Blog
Machine Translation Report
What is the optimal MT Engine for you? Find out in the latest MT Report by Memsource.
There’s been a lot of buzz around artificial intelligence and machine learning lately. Both technologies are being heavily invested in, and their practical use cases range from gimmicky to life-changing. Depending on who you ask, emotions surrounding the topic range from optimism to fear. Freelance translators often loathe the rise of machine translation engines - automated systems that can translate entire documents by themselves, and quite well too. These engines can pose a threat to our livelihood, and sketch a possible future where humans merely function as guidance, while machines do the real grunt work. The same work that we as translators love so much.
On the other side of the spectrum are, for example, translation agencies. This is not a knock on translation agencies in general, but there are several which are exploring the possibilities of using machine translation engines as a primary source of translation, with humans checking the result to make sure the final product is decent. The benefits? Lower costs and bigger margins, obviously, but also a less time-consuming translation process, as machines translate much faster than any translator of flesh and blood ever could.
This leaves many a translator fearing what will become of their job in the future. However, I believe the concept of machine translation has three impossible obstacles to first overcome: understanding context, understanding culture, and keeping up with cultural and linguistic developments.
Context. A translator needs to have a certain degree of background information and context in order to produce a translation that accurately portrays the source text. This is inherently difficult - if not impossible - for software to deal with, because it can’t think for itself (yet). For the sake of this post, let’s imagine that we have reached a point where it can. It’ll still have to deal with ambiguity, which is inevitable with languages, as well as badly written source texts. How will it make decisions in these kinds of situations, and how would we know that these decisions are even correct? The answer is a human set of eyes. Even well-trained, professional, human translators make mistakes here, despite all of their cognitive skills. Expecting an algorithm to manage this all just might be too much to ask.
Culture. Translating is far more than looking up what ‘X’ in language ‘A’ means in language ‘B’. Some things are exclusively tied to a particular culture. Consider Groundhog Day in the US for example, or ‘spekhappen,’ an old Dutch game, often played at kids’ birthday parties. The latter even comes with a cultural connotation: a feeling that it evokes among a specific group of people. How will software deal with this? Human translators can rely on their judgment to decide whether to keep the original term and add an explanation or to replace it with a similar phenomenon that their audience is familiar with. Sure, you could theoretically brute force your way to a successful machine translation engine by jamming every connotation for every word in every language into an algorithm, but that would be neither profitable nor feasible. You simply need a human to not only feel this connotation but also to know how to act upon it. In the field of marketing translation, this is crucial. As a translator, you need to feel what the author wants the audience to feel, so you can write something that can recreate the same feeling among a group of people from a completely different culture.
The third and final hurdle is cultural and linguistic development. Culture and language are living, breathing things. You can’t expect them to sit still - they develop constantly and at a rapid pace, which will only increase through technological advances. Keeping a machine translation engine up to date with both cultural and linguistic developments would be a nightmare for just one language pair - now imagine doing this for the odd 6,000 languages spoken across the world. For individual translators, this is a non-issue. We live, sleep, eat and breathe our working languages. I know about filibustering, outfits which are ‘on fleek’ and ‘throwing shade’, just like I know about ‘blokkeerfriezen’ and ‘excuustruzen’. I don’t require updates for idioms or ever changing slang. Whether or not a self-learning algorithm could ever do this as well is unsure, but it would be one heck of a job to develop one that does so for each existing language.
While I’m personally not a huge fan of machine translation, I cannot deny that there certainly is a place for it within the translation industry. Not every company needs a top-notch, watertight, and culturally appropriate translation. Some companies just want a translation. For them, this technology will be a welcomed addition to the offering that is out there today. For companies that insist on a certain level of quality though, translators of flesh and blood will remain a necessity. They aren’t just trained to master several languages, they’re trained to translate, which involves constantly making complex decisions. This requires knowledge, experience, and intuition – and at least for now, there is no engine that can successfully combine these three factors.
About the Author
Milan Verwers is a Dutch freelance translator who graduated from the Maastricht School of Translation and Interpretation. He loved languages as a child and since then, nothing has changed. His linguistic endeavors now include translating financial documents and marketing texts, reading newspaper articles, and writing the occasional article or poem.
