Can machine translation technology compete with humans?

November 29, 2018 Simon Hopes | Comments Off

Machine translation, an idea first conceived in 1949, has been a source of fascination for linguists, scientists, and inventors for decades. Countries around the world have been in a race to build machine translation systems that could eventually replace human translation services ever since those early promising experiments. But who will deliver this holy grail of parity with human translation? And more importantly, when? For more details of machine translation, you can visit the blog of Tomedes – a company providing professional translation services to clients worldwide.

The Slow But Steady Progress Of Machine Translation

Some companies had early success with machine translation, at least partially anyway. In 1978, Xerox began using its “Systran” system to translate technical documents. And the first commercial machine translation system was developed in 1991 at  Kharkov State University.

By the mid 90s, AltaVista’s Babelfish had begun to offer web translation, and the company was getting around 500,000 requests for translation daily. By 2012, Google made an announcement that Google Translate was rendering enough text to fill a million books every single day. Then, in 2016, it claimed that its latest neural machine translation (NMT) systems were capable of producing text that was “indistinguishable” from human translation.

Recently, Microsoft made similar claims, stating that it had achieved parity with human translation. And as recently as a few weeks ago, Baidu’s PR department sent out a press release about the company’s latest NMT research. Within a few hours, the media was all over it, making wild claims that Baidu had “cracked AI translation.”

The Major Roadblock To Successful Machine Translation


While machine translation is a remarkable innovation, there is one major roadblock that continues to stand in the way of complete parity, and that is…wait for it… machines are not human! Until a computer can successfully mimic all the complexities of the human mind, it will always fall short. To understand the subtleties of language, you have to follow the mechanics of human thought, something a computer program can’t do. In fact, there are many reasons why a human translator is always needed for an accurate translation.

We can programme machines to learn about a culture, but they will never be able to understand it. Each culture has its own linguistic nuances that are unique to that culture. To understand this, a machine would have to understand slang, regional colloquialisms, idioms, and so on. Native translation professionals are skilled in understanding these differences and thus can find the right equivalent in the target language.

Certain types of content require 100% accuracy. Legal documents, medical records, safety manuals, instructions for the use of pharmaceuticals, and many other kinds of documents cannot have translation mistakes of any kind. In cases where a translation is inaccurate, the result can be a loss of money, significant damage to a brand’s reputation, and even loss of life.

Style and tone
There are many texts where the style and tone of writing are very specific. It could be for persuasive marketing, expressing empathy, or exhibiting humor. Yet when translated by a machine, the result is flat and unoriginal. This type of content is meant to be understood by the reader in a particular context; otherwise, its meaning will change. Professional translation experts know they need to recreate a similar tone in the target language to keep the audience engaged.

Creativity and critical thinking
Many things come into play when a company is creating content for a website, from the wordplay that is used to particular word choices for the chosen audience. Machine translation is unable to account for either the audience or the industry and will choose what it deems to be the “correct” word. However, branding involves deliberate word choices creatively selected for their emotional impact and their suitability to the audience. Machines simply can’t make off the cuff decisionsaround these intricacies in the same way that a professional translator can.

Will Machine Translation Ever Replace Human Translation?

The rapid breakthroughs in AI technologies and the ongoing development in neural machine translation are exciting. We are living in amazing times. But even the best machine translation is still not able to match the skills of a human translator. Humans can make their own deductions, judge what a brand’s creative intent is, make tiny adjustments in translations where space is at a premium, and so much more that machines are not able to do.

There is a major push toward improving machine translation and an interest in this technology that has never been seen before. The world’s largest annual machine translation conference recently took place in Brussels, where the growing interest in anything related to AI drew double the number of participants the conference had in 2017.

The reason for so much hype within the machine translation industry varies. One could argue that it is fundamentally fueled by a hunger for prestige, power, and of course, wealth. And while many corporations claim they have finally perfected it, those same claims have been made for decades. But one thing is certain: We are still a long way from the science fiction reality of machines replacing humans. At least for now.


Louise Taylor is the head of content for Tomedes, a translation and localization company serving clients in more than 90 languages.  A professional freelance writer, Louise has loved languages since childhood and holds qualifications in Spanish, French, German and Latin.