Artificial Intelligence and the translation industry

Künstliche Intelligenz ist nicht mehr wegzudenken

All the talk nowadays is of artificial intelligence and how our lives are going to change as a result of it.

But in fact, AI has been with us a long time now, and it has been impacting my life as a translator for some time. My opening above means “Artificial Intelligence is here to stay”.

This article is about the life of a freelance translator over the years and provides some suggestions as to how translators can survive the onslaught of what we still call “machine translation”.

“Artificial Intelligence is here to stay”.

There was no email when I started my professional translation career in 1990. Translations were typed onto a standalone computer, placed on a floppy disc, and dispatched by post. The originals from which we worked arrived by post or by courier. I recall one van arriving with ten huge ring binders containing user instructions for a computer program.

Meanwhile, and indeed since the 1950s, the clever boffins were beavering on their enormous mainframe computers, developing what we now know as automated translation. Early output was pretty poor, as it worked on a rules-based system.

By this time, the information super-highway was already on the horizon, and it wasn’t long before freelancers could send translations to their clients via file transfer protocol - FTP. This involved logging onto the client’s server and uploading the finished work. Most of my work was sent to me by fax at this time.

Meanwhile, in the background, the boffins were refining the systems, still on mainframe computers used by huge multi-nationals. By now, they had developed a statistical method, which was better, but not a lot. I recall reviewing and revising 20,000 words for a well-known conglomerate over an entire weekend. Frankly, the output from the machine was dreadful and it was impossible to produce a decent job in such a short time.

Once email arrived and search engines appeared, things started to speed up – and improve in quality, as translators could research much more extensively.

The next significant development was CAT tools (Computer Aided Translation Tool). In principle, a CAT tool is a database of every translation I have done since starting my first one in 1999. Translations are stored in segments (usually a sentence at a time) and can be easily searched and reproduced if and when a similar segment occurs in the future, thus maintaining consistency and improving quality. It also meant that missed paragraphs, a common mistake in the early days,  could no longer happen. But it is merely a database, not a translator. On day one of owning a CAT tool, it is empty and useless.

Meanwhile, neural networks were replacing the statistical translation machines. Along came Google Translate, Deepl, IBM’s Watson – all the big boys developed their tools with increasing accuracy and success. Nowadays, you might think the world does not need us as linguists. Indeed, in many cases, we have been replaced by machines.

Yet, and yet: the human still needs to check over the output. There are context errors, word order errors, false friends are often translated literally – a false friend is where a word that sounds the same has a different meaning, such as “actuellement” in French which is not “actually”, but “at the present time”. Or “location” which means “renting”, rather than a place

Or German “Chef “– isn’t the cook, but the boss (Koch is chef) or “Rente “ means pension and not rent, and “Gymnasium” is not a place to exercise, but a grammar school.

Much of my work these days is “proofreading” such outputs and adding soul to a soulless machine translation.

Definition of "translation" in a dictionary, highlighting its pronunciation and meaning related to converting text or speech from one language to another.

Literature is one area where machine translation can never prevail. It fails to capture the nuances of language that a human translator does. A translator of fiction, for example, will capture tone, playful language and humour, and nuanced vocabulary. A human looks at the broader context and is aware of cultural differences where some changes might be appropriate. In this field in particular, machine translation cannot be king.

Machine translation can handle vast swathes of text – especially boring ones, software strings, or long lists. And they are getting more and more accurate. In principle, the longer and better the input, the better the output. And I use it to help me in my own work. But I always need to do a lot of editing.

Why? Because I still believe that the machine has no soul. And a translation needs to look to the reader as though it was written in the target language in the first place. The style and customs of the target language must prevail.

Good translators are learning to offer other services, such as copy-writing, SEO research, and even outside translation, teaching, or doing something else entirely. I diversified by offering training services to translators alongside my translation services, before retiring from mainstream translating. Even in “retirement”, I still have a few clients who appreciate my service and attention to detail.

Good translators will survive but must diversify to maintain their incomes.

And we have to accept that artificial intelligence is here to stay.

Register for our next club meeting