In the midst of the Cold War, Soviet premier Nikita Khrushchev said during a public speech that the USSR would have outlasted the United States. This sentence didn’t sound friendly at all, but news reporters made it sound even worse when they wrongly translated it as “ we will bury the United States”. To American ears, this statement equalled the threat of an imminent nuclear attack, and that fear caused the conflict between the two powers to become even harsher. Thank God we all know that no nuclear weapon was deployed.
An article published in The Guardian at the beginning of April 2022, reported that the French newspaper Le Monde has launched an online English language edition and that the translation from French into English is partly carried out by AI tools.
As shown by the example at the beginning of this article, wrong or inaccurate translations can lead to serious diplomatic incidents, if not worse. This is why over time the importance of professional translators has been recognised and they are now largely employed to carry out accurate and sensitive translations. Sure, we all use Google Translator whenever we need to contact our business partner who doesn’t speak English or want to help our kids with their homework. But when it comes to mainstream media and politics, can we trust AI tools to provide the correct translation of extremely sensitive content?
How to nail the perfect translation
Before delving into the technology behind AI translators, we need to understand the process and the level of difficulty which lie behind a good translation. Many words in the English language have more than one meaning. Let’s take the word nail for example. When using this word you could be referring to the nails on your fingers, the nail you’ve just used to hang a painting on the wall, or again you could use it as a verb to indicate that you did something very well, I nailed it! Same word, three different meanings. But there’s more to it. Let’s take the last example I nailed it! If you shout this sentence while holding a newspaper in your right hand and pointing at a dead fly on the floor with your left one, you probably want me to know that you killed the fly. It’s complicated, we know!
All those examples are useful to show you that there is no such thing as the perfect translation of a single word, because it all depends on the context. So, if the human brain, which is capable of processing words, sentiments and many other internal and external inputs at the same time, finds it hard to come up with the correct solution, how can AI translators do that?
Recreating the brain of a human translator: neural machine translation (NMT)
The first examples of machine translation lacked accuracy, mainly because these early technologies didn’t take into account the different meanings and sentiments expressed by the same word or sentence. The real breakthrough in assisted translation happened with the deployment of neural machine translation, which applies artificial neural networks to predict the likelihood of a sequence of words and produce a correct translation. Artificial neural network is a machine learning tool which is at the core of deep learning. It is actually one of the first machine learning algorithms which were invented in the fifties and its structure resembles the one of the human brain. The same way billions of neurons work inside the human brain, an artificial neural network contains many nodes connected together, which are capable of producing an output in response to many different inputs. Quoting IBM’s website “once these learning algorithms are fine-tuned for accuracy, they are powerful tools (…) allowing us to classify and cluster data at a high velocity. Tasks in speech recognition or image recognition can take minutes versus hours when compared to the manual identification by human experts.”
The neural machine translation tools which are currently available on the market are trained to recognise and reproduce different meanings and sentiments and are way more accurate compared to previous translation assistive technologies. In other words, we can say that neural machine translation reproduces the brain of a professional human translator.
Will AI powered tools replace human translators?
The answer is no. At least for now.
While the use of artificial intelligence has significantly improved the accuracy of machine translation tools, they still produce mistakes. They struggle mainly with understanding the context, tone of voice, and the cultural and colloquial nuances of everyday language. Neural machine translation it’s definitely a huge help in speeding up the translation process, but it still needs human reviewing and editing. On its website, Le Monde explains that “the team (of its English language edition) is composed of eight journalists (…) responsible for selecting articles for translation, editing the translated versions and managing the English home page of the site. The translation is done by international agencies, with the help of an artificial intelligence tool. The selection and editing of the articles is done by native English-speaking journalists”.
We are sure that further development in technology will improve even more the accuracy of machine translation, but for now we are reassured to know that AI translators are being supervised by professional translators who correct the final output in order to avoid the publication of incorrect or fake news. Or so we hope.
Author Manuela Armini