Machine translation is improving every year and technology companies are constantly telling us that AI algorithms are close to replacing human translators. The marketing campaigns are certainly impressive and the phrase “artificial intelligence” never fails to excite.
What’s the reality of machine translation and AI in the field, though?
We use this technology, which has been around for many years, on a daily basis – this means we have a pretty good idea about its strengths and weaknesses, including what kind of progress can realistically be made over the next five years. So what’s the current state of AI translation and what does the future hold for AI technology?
Machine translation & artificial intelligence in 2020
Artificial intelligence has changed the nature of machine translation, turning it into a tool that can improve and learn by itself. Before recent breakthroughs in AI technologies, such as deep learning and neural networks, machine translation algorithms had to be manually created, tested and managed every step of the way.
Minor improvements required hours, days and even years’ of programming labour.
However, the latest deep learning algorithms are able to consume millions of data points every day to improve the accuracy of their output, while innovations in neural language processing (NLP) make these algorithms more capable of understanding the contextual meaning of input dialogue – in the same way Google knows when you’re talking about Apple computers instead of the fruit.
We are now at a point where machine translation can learn by itself. Minor improvements are constantly coming, without any extensive input from programmers, while data scientists are continuing to engineer new breakthroughs that result in significant progress every few years.
Is AI translation ready to overtake humans?
Like many technology fields, there is a lot of hype coming from all angles related to AI technology, but we are not going to see any rise of the machines in the near future.
You may remember, in 2015, when Japan wowed the world with the first robot hotel, staffed entirely by machines. This was at the peak of AI hype and a glimpse of things to come for humans that will be doomed to redundancy. Except, four years later, half of the robot staff were fired for not being very good at their job and annoying the human clientele that pays for everything.
Humans currently staff the reception desk and handle key services at the world’s first robot hotel.
Luckily, machine translation has fared much better in the age of artificial intelligence than its cousins in hospitality. However, there is a familiar distance between hype and reality when you read the latest article claiming that AI translation is ready to replace humans and actually use the technology for yourself.
In reality, we’re a long way from algorithms replicating complex cognitive behaviour, such as the decision-making processes made in translation, without any human input.
This doesn’t mean progress isn’t being made, though. The issue is that expectations are very high as a result of the hype culture that’s so prominent in the tech industry.
What does the future hold for AI translation?
There will continue to be breakthroughs in AI translation in the coming years but the technology’s role will continue to be the same: enhancing the performance of human translators. As things currently stand, we can use the latest machine translation tools to instantly translate text and, for basic pieces of content, the accuracy can be good enough to act as a first draft for professional translators to edit and refine.
This can reduce the manual workload of a professional translator by 20-40% if the original text is simple enough for AI translation to achieve its maximum accuracy.
Even for more complex texts, machine translation can play a role and the time savings will only increase as the technology continues to improve – and this is where AI translation is going to make its impact.
As the technology continues to improve, this process will not be limited to text-to-text translation, either. We are at a point now where real-time translation and speech-to-text translation are starting to make genuine progress. Accuracy issues still remain and complex language often poses challenges, even for the most advanced AI algorithms – not to mention the variables of accents, pronunciation, background noise and interference.
Sadly, progress will continue to be slow because the complexity of language is so vast.
The realistic breakthroughs will be in computer processing that enables algorithms to complete “real-time” tasks much faster or interface designs and deliverability that allow human translators to edit and deploy machine-translated content faster – for example, at live events.
This is no different to how Microsoft repackaged existing translation technology into Skype Translator in 2014 – the same technology in a more innovative interface.
The accuracy of machine translation has improved marginally over the past five years, but the technology has improved enough in other areas that it is a more capable tool for human translators, who are able to deliver the required quality much faster, thanks to AI translation.