Image from

You’ve probably found yourself, at least once, stuck in a situation where you’re trying to grasp every glimpse of meaning behind someone’s words to understand simple directions in an unknown country. 

You’ve also likely wished that you could flick a switch and translate everything coming your way. Not unlike the Hitchhiker’s Guide to the Galaxy where the Babel fish, a leech-like creature that once placed inside your ear, instantly translates every language under the sun. 

But, how far are we from this kind of technology, and how is artificial intelligence affecting the translation industry? 

Let’s run through recent developments and see what the future may hold.

A brief history of modern translation

Translators nowadays have a wide array of tools that can make their job easier. CAT tools help translators convert texts, segment them, and alleviate the editing process. However, they often deliver rigid, machine-like segments that need a lot of editing. 

In a nutshell, they’re an excellent option for saving time, but they are not ready-made programs that can translate whole documents. 

On the other hand, machine translation programs can feed a document into an algorithm that automatically translates it into another language. 

Machine translation began with word-for-word translation models where a machine translates the text by finding the corresponding word in the target language. 

A word-for-word model was soon replaced by phrase-based statistical translation systems. These are more accurate considering that they use statistical methods to find whole blocks of words or phrases to translate the text. Even though this model manages to translate with higher accuracy, it still misses some of the high points of natural languages. 

The latest machine translation system that made an enormous leap is neural machine translation or NMT. It uses neural networks inspired by the architecture of the human brain to produce more accurate translations that correspond to the actual human language. 

Neural machine translation is now the most widely used translation system, and since 2016 it has been implemented into Google Translate to replace previously used phrase-based translation.

The main advantage of these systems is that they predict the probability of word sequences and they’re self-learning tools that improve over time as you feed them more data. However, one of the greatest shortcomings of NMT models is that they still have to be trained by humans to learn and improve themselves. 

The future is now

You may imagine future translators walking around in hazmat suits and trying to decipher alien languages like Amy Adams in Arrival. But, the truth is, they have enough on their plate with human languages alone.

Even though machine translation has progressed significantly over the years, a real-time translation is still a struggle. Luckily, researchers have been putting enormous effort into developing more accurate translation mechanisms to break language barriers.

For instance, Skype now has a real-time translator capable of translating nine languages. However, while it doesn’t have any problems with translating simple phrases, it often falls short when translating natural speech. 

Google is fighting the same battle. They recently launched Pixel earbuds that can translate spoken language in real-time. Thanks to AI, translation tools nowadays analyze the waveforms of a spoken language to find corresponding utterances in the desired language, making the whole translation process much quicker and more efficient.

As for now, real-time translation technology has an accuracy of around 85 %, and it takes 2 to 5 seconds to translate an utterance. Even though improvements are impressive, the translation still lags behind everyday human speech, making it pretty challenging to maintain a normal conversation. 

Will computers replace human translators?

From the first industrial revolution to the present day, people have dreaded technological advancement. 

For instance, in the 1800s, workers smashed machines because they feared they would put them out of work. While a certain amount of technophobia will always be present as on some level we’re conditioned to fear the unknown, fearing that every step in technological advancement will lead to job losses is groundless. 

Thus, technology isn’t going to completely replace human translators any time soon. 

For a start, even though it’s cheaper and faster, machine translation isn’t a one-size-fits-all solution. Some types of creative content like literature as well as marketing-related content are much harder to translate. On the other hand, MT proved to be an excellent choice for simple pieces like user manuals, guides, and similar technical documentation that is usually high-volume and needs to be translated quickly.

Another area where machine translation comes short is localization. Cultural differences and tone shifts often can’t be accurately transferred into another culture, especially if they’re distant. 

Data marketplaces

Artificial intelligence technology is built on data, and nowadays, data tends to be more precious than oil and gas. Considering that the machine translation market reached more than 550 million dollars in 2019, and it is expected to hit 1.5 billion by the end of 2024, we can expect translation data to reach even higher value in the future. 

Unfortunately, translation memories like the ones in CAT tools are often repetitive and not maintained well. Also, a significant issue tends to be big tech and language industry companies with a monopoly over the whole translation data industry. Since the future of the translation business depends on data, it’s crucial to break the monopoly on language data. This is where data marketplaces can become essential. For instance, TAUS and SYSTRAN are translation data marketplaces where any translator can buy and sell translation data sets needed for their MT systems. This way, they can monetize their knowledge while feeding their systems with the most accurate data.

How can AI change our communication?

According to Elon Musk’s prediction, human language will soon become obsolete. Brain implants like Neuralink will allow humans to communicate without uttering a single sentence. We will simply be able to think of something and share that thought with someone else. In this scenario, languages, as we know them today, are likely to become redundant since the brain-to-brain connection will allow us to overcome the language barriers. 

Eliminating language barriers has been on humanity’s bucket list since the dawn of time. From the tower of Babel to Musk’s Neuralink, all of our efforts have been put into understanding each other without any obstacles. 

But, even though language singularity can simplify our lives and bring us closer together, there are some areas where technology falls short. Languages don’t have a single purpose of carrying information; they are vital for every society since they’re rich sources of culturally unique nuances of meaning that could be lost if we switch to a universal language. That being said, we’ll have to make sure to include a wide range of culturally diverse meanings to teach the machines how to become culturally sensitive and cater to all our social needs.