As one of the main areas of AI research, automatic natural language processing would seem to spell doom for human translators. But maybe not, writes associate professor Sylvie Vandaele.
Machine translation is a complicated affair. Back in the 1950s, it was already one of the strategic battles of the Cold War. In this century, deep learning entered the picture, with Google releasing a new, neural-network-based version of its online automatic translation tool in 2016. In 2017, Linguee, a German firm that has since been renamed DeepL, surprised everyone when it released an online system that was quickly hailed as the best in the world.
Université de Montréal recently achieved its own automatic translation milestone: Deep Learning, a textbook written by Professor Yoshua Bengio with Ian Goodfellow and Aaron Courville, was translated into French in record time thanks to a joint effort between DeepL and Quantmetry, a Paris-based company. This achievement prompted some media sources to conclude that human translators are becoming a thing of the past. According to Nicolas Bousquet, scientific director of Quantmetry, this is especially true for those who translate such everyday documents as vacuum-cleaner manuals.
But this reductive view doesn’t consider the many different kinds of contexts and specialized fields that translators work in. It’s based on a cliché that categorizes all writing as either literature or “the rest,” as though nothing lay between manuals and literature, as though scientific and technical writing were not themselves products of the cultures from which they emerge.
Context matters a lot
All translations must still be checked by a human – and a skilled one at that. The corpus of texts used to “train” machines are the key to successful automatic translation. These texts must be extensive and of high quality. AI algorithms formulate translations based on the corpus. In real life, however, context has a huge impact on how the words in a text are interpreted, and translating loosely is often the only way to effectively convey the intended meaning. And for that, skilled human translators are needed.
This is precisely why employers and clients should think twice before concluding that the increasing potential of AI justifies cutting employee salaries or freelance rates. While machine translation can drive down costs by making the process faster, properly post-editing the translation remains an essential investment.
Following the French translation of their seminal work, Bengio and his colleagues put forth an interesting hypothesis that could be of particular interest to Quebecers. They noted that effective machine translation could lead to a resurgence of interest in using the world's various languages– including French – in science and technology. This, in turn, could weaken the dominance of English and result in a translation boom, giving the scientific community access to ideas drawn from a wide range of languages and cultures.
So instead of spelling the end of human translation, the trend towards automation could actually see professional translators, particularly those who specialize in specific fields, serving more clients interested in receiving top-quality translations.