An embarrassing translation error
It was 2018 and Netta Barzilai had just won the Eurovision Song Contest.
It was Israel’s first Eurovision victory in 20 years and Prime Minister, Benjamin Netanyahu, tweeted his congratulations in Hebrew.
“Netta, at kapparah amitit.”
He used the word ‘kapparah’, which is a slang term of endearment — and the sentence should have translated into English as “Netta, you are a real darling.” But it didn’t.
Instead, the English-speaking world read his tweet as “Netta, you are a real cow.”
The embarrassing error was made by Twitter’s integrated translation tool, which is automated by Microsoft Bing.
The rise of on-demand translations
A human translator would never have made a mistake like that.
But, unfortunately, for any platform that offers translations on demand, machine translation is the only viable option.
For example, according to a poll by The British Council, translation apps, like Google Translate have become a go-to for young British travellers visiting foreign countries. More than 60% of 16–34-year-olds said they’d used their smartphone to understand the local language.
And 21% of those who’d used their smartphone had reported inaccurate translations, leading to misunderstandings.
But, while young people on holiday might be able to laugh off a mistake, it’s a different story for companies wanting to do business.
We previously posted about 15 Unfortunate translation failures that were not only incredibly embarrassing, but cost companies huge amounts of money.
Why machine translations fail
There are a number of reasons why a machine’s translations might be inaccurate. Here are some of them.
Words can have more than one meaning
Unlike humans, machines can’t recognise context. So if a word has more than one meaning, the machine translator will go for the most obvious one rather than the correct one.
Character strings can take on new meanings
In Chinese language, an individual character, used alone, can have one meaning. But that same character can have an entirely different meaning when it’s used together with another character. Sports brand, Nike, found this to its cost.
Nike launched a pair of trainers to commemorate the 2016 Chinese New Year, with a Chinese character embroidered on each heel.
The left heel had the character ‘fa’, which means ‘getting rich’. The right heel had the character ‘fu’, which means ‘fortune arrives’. But read side by side, the two characters formed the new phrase ‘Get fat’.
A human translator, or native Chinese speaker, would have been aware of this problem, but Nike didn’t consult either.
Language is full of nuances
There are lots of subtleties and cultural elements to language, like slang, evolved meanings, disguised insults, regional variations and idioms.
As humans, we’re learning these things as soon as we learn to speak and read. But we can never expect machines to master such complexities.
Machines run on formulae
Human translators can do language gymnastics. They’re experts at bending and breaking certain rules to make sure the words carry the same meaning from one language to another. A machine will only ever follow the rules it’s programmed with.
Machines can’t read the tone
The tone refers to the way the words are delivered. Maybe they’re supposed to be amusing, or have a poetic style. Perhaps they’re designed to shock, or motivate the audience to take action.
A machine performing a flat translation won’t pick up these style cues and the meaning of the message will be lost.
Will machines replace human translators?
Tech companies are always telling us AI is going to replace human translators. And there’s no doubt the technology is improving.
But the tech companies are overselling what their machines can actually do — and what they might be capable of in the future.
As sophisticated as their machines get, they will always have to be programmed and follow predetermined formulae, rather than thinking critically and rationally.
And though they might be able to translate basic sentences and formulaic documents quickly, we should be wary of trusting them completely without some human scrutiny.
Where documents and projects are more complex, the style, tone, meaning and intent are vital. And machines simply can’t compete with human knowledge and understanding.
They’ll never be able to learn and apply, through programming, what a human has learned in their lifetime. Or make a true human connection with real people.
Do you need human translations?
Lotuly never uses machines to do a human’s job. The potential for error is just too great — and you have a reputation to protect.
Instead, we work with experienced, native-speaking translators and writers all over the world. Skilled translators who understand exactly what you want to say and give you the right words to say it.
Are you a translator?
If you’re an experienced translator and would like the chance to work with a progressive, ethical and environmentally conscious company, Lotuly would love to hear from you.