The birth of the internet changed human communication beyond recognition.

In the mid-90s, a person with a connection in San Francisco could talk in real-time with an internet user in Tokyo: something that was simply impossible just a few years earlier.

Amidst all the excitement there was one big stumbling block, apart from the 16-hour time difference: how could they understand each other?

Let’s presume that neither spoke the other’s language, surely having a conversation would be virtually impossible?

Early online translation services, such as Google Translate, tried their best but were littered with errors, often with hilarious consequences. It stayed this way for over a decade until a major breakthrough was made; something that changed online translation forever.

That something was Neural Machine Translation (NMT), and it’s about to shape the future of communication as we know it.

Neural Machine Translation - what is it, and how does it work?

According to a specialist website:

Neural machine translation is the use of neural network models to learn a statistical model for machine translation.

In plainspeak, this is an advanced computer system that converts sequences of words at a time, rather than on the old-fashioned word-by-word basis.

It works by using a neural network model that predicts the meaning of chunks of words and then automatically translates them, normally to a high degree of accuracy.

The process can be split into two stages:

  1. The analysis phase – The model encodes the words into vectors that represent word meanings. Each one is then assigned a context based on the previous word. Using this context, a translation is selected from a list of possible options.
  2. The transfer phase – The words are decoded and placed into sentences in the target language.

All parts of this model are trained jointly, otherwise known as ‘end-to-end’, which maximizes performance. It also uses much less memory than previous models. Of course, this is a brief overview and there are more in-depth explanations online.

There are still some drawbacks, however. The system lacks awareness of human quirks, like wordplays and cultural differences, so users have to bear this in mind if they want to avoid some awkward moments.

Current uses

The beauty of NMT is that you barely know it’s there. You don’t need to download any extra software or read an online guide: it’s simply integrated into the website or app that you’re using, quickly and quietly translating text as you type or read it.

On a social level, Google Translate is perhaps the best-known NMT service, but there are a host of rivals that compete to translate dialogue in real-time, offering features such as speech translation and image recognition.

Online platform Airbnb has made full use of it by powering its in-app messaging service with NMT, thus bridging the translation gap between users. It goes without saying that this feature helped to push up sales significantly.

Similarly, Casino.guru, a multi-lingual iGaming website, has decided to launch an online forum with an automatic translation feature to allow users from different countries and cultures to communicate . Video chat platforms have also got on board, such as Skype with its Translator feature, which has the capacity to translate text in group chats of up to 100 people.

The future impact of NMT

The model is likely to shape the future of translation as we know it.

The DeepL Translator is a much-publicized start-up that uses NMT as its focal point across ten different languages, employing supercomputers that can process a million words in less than a second.

Such technology already allows corporations to hold meetings in several different languages, translates whole documents with one click, and assists developers in creating multilingual apps. With the arrival of 5G, further possibilities lie in store.

Simultaneous oral translation, where speakers simply talk into an app that instantly replays their words into the target language, is likely to be widespread over the next few years. Enhanced connection speeds would also mean internet users can translate huge documents in just a few seconds, in much the same way they’ll be able to instantly download a movie.

The rise of such tools raises an important question: Will it replace human translators?

The Managing Director of DeepL doesn’t seem to think so.

In a recent interview, Jaroslaw Kutylowski spoke of the ongoing need for human processes in professions such as marketing and law, where tiny linguistic nuances can often make a huge difference.

He also states the need for human creativity, which is something that machines will continue to lack for the foreseeable future. Not to mention the fact that he wants his kids to learn languages with a human, not a machine.

While there might always be a need for human translators to modify NMT translations, it doesn’t hide the fact that such technology could take a huge chunk out of their market.

Correcting work is a much lighter task than wholesale translation, and there might be some cases where just a general idea is enough for customers to complete their tasks without involving another human being. Market saturation might lead to lower prices as human translators compete for work with, of course, fewer big jobs available.

There are even fears that an overreliance on machines might lead to a decline in human knowledge – why bother to learn a language if a machine can do it for you? The same theory applies to other areas of life, such as engineering and teaching.

Harnessing the power of NMT

A topic that often comes up when talking about new technology is the need for humans to use it in a way that most benefits mankind. The area of translation is no different.

Interpreter Paolo Cappelli is one of a number of people who believe the power of 5G can benefit human translators hugely.

In an interesting article, he cites the possibility of VR video conferencing and holographic calls, which effectively puts the translator into a meeting far from his or her location.

The same concept applies to machine translation. Rather than letting it dominate the industry entirely, it should be treated as a tool rather than the main player in the process.

It could be a commodity product, helping people with their day-to-day tasks but not used for highly sophisticated tasks where an experienced human would be much better suited.

In this way, human language skills will still be able to command a premium; after all, someone with the language in their head will always preside over a device.

Only by keeping this fact in mind, will we be able to make the best use of Neural Machine Translation and its related technologies; and not let it erode other areas of human progress.