Artificial intelligence, dehumanisation and precarious work: translators on the frontline of tech-induced job degradation

Translators work in a changing and competitive field, where the use of new technologies – currently AI – is once again leading to the dehumanisation of their work and the deterioration of their working conditions, a trajectory that looks set to be the future for many other specialised professions.

(JAD-MCH/Equal Times)

Professional translators have been on the frontline of the impact of new technologies on the world of work for decades. Not only has this forced them to adapt to an increasingly uncertain and changing environment, but it has also shown them, over the years, how each new technological development, while providing them with useful tools for their work, often adds another layer of complexity, dehumanisation and a loss of control over the conditions in which they translate, and even over the very nature of their work.

In a trade where the mastery of the languages used is fundamental, with their wealth of uses and nuances, their linguistic codes, their subtleties of tone and meaning, their ambiguities, and where only genuine cultural immersion and a good knowledge of the context can ensure a result that is as accurate and fluid as the original, the value of a translator lies precisely in his or her ample experience, sensitivity and personal judgement.

Technology, masquerading as progress, is however being used to steer the profession towards an ultra-capitalist logic, where profitability takes precedence over quality – and where the worker is not at the centre, or even part of the equation. It is a trajectory that seems to presage with considerable accuracy the direction in which many other specialised professions could be headed.

As the capabilities of artificial intelligence (AI) grow, not only is the world of translation transforming. Administrative personnel, auditors, lawyers, recruiters, managers, advertisers, analysts, journalists, artists and creative professionals are also seeing the approach of a threat that is already a reality for many translators.

Technology-induced changes to the profession

There are an estimated 640,000 professional translators worldwide, and three out of four of them are freelancers, according to one of the most extensive surveys of the sector in recent years, with more than 7,000 respondents from 178 countries. It is this majority of translators who are experiencing a rapid, technology-driven decline in their profession. For this article, we spoke to a dozen of them, in different countries of Europe and the Americas, all of whom partly work as freelancers regularly subcontracted by translation agencies.

Jean-Jacques (not his real name, like all the translators mentioned in this article), who has almost 30 years’ experience working with French, English, Spanish, Dutch and other languages, tells Equal Times that, after passing a series of tests, freelance translators are able to join the translation pool of large companies that “generally have clients who need regular translations and a degree of operational security”. “They take a commission, of course, as intermediaries, and they can put downward pressure on translators’ rates, as they usually provide regular work.” It is the larger agencies (which account for a fifth of the market) that are responsible for the job not being what it used to be, as “they incorporate as much technology as they can to speed up the work, lower their costs and maintain or expand their profit margins,” he explains.

Jean-Jacques has always embraced every new technological development in his profession with openness and curiosity. He is anything but a technophobe, and yet he has experienced first-hand how the conditions and the very nature of his work have been gradually downgraded in the process.

AI and translation have been linked since the 1940s, explains this translator who began to see the limitations of machine translation as early as 2003, and then watched with interest as neural network-based translation made it possible, as of 2016, for the major agencies to integrate it within their computer-assisted translation (CAT) tools.

“Their main task is to segment texts into translation units, which are usually sentences, but can also be a single word, such as a title. They then present the document to be translated in a grid format, with the sentences in the original language on the left and a box for its translation on the right. The CAT tools then store each translated segment in databases or ‘translation memories’ (TMs), which are fed with the work of many different translators, and can grow to be incommensurately large. European Union institutions, for example, make their TMs publicly available, with billions of segments with translations that are already in place for the future.

“If a sentence appears in the translation memory that is the same or similar to the sentence to be translated, the CAT tools propose it to the translator, to speed up the work,” he explains. “The agencies are, of course, making the most of this ability to utilise pre-translated texts to cut costs by paying less for every sentence that is already in the translation memory”, with “different rates” per word “according to the percentage of perfect or fuzzy matches between the segments to translate and the translation memory”. Sentences that are new are paid in full but as little as 30 to 50 per cent of the original rate can be paid for those considered to be a high match.

This automatic pre-processing of texts has become standard practice in the industry, and “working with agencies automatically means accepting this type of pricing”. Worse still, the cognitive effort involved and the very nature of the work are completely distorted, leading to a dehumanised and alienating work environment.

With companies automatically pre-translating as much as possible, instead of receiving a clean text to translate freely, “we almost always receive files that are already segmented, with sentences already translated and machine translation suggestions for the segments with no equivalent in the translation memory”. So the translator ceases to be a translator: “It’s not really a matter of translating anymore but revising and correcting the segments proposed by the machine,” says Jean-Jacques.

Aside from the mental load of revising machine translations rather than translating freely, “these tools don’t understand the text, so they may offer you translations that are a very high match, but they don’t fit the context in which they are found”. And so, “I’m paid a lower rate because the text seems to match, but I have to correct the mistake and rewrite it from scratch”. Added to that, machine translation tools, like other generative AI, suffer from “hallucinations” and “can add or remove parts of a sentence out of the blue, so correcting this type of text becomes an exercise of intense focus, as every detail has to be checked”.

Like dairy cows

Rosa, who translates from English and French into Spanish and has two decades of similar experience behind her, fully agrees with him. “Many people have no idea that a machine cannot replace a human being, or that machine translation leaves a lot to be desired,” she says. Although she enjoys her work with her direct clients, with agencies “profit is the only thing that matters, and translation has become like a commodity that they extract from us at the lowest possible price, at the expense of treating translators like cows in a milk factory”. Sadly, “they don’t care about anything, neither the quality of the product, as long as it is minimally acceptable – and I spend my time correcting absolute horrors – nor how we are treated; they just want to make a quick profit. They’re the ones who demand the most, pay the least and treat us the worst.”

Rosa continues: “One of my clients is a major company that outsources the management of its translations to an agency, which in turn outsources the translations to a translation agency, which in turn uses freelance translators to do its translations from a hellish platform,” she explains. “I find myself translating segments in tiny little boxes, where there are also terms that are underlined in different colours – to take advantage of the system’s translation memory and discount those terms from our bill, of course – in the middle of a page with 50,000 functions, which is very confusing and visually exhausting, and is part of a ridiculously complicated overall system,” which wastes “a lot of time” and energy on matters totally unrelated to the translation itself, and pays “less than half of what I usually charge”.

“I accept it out of necessity, but what I’d really like to do is to tell them to take a hike,” she says, because “they try to automate everything as much as possible to cut their costs, but we have to spend our time going through countless administrative, IT and bureaucratic tasks to complete the process of a simple translation or revision”.

“And if the process is disrupted by a correction or an error, such a mess ensues that, between messages and alarms, the revisor, as the person ultimately responsible, is forced to fill in several documents explaining what has happened, how to avoid it in the future – grovel, repent – and is penalised with time without work. It’s insane,” she says.

“These are big companies that know nothing about translation, only about profit,” adds Rosa. “As an extreme example, there are the completely automated platforms,” which pay up to seven times less than usual, and which distribute the work by sending “an automated email alert that a job is available on the platform, which you have to rush to get onto and, like hungry dogs, try to get a bite of the prey to be shared among many, as you get to translate the segments that are still free” (decontextualised words or sentences) “and, in a matter of minutes, if not seconds, the translation is complete”. Once it is revised, “if they’ve corrected anything you’ve written, they also automatically threaten to disconnect you from the system if you have any further corrections, which are sometimes highly debatable”. This tendency to put profit before quality is the most destructive thing of all, as Rosa explains: “There are platforms that auction translations, and whoever is willing to charge the least gets them. The translation then goes through an editor who gets paid according to the corrections he or she makes, which are deducted from the first translator’s pay. I don’t know how those translations turn out, but there’s reason to doubt about their quality and most certainly about the quality of life and satisfaction of the translators.”

AI: a tool and a threat

The technology enabling automated pre-translations, which some companies and organisations are already using to replace human translators, particularly for more predictable or impersonal texts such as accounting tables and bureaucratic documents, is the neural network-based AI that Jean-Jacques saw emerge in 2016.

Responding to Equal Times, José F. Morales, professor of computational logic and researcher at the AI department of the Polytechnic University of Madrid (UPM) and the IMDEA Software Institute, says: “Generally speaking, AI without human supervision can maybe do monotonous jobs, but it often even does very technical jobs badly.” An AI translator “will struggle to understand the nuances of a text, or to put sentiment where it is needed”, and its use is already starting to have a damaging impact on language. “There are strange uses of English that are being normalised by AI and machine translations, which means that they appear in more and more texts and people start to consider them acceptable; then the AI itself feeds on them and trains with them, and so we get caught in a loop”, a vicious circle that could get exponentially worse in the coming years, he points out. Nevertheless, the professor qualifies, it is a useful tool, and “when it comes to translations, as with almost everything else, much the same applies: we should treat AI as if it were a student that we supervise: we can ask it to do a job for us if we’re discerning enough to know whether the result is good or not.”

There is also the question of readability, and the style that results from AI misuse can be torturous for a human reader, as Rosa points out: “A machine cannot replace a human, except for texts that are impersonal, repetitive and devoid of all literary style.” And if translators are squeezed to the point that their job is made unviable, she insists, “I worry about the future not only for myself, but also for the art of writing, because in the long run, if nothing is done about it, the same thing could happen to journalism and literature.”

Alina, who translates mainly from Russian into Spanish and English, but also has some knowledge of Arabic, Swedish, German, Ukrainian, Tatar and Belarusian, is clear: “AI is the eternal dichotomy, because it is both a tool and a threat,” she says. “It’s baffling to think that AI is learning from us,” that “we ourselves, the translators, are teaching it how to translate, how to improve. We are teaching the machine to replace us.”

What to do? Resist, like Hollywood

In search of answers on how to address this challenge, Equal Times spoke with Lindsay Weinberg, director of the Tech Justice Lab at Purdue University in the United States, and Robert Ovetz, a political scientist specialising in non-profit organisations and labour movements, of San José State University, also in the United States. Both suggest rejecting AI interference outright, and the red line could lie in defending the creative work that only a human being is deemed capable of doing (and charging for), learning from the successful resistance of Hollywood writers and screenwriters.

As Weinberg however points out, if they resist, the translators, like higher education workers, are likely to be tagged as “technophobic, anxious, unwilling to change with the times”, rather than defenders of “the quality and integrity of our working conditions”, particularly when “AI translation is being touted as so much more superior or effective or reliable than a human translator, although we see so many instances where this isn’t the case”. Far from being a mechanical “zeros to ones” type of task, translation “requires cultural sensitivity and awareness of context, types of knowledge that are very qualitative and that resist automation”.

Then there is the fact that freelance translators are isolated from one another by definition, and the professional associations in each of their countries or regions rarely have the force of a union or unified force with the power to mount organised resistance. And while they have considerable awareness about the nature of the problem in some countries, in others they have a more naïve take on AI.

“You’re highly fragmented, you’re all working at home, you’re using intermediaries […], you don’t have any contact with the staff who use your work,” says Ovetz. “If you want to address this, you have to map out your work structure, and you have to identify who’s downstream in the supply chain of your product. You have to identify where the work being fed to you is coming from, how is the work distributed”, because “the key to organising is understanding the supply chain, so you can identify the weak points, to disrupt it and defend what you want”.

And that is where translators can “find allies”, as some of these clients will have unionised employees with whom they could take joint action, or who they could encourage to change the way work is assigned, or how they are paid, etcetera, as with the clients that are not agencies.

Ovetz recommends that translators find ways to access the original text, so they can point out “how the final product is actually flawed because of the automation process”. A “tactic” would be to “build organising around that, and discredit the work process”, bearing in mind that technology is being used to streamline the task, break it down into small components and outsource or automate different parts of it, and give the leftovers to human translators.

As Jason Resnikoff, assistant professor of contemporary history at the University of Groningen in the Netherlands, underlined at a recent conference on automation, held by the European Trade Union Institute (ETUI), “narratives of technological progress are generally a losing game for workers, and I understand that resisting these narratives is in itself perilous”, but the unions and the workforce must do so. Even though, as Weinberg points out, the employer across the table from you may think “this person is not realistic, they don’t realise where the economy is going”.

But Resnikoff insists: “I think that’s a hit that unions are just going to have to take for a little bit. It falls to unions to find ways to be able to reject employer-initiated changes to the means of production, without appearing to be opposed to progress. And that requires unions to put forward their own very specific definition of progress.”

The cause of the bossless

Perhaps we need to keep sight of the idea of a “fair society”, so that automation ceases to be, as it has been since the 19th century, a “disruptor of the labour fabric”, argues Resnikoff. Then there is the question of how to halt any further deterioration in the status of translators, as well as other highly skilled professions, as so aptly portrayed by Elena, an English and French to Spanish translator with over three decades’ experience.

Rather like the “sans-culottes” of 1789, she considers herself to be misemployed, or “bossless”, because “fiscally, we are treated like a company, but I believe we are wrongly categorised,” she argues. Elena considers herself to be “an involuntary self-employed worker” because, like many freelancers, she does work that is similar in practice to that of hired employees, but she does so as “a worker who has been disassociated from an employer and classified as self-employed”, thereby “losing the rights and protections associated with salaried employment”. That is why she is appealing to trade unions to take up the cause of “employer-less employees”. “We are treated like a company, when we are not, because as individual freelancers, no matter how hard we work, we’re never going to make a profit – as a company can – because our work trades our time for money, and our brainpower and efforts have physical limits.” This is precisely the problem that employers, through AI, are trying to take out of the equation, while ignoring the fact that without human translators, the quality needed to make texts accurate and useful, if not intelligible or simply legible to their human readers, is impossible.

That is why it is so important to organise and make everyone aware of the weakness of this system, insists Ovetz: “One of the things you need to do is always be looking for the people who are entering the field and inoculate them against what they’re about to face […], introduce them to your organising, get them onboard. Because if you don’t, you know you’ll be divided.”