Meet Audrey Possemiers, who’s helping NATO’s translators deploy cutting-edge AI tools

Life as a CAT coordinator

Audrey developed an early interest in foreign cultures and languages, and so she decided to study Translation and Multilingual Communication at university. She wanted to work as a translator but soon understood how the nature of the job was changing with the introduction of new technologies.

“When I signed up for my studies, my parents told me that translation would one day be done by machines,” Audrey says. “Today I can say that that is not entirely true, but the translation landscape is changing, there are new job opportunities emerging and it is thanks to technology that I am here today.”

NATO’s translators are recruited for their ability to produce carefully crafted texts that faithfully reflect the intention of the original, while always respecting the sensitivities in NATO’s different lines of work. In their work, NATO translators make extensive use of computer-assisted translation (CAT) tools, which rely on translation memories and terminology databases. The Translation Service is now preparing to deploy AI in its translation processes, initially in English and French, the two official languages of the Alliance. Other language pairs are set to follow soon.

NATO’s CAT coordinator since 2016, Audrey is the point of contact between translators and IT specialists, staying abreast of the rapidly evolving translation technology and ensuring that the translators have the best tools to perform their work. As such, she is spearheading the digital transformation of translation at NATO Headquarters.

“NATO is an international organisation that fosters multilingualism and diversity,” says Audrey. “The variety of subjects and the sensitivity of the content makes it an exciting workplace for translators, who have to give their best to deliver the high quality that is expected. For all of these reasons, NATO is the perfect environment for language technology to help translators keep up their pace and make the Alliance’s messages stand out across borders in a challenging world.”

From the origins of machine translation to today

Machine translation can be traced back to the 1950s. At that time, machines performed rules-based translation, meaning that they were taught the vocabulary and grammar of multiple languages in order to do automated translation.

"But in grammar, we know that there are as many exceptions as there are rules, so the machines would too often make mistakes that were highly detrimental to translation quality," Audrey says.

Three decades later, in the 1980s, statistical machine translation emerged. It made use of large amounts of data and the expansive memory of computers to predict the most likely translation. Millions of words were necessary to train this technology in each domain. However, though the results were good in scientific and technical writing, statistical machine translation presented many flaws, such as the machine’s inability to interpret colloquial or artistic language.

The latest innovation in the field of translation technology took place more recently, in 2016, when neural machine translation (NMT) was first introduced. Instead of performing basic automatic translation based on statistics and data sets like the previous technologies, NMT is an AI-based technology that relies on deep learning to mimic the way human neurons send signals to one another.

“The machine is able to check whole sentences and makes connections between individual words to better capture meanings," Audrey explains.

The tool breaks down sentences, calculates probabilities and matches individual words with the most accurate equivalent in the target language. As a result, neural machine translation output is considerably better than previous machine translation systems.

This new machine translation technology retains information from past translations, adapts to new contexts and learns from human corrections.

"You can feed the machine with generic language, but also improve it with your own data. Our plan is to train it by familiarising it with NATO’s syntax and agreed terminology," Audrey elaborates.

 

Why AI-based translation will not replace human translators

The ability of AI-based tools to translate quickly and even to use common idioms is impressive. However, despite continuous progress, this technology has its limitations. NMT output is only as good as the data that it has been trained on, hence the need for first-rate human translators.

In NATO’s official statements and other documents where the image of the Alliance is particularly at stake, there is no room for mistakes. AI tools are not equipped to understand the emotional nuances and the political or historical context of a certain text, particularly when it concerns the use of subjective terms, adjectives or expressions with a hidden meaning. Human translators will always be needed to verify any AI translation.

“Due to NATO’s political nature, words are of paramount importance,” explains Audrey. “The words that are used in our texts are not picked at random; some committees may negotiate for hours to come up with the perfect word to convey the correct meaning without hurting sensitivities or leaving room for misinterpretation (or on the contrary, deliberately introducing ambiguity).”

NATO’s experienced translators have the expertise that comes from knowing the context of the source material and their extensive understanding of the Alliance’s history and particular language. Still, deployment of NMT in the Translation Service further enhances the translators’ ability to produce quality and consistency in their work.

 

From languages to scuba diving

“To me, translation is about opening our minds to other cultures. Having people from different countries and backgrounds understand each other through translation makes us the ultimate peacekeepers,” says Audrey.

This interest in other countries and cultures extends beyond Audrey’s work in the Translation Service. When she is not keeping up with the industry’s latest trends, she likes to put her French, English and Spanish language skills to the test and travel around the world on the lookout for scuba diving spots.

"I have a passion for the sea and sea life. Before getting my scuba diving license, I would visit aquariums, but now I can see the creatures in their natural habitat. I have been to many countries in Latin America, the United States and New Zealand, but my favourite place on earth is in Yucatan, Mexico, where there is clear water and a big coral reef."