editorial line is respected.

Human intervention remains recommended for certain areas

Emotions are a real difference between a computer program and a human. Their respective translations stand out with sentiments recognizable by the choice of terms. An artificial intelligence is not capable of reflecting this particularity, despite a tendency to make a translation faithful to the original. Thus, human intervention becomes inevitable when it comes to appealing to emotion. This is the case for content whose purpose is to make people feel a need and invite people to purchase. Political speeches also fall into this category. On the other hand, artificial intelligence may be sufficient if it is simply necessary to inform.

Despite numerous efforts, it is still difficult to take into consideration the cultural adaptation of automated translation. In other words, the same phrase can have multiple meanings in different countries. This delicacy is more marked for certain fields, including advertising and marketing. All sectors related to sales must be extra vigilant so that the message is clear, especially if the announcement takes a second-rate tone. Then, the intervention can also be partial, total or selective depending on the power of artificial intelligence for translation.

The AI ​​technologies currently available are mainly general

Translation engines like Google Translate are at the heart of a debate. On the one hand, professionals do not even consider them as relevant solutions. They prefer to turn their back on these general tools in favor of much more specialized programs in a specific area. For their part, the general public is satisfied with online providers or open source software. These options have the merit of being free, at least in part. Regardless, DeepL and its ilk are sufficient when occasional translations are required. Paradoxically, for businesses, we need algorithms that can be trained for specific tasks using artificial intelligence.

The presence of more efficient translators does not yet constitute a threat to the humans engaged in this profession. These specialists even benefit from their existence since they can finally expand their range of turnkey services. In any case, machine learning takes time. It takes on average 20,000 standard sentences to train the algorithm to translate like a human. This figure increases tenfold if quality work is required. Furthermore, the initial text must be impeccably written. In short, companies believe they have computer programs adapted to their needs, but they just have a barely developed support tool. This is why they entrust their translation to agencies.

What about translation using artificial intelligence?

These 5 points summarize the situation of AI applied to translation.

  • Thanks to artificial intelligence, the job of the translator-interpreter is no longer the same as before. It plays more of a validator role.
  • Powerful IT tools offer the possibility of progressing quickly with work, regardless of the volume to be processed. Which justifies the fact that they are chargeable.
  • Human intervention is still essential, especially if quality translation is required. In particular, we need to work on the emotionality that algorithms do not have.
  • Online solutions have acquired a certain maturity. However, free access is a problem. Google and its ilk have not committed to keeping the content copied and pasted into their translator confidential.
  • At the current pace of machine learning, the next generation will be able to hope to benefit from translation via artificial intelligence that will match human sensitivity.