The media update team explores this development. 

Machine translation, also know as automatic or automated language translation, has been around for many years – since the 1950s, in fact.

But traditional ways of translating sentences with machines can be problematic. Part of the challenge is that these methods translate a word or phrase by considering the words around it – instead of its context within a sentence.

The resulting sentences sound incoherent or even illogical.

AI, and specifically machine learning, is now being applied to machine translation to help solve this issue.

How AI-powered machine translation works

Neural networks, which is a form of machine learning, can make translation more accurate.

These systems start by ‘analysing’ the original sentence in a process called encoding. At this stage, the words in the sentence are converted into something that looks like a lot of numbers. What happens from here on out is a mostly mysterious process best explained by machine learning specialists.

The most important thing to know is that neural machine translation is more accurate than other systems because it translates whole sentences at a time, instead of pieces of a sentence.

Another benefit about this form of machine translation is that developers don’t have to create the same amount of rules and algorithms than they did for earlier systems. The new systems ‘teach’ themselves based on millions of examples of correct translations.

Other recently tested ways of applying neural networks to translation tasks are also showing promising results.

Many AI-powered systems require massive amounts of translation examples to learn how to process language. Some scientists are suggesting that machines be trained via unsupervised learning, which is when they learn from recognising patterns in data.

There would be no need to have millions of examples of how Mandarin is translated into English, for instance. The machine would just be let loose on databases of whole languages and be left to draw its own conclusions as to how languages work.

How it’s used

Google launched its Google Neural Machine Translation system in 2016. The system processes languages with fewer errors than the previous Phrase-Based Machine Translation system it used. It has been put into action on Google Translate with a total of eight language pairs.

Microsoft is using its AI-powered machine translation technology in a smartphone app that can recognise Standard Chinese words being said by a person learning the language. It evaluates the speaker’s pronunciation, rhythm and tone by comparing it to that of native speakers. The app then shows users where they can improve and plays samples of correct pronunciation.

There are plenty of other examples and new players, offering improved translation accuracy, that continue to enter the market.

Other companies are using AI to assist human translators. This form of augmented intelligence either shows humans suggestions or requires humans to edit translated content.  

Whatever the learning method, AI-powered translation is still not perfect – or comparable to what human translators can do. But this technology has its place and is worth developing considering the language barrier people face in both social settings and the workplace.

A human translator is not always available when you need to ask for directions or communicate with a colleague whose native language is different from yours. In these situations, any solution will help. AI-powered translation looks like it could be that solution.   

Want to stay up to date with the latest news? Subscribe to our newsletter.

AI is also improving machine-human communication. Read how machines are ‘speaking’ to humans more effectively in our article, What is natural language generation?