12 April 2024
The evolution of the translator
As the subject of artificial intelligence continues to dominate discussions about our changing roles at work, many people are beginning to feel anxious about how its potential applications could affect their future. For anyone working in localization or content creation, this is nothing new. Translators have had to deal with similar scenarios many times, thanks to the progression of computer-assisted translation, the advent of neural machine technology, the accessibility of Google Translate and so on.
In spite of this, translators have remained at the core of most localization efforts. While their working styles and the tasks they undertake might have evolved, translators have shown incredible resolve in adapting to the changing nature of the tech that impacts their world. In a series of interviews, the team at Alpha CRC shed some light on how the role of translator has developed over the course of the last 40 years, and share their insights into how it might evolve in the future.
I actually started translating back in the days of the typewriter. I even used to know a translator that did their work longhand. It’s been interesting seeing how advances in translation tools have changed the ways we work over the years, especially as someone who is quite supportive of using technology wherever possible.
When we started to use computers for translation, we just used general office applications, but in the 1990s we saw the advent of Trados – the first specialized software for translators. I think this was probably the very first technological revolution in the industry, one that I’d call machine-assisted translation. Even then, we were mainly just using Microsoft Word – Trados was effectively a plugin.
We then saw translation software become more autonomous and switch to the now instantly recognizable grid format. Software such as Déjà Vu paved the way for the now ubiquitous memoQ. As I mentioned, I’ve always been a strong proponent of getting the most out of our computers.
I actually developed my own system that worked with memoQ to allow pretranslation of English to Spanish using a vocabulary database that I created myself. Over time, this grew to over 80,000 words.
That said, it still needed a lot of human supervision, as the results were imperfect. I was still very much ‘translating’. Then tools such as Google Translate came along, and began to change the landscape again. At first, it acted as another form of help for the translator, as any output required a lot of post-editing and reworking. I’d say that my role evolved to one of a senior translator correcting the work of a junior.
When neural machine translation become more widespread, we began noticing a vast improvement in the quality of translated output. I found that I was having to translate less, and was now moving more into post-editing. I’d liken the current process to that of reviewing work from a colleague in many ways.
I’d like to point out that the majority of what I’ve said is suitable for industrial translation. Creative translation however (think literary translation, or translation of marketing collateral) moves closer to the realm of interpretation, and I believe this is where the human translator still has a leg up on the machine, so to speak.
In the future, as AI continues to evolve and spread throughout our industry, I think we might see yet another shift in the role of the translator. Perhaps we’ll see a consolidation of specialties – whereas before we had individual writers, translators, editors, proofreaders and so on, we might find that these tasks are being overseen by ‘language specialists’.
I’ve only been working professionally as a translator for three years, but even in that relatively short time I’ve noticed that I’m now doing more post-editing than raw translation. In fact, we’ve recently been working on some client projects that have given us some interesting insight into the potential future of translation: a shift towards machine translation for less-viewed content, with only the top 30% being sent to human linguists.
That’s in part thanks to the tools themselves getting better. Machine translation performs especially well with technical content. In the past however, there have often been very specific terms that clients request us to use. Now that we can feed that terminology to the machine translation engines, that’s becoming less of an issue.
It’ll also be interesting to see how machine translation improves specifically in regard to French. As a language with linguistic gender heavily influencing grammar, causality is something that is very important. Yet it’s also something that machines can struggle with. When reviewing machine translated content, I often spend a lot of time changing the gender of pronouns, adjectives or verb endings.
I think the advances we are currently seeing are also leading me to reconsider what it might mean to be a linguist in the future. I think there will always be a need for human translators and reviewers, but it might be that we’re expected to pick up a wider skillset in order to differentiate ourselves from the machine. Where machine translations excel with formulaic content, it tends to struggle with more creative stuff. I think that’s where we might be expected to flex our transcreation and copywriting skills.
Now I’m not a linguist specifically, but it’s been interesting for me to see how the role of the translator has evolved from a more administrative point of view. Especially when you stop to realize that, while the translator’s tasks have changed a lot over the years, these changes aren’t always brought about by technological advances as much as they are attitude changes.
When I think back to when I first started in the industry, clients weren’t particularly interested in talking directly to the translators. Instead, project managers were pretty much always the sole means of communication.
That’s now changed quite drastically. We have a lot of clients who want to know which linguists are working on their projects, and regularly invite them to meetings. Translators are therefore having to take on a much more ‘client-facing’ role than they used to.
Similarly, we used to have teams per language that dealt with all requests for their locale. That would mean that a translator could be working across multiple sectors with various clients. Over time, we began to expect linguists to specialize further, which is how we came to the position we’re at now, wherein translators aren’t grouped solely by language, but by sector. With machine translation now being implemented more and more, these specializations mean linguists are able to take on an expert role when reviewing content.
In the future, I do see technologies such as artificial intelligence changing the way linguists work again. It might be that we see a shift towards employing human resources for more high value content, or in developing prompts, feeding engines and verifying translated output.
Artificial intelligence does signify potential changes in our industry, but I believe that as long as we are able to adapt to the new opportunities it presents and modify our roles accordingly, we won’t have anything to worry about. Our role today is very different from what it was twenty years ago – I’m excited to see how it will continue to shift and evolve in the future.
I started out on a mechanical typewriter and an array of heavy-weight dictionaries, in 1970. If you missed out a sentence or changed your mind about a phrase, you had to retype the entire page.
During the miner strikes and the power cuts, I was working at an engineering company in Leicester. A few of the secretaries were fortunate enough to be given an electric typewriter, but I, as a translator, had to make do with a mechanical one. So on dark winter mornings, I was the only one tapping away while the others were having a good natter.
A slight improvement came with the IBM golf ball machine and that marvellous invention, the correction ribbon. By 1985 personal computers and word processors made a translator’s life significantly easier. I started out with an Acorn BBC, then moved to the first Macintosh, where PageMaker came on 4 diskettes. Real progress that was!
However, the Internet wasn’t around, so researching terminology and subject-matter knowledge was infinitely more time-consuming than it is now. Quite often, a trip to the university library was needed. In fact, looking back I don’t know how we did it.
On the other hand, those early days did have some benefits compared to life as a translator these days: Clients paid upfront or on delivery, and they were hugely appreciative. If you did a job particularly well or fast, you might even get a bouquet of flowers the next day! You were made to feel like a diva, or an artist. Last time this happened was 1997. I remember it fondly.
Since then, sadly, the status of the translator as a diva has eroded. You may have heard the expression ‘translation as a commodity’. Just as you buy sugar or flour by the weight, you can buy translation by ‘the word’ – the commonly accepted unit is actually ‘1000 words’. And the unit price has been going down and down; most certainly it has not kept pace with inflation.
Looking back, this downward spiral really started in earnest with the introduction of CAT tools in the late 1990s, and then, in December 2016 took a deep dive, when Google Translate and other MT engines started using neuronal networks and machine learning. That caused a real upheaval and conjured up the spectre of job losses. And the saga has continued over the past years, with AI and ChatGPT coming onto the scene. It is hard to convince people that translation requires more than a press of a button and a prompt. But while it is true that millions of words can be churned out overnight, it would be foolhardy to trust the results without careful sanity checks – using a process that is called post-editing.
Examples of hilarious misunderstandings by Google Translate and DeepL abound. These certainly help to make translators feel better and convince them that they are still needed. And that is indeed the case. Clients need reassurance. They are asking for light, medium or heavy post-editing. ‘Light’ being the most difficult – because we are expected to spend very little time, say 4 hours on 10000 words or more, but at the same time find and correct the ‘fatal’ errors.
That means it is not good enough to read through the target text, you do have to compare source and target to spot any discrepancies in meaning, omissions and additions (and those are not at all uncommon). And if it is ‘heavy’ post-editing you are asked to do, then the expectation is to bring the MT output up to human standard.
If in the early days you could make a living by translating 1,800 words a day, we are now expected to post-edit 20,000+ in a day. That’s a lot of words to process with your brain! Needless to say, MT output varies quite a lot, depending not only on the language pair, but also on the text type and on the quality of the source. And this indeed is where the problem often lies: the source is poor and illogical, too wordy – perhaps written in a rush, carelessly, by non-native speakers, or itself machine-generated … And it is us, the translators aka post-editors, who are tasked with putting it right.
So yes, if you ask me, our job has changed a great deal.
