![](https://www.techmonitor.ai/wp-content/uploads/sites/29/2018/09/Parrot.jpg)
When it comes to AI-powered translation work, robots rank higher than talking parrots, but they wouldn’t get very far without human translators. On International Translation Day, let’s celebrate humans, the most important part of the AI translation pipeline
In May of last year, the United Nations recognized September 30 as International Translation Day in order to celebrate the work of language professionals, but you have to wonder: with the inevitable rise of AI, how many more International Translation Days do we realistically have left?
Will we be paying tribute to human language professionals on International Translation Day in five years, or will it be a day forgotten amongst jobs lost to the digital workforce?
Don’t Believe the Hype
The reality is that while AI-fuelled machine translation has made unprecedented leaps and bounds to ultimately become a human translator’s best friend, we’ve got to beware of the hype: AI can’t be left on its own to provide accurate human translation, especially in the case of business. The translate app on your phone which may work for ordering food in foreign countries probably isn’t going to cut it when trying to troubleshoot, for example, customer service travel itinerary queries in different languages.
Think of it a bit like those ’10 kid’s test answers that are too hilarious to be wrong’ blog roundups. You know the ones. What ended in 1896? Answer: 1895. Or, where was the American Declaration of Independence signed? Answer: At the bottom. Like these kids, AI can deliver very literal results. On top of that, it can often make errors that have the potential to be extremely misleading from a customer service perspective, and which could have grave consequences for any business relying solely on machines for translation purposes.
Here are a few of the reasons humans really still need to play a big part in the translation process, especially when it comes to managing crucial translations amongst multinational businesses.
Neural Machine Translation isn’t Business Translation
Popular machine translation apps use billions upon billions of data points in order to help make sense of online content in another language, and recently we’ve seen that they’ve all upgraded to ‘Neural Machine Translation,’ which can increase fluency and accuracy in translation.
However, general purpose Neural Machine Translation systems are trained on domains like news articles and parliamentary proceedings. Drop them into customer service email and chat conversations with international customers or throw business critical information at them such as product descriptions, and things will start to sound odd very quickly. Brand names may not stay consistent, formal and informal tones may become mixed, and entity names may be completely mistranslated.
See also: Unbabel and Microsoft Roll Out New Translation Services
A human in the loop is definitely required when it comes to managing Neural Machine Translation systems to deliver fit for purpose translation in an enterprise environment. In this situation, most human translators will also have and apply a glossary for any given client, with specific instructions, brand guidelines and tones of voice adhered to.
Here’s an example of this type of machine translation peculiarity in the form of an unexpected translation of an out of vocabulary word:
Source (Russian): Наш хостел расположен в деревне Туришкино, которая находится в 60 км от Санкт-Петербурга.
[Our hostel is located in village Turishkino, which is 60 km away from St.Peterersburg]Machine translation (English): Our hostel is located in village Tururushkaino, which is 60 km away from St. Petersburg.
Since the Neural Machine Translation system did not have the name of the village (not a huge surprise given it’s a pretty rare word) “Туришкино” in its vocabulary, it had to translate it subword by subword. In this particular case, it translated the name of the village in a very accurate way that could have led to some seriously confused travellers in search of the forever elusive Tururushkaino.
Culture and Context Reign Supreme in Customer Support
![unbabel ceo](https://www.cbronline.com/wp-content/uploads/2018/09/Vasco-Pedro-1024x1024.jpg)
As humans, we can understand both the context and culture of specific languages, which is extremely key in translation. But Machine Translation systems are trained to read parallel sentences, which is a bit like teaching a parrot to talk; the parrot may be able to do it, but they are never going to truly understand what they are saying.
Think about it: would you trust Machine Translation to translate everything you need to tell your customers? Machines may not understand formality and tone of voice, and may make culture-specific mistakes as in this example:
Source (English): Make sure you have the latest operating system on your device
Machine translation (German): Stellst du sicher, dass du das neueste Betriebssystem auf deinem Gerät hast
[Make sure you [informal] have the latest operating system on your [informal] device]The choice of formality is usually defined by customer, however the use of inappropriate formality (like the use of informal “du” instead of formal “Sie” in this example) can be a real threat for communication with customers, and one that can seriously affect customer satisfaction levels and which may ultimately lead to customer churn.
These are only a few examples that show the limitations companies are facing with machine translation, despite how impressive advances in Neural Machine Translation have been. We’ll continue to see advances, and, from a customer service point of view these will increasingly save translators time and effort.
But September 30 is a day for humans – a day to celebrate and appreciate the hard work of the world’s network of language professionals who break down language barriers daily in order to bring nations together, facilitating international trade and new opportunities for e-commerce not restricted by language, culture and context.
Dr Vasco Pedro is the co-founder and CEO of Unbabel, which combines bespoke Neural Machine Translation (NMT), advanced automated quality estimation, and a global human network of tens of thousands of ‘Unbabelers’, who post-edit the automatic translations to professional quality.