Back in the early 1950s, the leading researches in the field of machine translation announced that it would be just three to five years before they solved the problem of machine translation services entirely. Clearly, their prediction was a little off; even five decades later nobody had managed to crack it. But could Deep Learning be the key to doing so? And will human translation and the concept of a translation agency as we understand it today become obsolete as a result?
What Is Machine Learning?
Machine Learning is when computers use algorithms to parse data, learn from it and then make informed decisions based on what they’ve used. This article on Machine Learning algorithms from Zendesk takes an in-depth look at the concept if you want to know about it in more detail. For now, suffice it to note that Machine Learning models mean that computers can get progressively better at the task that they are performing. That’s why many of those working on machine translation projects are excited about the potential that Machine Learning has to shake up the industry.
What is Deep Learning?
Deep Learning is a form of Machine Learning where the computer is able to identify and then correct its own mistakes (with both being subsets of the wider field of artificial intelligence). Deep Learning algorithms support the computer to use logic to draw conclusions in a similar way to the human brain. It does this through the use of neural networks and Deep Learning is thus a far more capable form of machine intelligence, as this article on Deep Learning vs Machine Learning from Skymind explains. It is used for a wide range of purposes, including translation.
What Does All This Mean for the Translation Industry?
Deep Learning definition aside, the potential for the deep neural network to crack machine translation is clear. The issues with machine translation have traditionally been around the poor quality of its results in terms of word choice, grammar, and sentence structure. Essentially, machine translation software delivers language that doesn’t sound natural, despite being fed tens of thousands (if not more) examples of written language.
Neural machine translation, which replaced the use of statistical machine translation back in 2015 and marked a significant leap forward, as a result, is therefore incredibly exciting. However, it still requires the machine to be fed comparable sentences in each of the languages it learns in order to translate them.
The really exciting part relates to the introduction of unsupervised neural machine translation, which this article on machine translation from Forbes explains in detail. Facebook, working with NYU, the University of the Basque Country and Sorbonne Universities, has produced a Deep Learning AI model that uses byte-pair encodings, language models and back-translation to take the next step forward in solving the machine translation problem. The new system means that a computer can translate from one language to another without having to be fed comparable phrases in each language.
The Future of Translation?
Such developments are bringing us significantly closer to flawless machine translation. A machine that learns to identify its own linguistic errors and correct them to the point of being able to translate accurately in any language would be an incredibly powerful tool. The benefits of artificial intelligence in this respect could be huge – the ability to translate accurately and instantly could open up communications around the world like never before, as well as providing global access to historic documents, literary works and so much more. Such a system could also be used to translate documents in most languages as well as rare and dying languages.
What Do Deep Learning Frameworks Mean for Human Translators?
While all of these Deep Learning applications have vast potential, we should remember that this is not the first time that researchers have been excited about the possibility of cracking machine translation. When those working on the Georgetown experiment in 1954 successfully translated more than 60 sentences from Russian to English, it led to a rush of funding for the brightest minds in the machine translation sector, who were confident in their ability to solve the issue in just a few years.
We’re in a very similar situation today, albeit with vastly more powerful technology being used. Companies such as Google, Microsoft, and Facebook are working alongside academic institutions, with plenty of funding being thrown at the problem. The results have been promising, but so were the results of those early experiments back in the 1950s. When it comes to machine translation, money doesn’t guarantee success.
It seems almost certain that machine translation will one day be on a par with translations produced by humans. However, that day doesn’t seem likely to arrive in the near future, despite the latest promising advancements. As such, it doesn’t seem prudent for translation agencies and the human translators who work for them to start seeking employment elsewhere just yet.