skip to Main Content

The (Unknown) Future of Translation in an Age of Artificial Intelligence

AI and the future of translation

Ever since our ill-fated attempt to reach the heavens from the city of Babel, humans have dreamed of instant, effortless and free language translation. After all, language remains one of the greatest obstacles to international trade and diplomacy, if not the greatest. The first obstacle was physical distance, but now that 4.4 billion people have internet access, and as the likes of Amazon have proven, any business on the planet can now enjoy direct access to almost 60% of the world’s population. The only real hurdles left are language and culture, which is perhaps why the translation industry has experienced such a boom over the last ten years.

And it’s also why the likes of Google, Microsoft, Bing, Baidu and a host of other players have invested billions in trying to perfect machine translation. In November 2016, Google announced that Google Translate would switch from phrase-based machine translation (PBMT) – which breaks sentences down into small chunks, or even words – to neural machine translation (NMT), which instead translates whole sentences, predicting the likelihood of a sequence of words. Google now claims to reduce translation errors by 60% compared to the previous version.

But what does this shift mean for buyers and sellers of translation services? Will clients soon have unbridled access to free, high-quality text translation? Are human translators’ anxieties justified? And what can translation-industry stakeholders do to prepare for AI?

Is parity between human and machine translation is likely?

Despite these watershed moments in machine translation, humans still appear to be on top. In a paper on machine translation published in March 2018, Microsoft stated that “quality continues to vary a great deal across language pairs, domains, and genres”. Furthermore, in a competition held in Korea in 2017, pitting human translators against the latest machine translation technology, four human translators scored an average of 25 out of 30, while the best of three translation programmes scored only 15. The other two scored less than 10 points.

One of the explanations for the failure of machine translation in this case resides in its apparent inability to interpret and communicate tone, emotion and culture, which are inextricably bound up with language and meaning. This is particularly true of creative texts – like the one used in the Korean competition – where style tends to take precedence over accuracy.

Furthermore, despite the fact that machine translation is now able to translate the meaning of an entire sentence, as pointed out by Samuel Laubli at the University of Zurich (cited by MIT), human translators are able to evaluate an entire text while machine translation only compares texts on a sentence level. Take the following complete sentence as an example, which can vary in meaning depending on the full context:

John may come to the party.

When translated into Spanish using Google Translate (March 2019), we get the following result:

John puede venir a la fiesta.

Back-translated into English again, this means:

John can come to the party.

Anyone who understands English will realise that “may come” does not mean “can come”, but has two possible interpretations, depending on the context:

  1. John might choose to come to the party.
  2. John is allowed to come to the party.

This simple example shows us that machine translation is still far from perfect. However, by definition, NMT can only continue to learn and improve.

The Way Forward – A Hybrid Approach

Despite the shortcomings that still exist with machine translation, the fact remains that it offers greater speed, scale and confidentiality. And this is a fact that businesses cannot afford to overlook, especially those with significant translation needs, and whose texts tend to be repetitive and formulaic (legal, financial, technical). Humans on the other hand are creative and have the benefit of intuition.

By combing these distinct advantages, we get the very best of both worlds: confidentiality, speed, scale, creativity and intuition. And in a world where the amount of content that we create and translate is only likely to increase, this can only be a good thing.

The future of machine translation
Trados Studio already integrates machine translation, with differing results

Prepare to Survive

There is nothing novel about the need to adapt to market forces, and there are plenty of examples of businesses and entire industries that have failed to survive the onslaught of modern technology. The translation industry is no exception and is changing fast. Machine translation technology has already been with us for several years, with all major CAT (Computer Aided Translation) tools now integrating machine translation features as standard (although how much they aid translators with speed and accuracy is open to debate). Nevertheless, the more that clients begin to trust machine translation, the more they will seek to use it, and to find suppliers that offer services such as post-editing machine translation (PEMT). In turn, translation agencies will look for freelancers that can help them meet this demand.

So, what can industry professionals do ensure they have a future in the industry? And how can we better prepare for and adapt to future trends? We’ve highlighted three ways in which both agencies and freelancers can not only adapt to the changing environment, but even prepare to thrive.

Offer post-editing machine translation (where appropriate)

As long as it is followed by human editing, the stigma around using machine translation will eventually disappear. Machine-augmented translation can help humans deliver translations with greater accuracy and speed, provided that clients give permission to use it. It is particularly helpful for longer and more formulaic texts, such as legal, technical and financial translations. For marketing-related translation and transcreation, however, our opinion is that post-editing machine translation should be avoided.

Adopt new pricing models

We also believe it’s time for the industry to adopt a new way of pricing tranlsation and localisation projects. For many decades, the dominant model has been that of per-word pricing. However, the advent of CAT tools and machine translation has made it trickier to determine how long projects will take. In cases where context is limited, such as product descriptions, and with shorter or more creative texts, machine translation doesn’t necessarily speed up the translation process. However, when the language is simple and the context is clear, machine translation can save significant amounts of time. Given this disparity, many now think that per-hour pricing would be a fairer and more transparent way of valuing the linguist’s time.

Mix it up

If language service providers and freelance linguists want to compete with computers, they have to offer their clients real added value. Freelancers and agencies can do this by providing additional services such as multilingual desktop publishing, typesetting, SEO and copywriting, proofreading and editing services as well as content population, interpreting, subtitling and transcreation. Because so many of these services overlap, by offering more than one, professionals will better position themselves to support their clients while remaining fully occupied with interesting and varied work.

Here at Keytext, we offer clients both per-word and per-hour pricing, with or without machine translation. For more information on our range of services, please don’t hesitate to contact us.

This Post Has 0 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top