There’s no denying that machine translation has made huge strides over the last couple of decades. But will AI replace human translators? Despite the millions poured into this industry, machines won’t replace human translators any time soon. Here’s why.
In the beginning…
One of the earliest experiments with machine translation dates back to 1954. Researchers from Georgetown University and IBM demonstrated a computer translating 60 Russian sentences into English. Its vocabulary consisted of only 250 terms, and it worked with no more than six grammar rules.
And yet, some of the participants held the wildly optimistic view that within two to three years, the problems faced by machine translation would be solved entirely.
… and now?
Almost seven decades later, machine translation has come a long way.
By pasting text into automatic tools, we get a translation that is accurate enough to help us out in most situations.
Need to get the gist of an email that’s in a language you don’t understand? Or want to know what you’re ordering from a menu while abroad?
Automated translation tools do the job, instantly and free of charge.
These tools are also a massive help for human translators, who can get a first draft almost instantly and edit this rather than translate from scratch. This makes for a speedier process and can therefore reduce costs for clients.
This approach is already used by many and will no doubt become the standard in most translation workflows.
Striving for perfection
There are many benefits of effective, accurate machine translation, from overcoming language barriers between individuals to facilitating the exchange of information on a global basis and helping trade.
It comes as no surprise, then, that the machine translation industry is predicted to be worth $1.5 billion by 2024.
Nor does it come as a surprise that machine translation is getting more and more sophisticated.
In 2016, Google announced the Google Neural Machine Translation system, which leverages an artificial neural network capable of deep learning to improve the quality of translation. Google – as well as competitor tools such as Microsoft’s Bing Translator or DeepL – will become increasingly proficient with the help of deep learning.
And although the accuracy of these translations isn’t perfect, they’re so convenient that we’re learn to live with their drawbacks.
Why human translation is better than machine translation
So why are human translators still needed? Although AI translations may have a high degree of accuracy, they also have serious shortcomings.
One of these is their lack of context and understanding of the world.
When translating the title of Paul Beatty’s satiric novel The Sellout into French, for example, Google Translate gives us ‘la vente’, a translation that means ‘the sale’. This doesn’t do justice to the connotations and layers of meaning of the word ‘sellout’ in English – especially within the context of the book’s narrative.
It’s easy to dismiss context and background as nice-to-haves when it comes to translation. Yet because of the way language works, and above all the complexities of mapping one language on to another, the same word could be translated in dozens of different ways depending on the field in which it is being used and even the organisation or individual using it.
This calls for the kind of judgement that a computer just can’t offer.
It’s little wonder, then, that the translator of the French version of Beatty’s novel steered clear altogether and instead opted for ‘Moi contre les États-Unis d’Amérique’ (literally ‘Me against the USA’).
Machines can’t see the bigger picture; humans can.
Our understanding of the world is infinitely more complex than a computer’s, and this comes through when we translate between languages.
The role of creativity and elastic thinking
Another area where machines are seriously lagging is creativity.
Will a machine one day be able to translate text where creative input is needed – a poem, a joke, a love letter – as well as humans can?
Experts in the field of translation wonder when ‘singularity’ – the point at which machines will be better than human translators – will occur.
In his book Elastic, Leonard Mlodinow argues that our brains have evolved to generate ideas, seek novelty and explore. He writes about the key differences in the architecture of humans and computers: the latter ‘execute their analysis by following a well-defined series of steps – a program or algorithm – in a linear fashion that is specified for the task at hand by a programmer’.
Computers can compose music, make art and beat humans at chess, but they do this by following specific algorithms created by humans, rather than generating new ideas.
The same applies to translation.
By following rules and algorithms, computers may get a large chunk of the work done. But even if they (theoretically) manage to complete 99% of the work involved, what about the 1% that depends on our human ability to think in a flexible, creative, elastic way?
In many scenarios, this won’t matter, of course. We don’t need poetry to ask for the right medication in a Spanish pharmacy or to communicate logistical information to colleagues who don’t speak our language.
But in other cases, AI’s inability to think like humans do – indeed, to think at all – will represent a deep lacuna: it will make the difference between a text with or without poetry, nuance or creative flair.
So will AI replace human translators? Not any time soon. The human touch will be needed for a long time to come.
You can find out more about our translation services here.