The ability for a computer to understand what we are saying – in any language – is now built into more areas of life than you might first realise. An algorithm determines what you see on the internet, based on the search terms you’ve inputted in the past. Voice recognition can control appliances in your home and even act as security for banking services.
When it comes to professional translation, should you put your trust in the emerging machine translation technology, or play it safe with human translators?
It’s no surprise the machine is moving in on the translation industry either – and to great effect in some cases. After all, who hasn’t turned to Google Translate in a sticky overseas situation? But is it any good?
Attempts to auto-translate began in the 50s but the latest technology – Neural Machine Translation (NMT) – is a very different beast. It’s far more intuitive as it mimics the “deep learning” processes of the brain.
What are the fundamental differences between machine and human translation?
There’s no doubt that the power of NMT – and the fact it continues learning as more and more data is fed into the system – is absolutely adequate to tackle many translation tasks.
Yet these machines, intelligent as they are, still have their limitations. Whilst machine translation is developing at an incredible rate, it would still struggle to deliver high percentage accurate translation and is not yet clever enough to pick up the nuances of language. Some machine translations are now good enough to produce excellent quality but this can be more dangerous – it makes it easier for mis-interpretations to be overlooked completely, which can often change the entire tone or sentiment of a message.
Machine translation is only as good as the data, known as corpora, that teaches it. For common languages there is a vast amount of bilingual corpora, but for low resource languages, the training data isn’t as comprehensive, and this can result in more questionable outcomes.
Furthermore, machines aren’t yet good enough to be trusted to translate alone. Even with the latest machine translation advances around machine learning and neural networks, human intervention is still needed. In the translation industry we call this Machine Translation Post Editing, or MTPE. Human input is still required for quality results.
Human translators also have the cultural sensitivities which provide context for the text. Only experience can inform you of the multitude of ways messages are best conveyed in different languages, with local idioms or native understanding.
And that’s before we even consider face-to-face interpretation. Body language, gesture – all the non-verbal communication – are also vital to communicating well.
Interestingly, 2020 has forced a reliance on machines as Covid restrictions have hampered in-person interpreting. Yet even when the only machine involved is a videolink, it is still a barrier to the sort of personal interaction which can be the key to enlightening interpretation. After all, it’s about so much more than just the words.
But will machine translation save me money?
Machine translation now has a very valid place in the translation industry. For certain types of text, NMT can add a lot of value, reducing your translation costs as well as the time it takes to get to market. If you do decide to go down the machine translation route though, it’s essential your agency has a robust postediting process.
If your project has social or cultural sensitivities, or has creative language, we recommend the human touch. No computer in the world will be able to understand the particular circumstances of each and every situation and machines lack the empathy and connection required to deliver a powerful message with sensitivity.
That’s not to say powerful technology won’t be used! We saved one client £15k on a human translation project using industry leading technology (see our other articles for a case study).
So yes – NMT is developing at lightning speed, but machines aren’t replacing humans just yet. For now, professional translators will always need to input somewhere in the translation process – whether that’s from the start, or in the post-edit.
It’s questionable whether robots will ever have the sophistication and emotion of a human brain – and until they do, nothing beats the accuracy and quality of professional human translators.