Will language translation tech ring a death knell for modern language learning?
“100 billion words a day? It is nearly unfathomable that Google’s neural machine translation can accomplish this”
Humans have been trying to find better ways of deciphering different languages for centuries, but it wasn’t until 1949 that the concept of “machine translation” really became a possibility.
According to a paper written by John Hutchins, Yehoshua Bar-Hillel was one of the first ones to take an interest in the field. He led a Georgetown University machine translation team and in partnership with IBM performed a demonstration of an automatic translation machine in 1954 known as the Georgetown-IBM experiment.
It was the height of the cold war, and the machine was capable of translating roughly 250 words from Russian into English. At the time, the demonstration generated a lot of interest, and it was predicted that machine translation would be perfected before 1960. However, computers weren’t advanced enough at this time to handle the complexity of translation, and subsequent experiments for the following few decades were lacklustre at best.
As the web became more established, interest in machine translation picked up again, most notably in 1997 when AltaVista’s Babelfish began receiving more than 500,000 translation requests as mentioned on this publication by Gaspari and Hutchins. By 2006, Google got in on the game, with Google Translate. While its functionality was initially fairly limited, by 2012 the company announced that that the tool had advanced to the point of being able to render the equivalent of 1 million books a day into different languages. Most recently, as stated by Google in a Super Bowl commercial that they launched last February, google translate accommodates 100 billion translations per day.
100 Billion Words A Day? Is There Even a Need for Human Translation Anymore?
It is nearly unfathomable that Google’s neural machine translation can accomplish this in one day. Using a concept called deep learning, it literally teaches itself how to translate as it goes along, and it gets better and better at it all the time. With this type of technology, one would think that the need for professional translators would have disappeared in the past few years.
Instead, as the world’s economy has become more globally focused, demand for human translation has remained strong. And while tools like Google Translate help immensely to get the job done, they still cannot replace human translators. Indeed, many experts predicted that by now human translation would be obsolete. And yet, it isn’t. Why?
Let’s say for example that machine translation has reached an accuracy level of 95% (it hasn’t, but just for the sake of the example, let’s imagine that it has). That 95% may be just fine for your twitter feed, ordering off a restaurant menu, or getting directions to a hotel room. Day to day casual communication between people can operate perfectly effectively with a 95% accuracy level.
“While tools like Google Translate help immensely to get the job done, they still cannot replace human translators”
However, business relies upon precision. With the economy geared at growth on a global scale, and companies dealing with numbers in the billions, even a 5 per cent margin of error could amount to millions in losses. And because neural translation is so complicated, it can be almost impossible to detect which part of the network is at fault. The only way to ensure 100 per cent accuracy is to combine the machine-assisted translation that you’ve undertaken with the input of a professional human translator. Check out these hilarious translation errors that show machine translation is not up to replace human translation yet.
The other issue is that language is enormously complex. We take for granted our native tongue and usually don’t stop to think that a large amount of what we say or write depends on the context or intent behind our words. Machines do not understand the context, nor do they understand intent.
“The other issue is that language is enormously complex”
Recently, there was an instance where a racist machine translation was generated by the AI app for WeChat – a perfect example of how machines cannot understand the complexity of language and what may or may not be appropriate.
Given the issues with machine translation, businesses are utilizing more cost-effective ways to combine human and machine translation to efficiently serve their translation needs. The combination of these two would fast track the completion of translation projects.
When Will Language Translation Technology Finally Override Humans?
Technology tools have advanced immensely, to the point where it is possible to comprehend time in the near future when machine translation will replace the hard work of some professional translation services.
Content of a sensitive financial, legal, or medical nature, however, cannot hope to be translated by machine, no matter how sophisticated the equipment is. It is likely to be a very long time before machines can ever replace the most exceptional instrument of all: the mind. Despite all of our shortcomings, machines will likely never have the adaptability, plasticity, and intuition of the human brain when it comes to language.
About the author: Ofer Tirosh is the CEO at Tomedes, a translation and interpretation service that works with clients around the world.