Translation as an industry and an activity has undergone massive change in the last decade or so. It all started with Google Translate, the first large scale machine translation (MT) which allowed the average user to translate documents and messages instantly. At the time, it started out with rules-based MT (RBMT). It quickly evolved to statistical MT (SMT), which included phrase-based MT (PBMT). These have been the standard for most of this time, using statistical models to discern what a speaker or writer was saying. In essence, it was an educated guess as to what was being said.
Today, the new standard at Google and Microsoft is neural MT (NMT). It’s changing how translations are done and, many people feel, that it will ultimately displace human translators entirely.
Neural machine translation (NMT) is designed to mimic how the human brain learns language. It starts with the same ingredients as a human, some simple parameters and training data.
For example, when a child is learning English, they quickly recognize that the sentence structure is subject-verb-object. They don’t know that those descriptions, but they know that to be understood, they need to structure their sentence a certain. A child in Ancient Rome would not have recognized that need. As a matter of custom, Latin was usually arranged subject-object-verb, but because the language uses prefixes and suffixes to define a word’s function, the structure is more or less arbitrary.
NMT starts with similar guidelines for structure and some basic words and phrases. The rest of its knowledge is added by users that correct the system over time. The computer learns from native speakers the best phrases and structures. “Over time” in this case is a matter of weeks or months. There are literally millions of users correct the system each day, making the learning curve very steep.
Because it’s based in the actual way that humans learn and speak, NMT is already changing the translation industry. According to studies, NMT is able to reduce erroneous translations by humans translators up to 60%. This means that it is more accurate than the average human translator. With Google and Microsoft making it readily available, it is also easy to use and have at hand.
While not free right now for many applications, in all likelihood, NMT will be free or very inexpensive in the near future as more and companies invest is it and more businesses and individuals learn to count on it for their everyday use.
Many of the world’s leading tech companies and translation companies are investing in NMT. Human translators are far more expensive and less reliable than well-built NMT technologies. Microsoft, Systran, Google, and KantanMT are all investing significant amounts of money and effort into creating the most efficient and correct NMT systems.
Google NMT (GNMT) was enabled for 8 languages: English, French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish. This month Google is expected to enable Russian, Hindi, and Vietnamese. The other companies working with NMT have created similar lists and continue to expand their repertoire.
The answer is yes and no. For everyday business and personal transactions, NMT will be much less expensive than human translators and infinitely faster. Since it available on mobile phones, computers, tablets, and even airport kiosks, it is far more user-friendly than dragging a human translator everywhere.
That said, there are some applications where a human is likely to remain superior. When C.K. Scott Moncrieff translated over 3,000 pages of Marcel Proust’s A la Recherche du Temps Perdu, it was more than simply translating words; it was a massive translation of poetry, imagery, and emotion. It is unlikely that a computer will be able to truly convey those images for quite sometime, no matter how vast its knowledge of a language.
The famous madeleine scene from A La Recherche du Temps Perdu by Marcel Proust
Original French: II y avait déjà bien des années que, de Combray, tout ce qui n’était pas le théâtre et le drame de mon coucher, n’existait plus pour moi, quand un jour d’hiver, comme je rentrais à la maison, ma mère, voyant que j’avais froid, me proposa de me faire prendre, contre mon habitude, un peu de thé. Je refusai d’abord et, je ne sais pourquoi, me ravisai. Elle envoya chercher un de ces gâteaux courts et dodus appelés Petites Madeleines qui semblent avoir été moulés dans la valve rainurée d’une coquille de Saint-Jacques. Et bientôt, machinalement, accablé par la morne journée et la perspective d’un triste lendemain, je portai à mes lèvres une cuillerée du thé où j’avais laissé s’amollir un morceau de madeleine.
Google Translate to English: It had been many years since de Combray, everything that was not the theater and the drama of my bed, no longer existed for me, when, on a winter’s day, as I was returning home, Seeing that I was cold, suggested that I should have some tea brought against my habit. I refused at first, and I do not know why. She sent for one of those short, plump cakes called Petites Madeleines, which seem to have been molded into the grooved valve of a scallop shell. And soon, mechanically, overwhelmed by the gloomy day and the prospect of a sad morrow, I carried to my lips a spoonful of tea where I had let a piece of madeleine soften.
Moncrieff translation: Many years had elapsed during which nothing of Combray, save what was comprised in the theatre and the drama of my going to bed there, had any existence for me, when one day in winter, on my return home, my mother, seeing that I was cold, offered me some tea, a thing I did not ordinarily take. I declined at first, and then, for no particular reason, changed my mind. She sent for one of those squat, plump little cakes called “petites madeleines,” which look as though they had been moulded in the fluted valve of a scallop shell. And soon, mechanically, dispirited after a dreary day with the prospect of a depressing morrow, I raised to my lips a spoonful of the tea in which I had soaked a morsel of the cake.