Lost in Machine Translation - Why Hire a Human in the Age of Google Translate?
Updated: Mar 28, 2019
January 1981 was not a great time in Britain: strikes, riots and the IRA dominated the news. As a child, however, I was completely ignorant of all that. Each week I sat glued to the TV, lost in the incredible world of The Hitchhiker’s Guide to the Galaxy, dreaming of intergalactic escape from Middle England.
Even at that age though, I knew it was pure fantasy. Robots with psychological issues? A fish you could stick in your ear to understand any language in the Universe? That last one would have been useful for French school trips though (Brittany was the furthest from my universe I had ever been). Je suis désolé, je n'ai pas la moindre idée de ce dont tu parles.
Great idea, shame it was never going to happen.
Life imitates art
Nearly four decades later, I looks like I was wrong. Innovations such as Google earbuds now enable consecutive speech interpretation. Soon, students may no longer even need to learn a language to study abroad since tools exist that generate real-time, translated transcripts of lectures straight onto their screens. A host of translation apps are also available to install on any smartphone.
The market for this is huge, particular in emerging markets like Asia. Google estimates 50% of Internet content is in English, but only 20% of the world’s population speaks the language. The trend for speaking to devices is leading to people overcoming their natural reticence about interacting with robots, meaning uptake of machine translation (MT) will only increase.
But is the technology any good, and how does it compare with a human translator?
Ask many professionals a few years ago and they would have been confident their jobs were in no danger. Now, however, it’s a different story. MT software and AI have made huge advances thanks to changes in the way machines analyse text. Previously, MT looked at each word individually and provided a literal translation. Great for straightforward, formulaic texts, not so good when multiple interpretations of a word exist. How would the machine know which one to pick? Now, however, AI has evolved to consider how each word depends on and is modified by others around it and choose from a range of possible meanings. Essentially, it now does what humans do: understand language in context.
Work hard or work smart?
As a professional translator, how you feel about this will depend on whether you see it as a threat or an opportunity. Anyone who has spent endless hours manually transcribing a video or audio file will know what labour-intensive work it is. It therefore makes sense for companies like Video Translate to use the available technology to speed up the process. They report 80% accuracy and soaring productivity as a result, though recognise each step still requires human verification and people still prefer a human voiceover on the finished product.
I recently started using MT for written translation. The software searches global databases to provide suggestions for each section of text. I have to say I have mixed feelings. The pre-translated sentences are still clearly machine translated and have to be rewritten to look natural. Also, its choice of vocab, while definitely improved, is still often a little odd to say the least. Long story short, Ok to get the gist and saves some typing time but you wouldn’t want to trust it blindly.
There are even times when using MT can be false economy. Content creator Suzanne Wales warns “A thorough edit and proofread of a MT text can sometimes be more time consuming than just having a professional translate it in the first place.”
Essentially, MT remains good for formulaic language or text that enables people to follow a series of instructions (beware of operating heavy machinery based on a Google Translate though). However, it still lags behind as a tool for understanding the full breadth, depth and richness of language. Improvement is needed before machines can discern subtlety, abstract thought, idiomatic usage, wordplay, culturally appropriate use of language and the etiquette that make human communication so layered and complex. As content creator Nick Sirris puts it “There are multiple ways to say the same thing, but you would use certain phrases only in certain context. Also, how would a machine understand tone of the person you're talking to? When would I say ‘what's up’ and when would I just say ‘hello’?”
When no means no, and so do a lot of other things
Obviously, the odd linguistic misstep isn’t always the end of the world (or the British would never have become such great travellers). In certain situations, however, there are times when you just can’t afford to get it wrong, like high-stakes international negotiations, for example. Many cultures have strict and often opaque rules about communication etiquette. If a Japanese businessman tells you “It’s very difficult” what he really means is “No, no way, not ever, please don’t ask again.” And then there’s the fact that much of our communication is unspoken, relying on tone, facial expression and body language.
Successful communication therefore relies on more than just being able to bolt a sentence together. To overtake human interpreters, earbud technology will need to speed up to provide real-time simultaneous interpretation. AI will also need to be able to pick one voice out from many, or follow a conversation despite distracting background noise (as humans can). Plus, it needs to be coupled with excellent facial recognition technology. Until this becomes a reality, people who really ‘know’ a language are likely to retain an advantage over those reliant on technology. 3CPO might have been fluent in over 6 million languages but protocol droids are useless when it comes to understanding sarcasm.
Clearly though, the arrival of smarter and better tech is only a matter of time. Therefore, translators need to be ready for it. It doesn’t spell the end of the profession though, and a future in which AI and humans work together is more appealing than one in which we are replaced.
Adam Wilson believes “Artificial Intelligence will continue to improve and the 80% of our processes will become 98%. Artificial voices will become so good that you won’t be able to tell the difference. However, there will always be a need for humans for the finer details. Also, human work will evolve into something else that’s still needed. For example, human translators will be able to devote more time into really perfecting the humor, cultural differences, imperceptible nuances that will really enhance the translation, which is something AI still won’t be able to do”.
The bottom line is MT works best when, instead of replacing the translator it works as a tool for translators. The smart cookies (or biscuits if you’re reading this in UK English) in the near future will be those with the foresight to adapt and the ability to leverage the opportunities AI offers. Like the industrial revolution, the digital one may mean job losses, but also gains as people are forced to either retrain or focus on more specialised tasks.
As Adam Wilson says, “If we don’t adapt and evolve, we’ll become extinct”.