Thoughts on Computer-Generated Translation

Last week, Smartling’s CEO Jack Welde gave a presentation in which he used a term that caught my attention – “computer-generated translation.” This term is simple, straightforward, and easy to understand. However, we rarely hear it within the narrow confines of the translation industry. It caused me to reflect on the terms we more commonly use in the industry, and whether or not “computer-generated translation” might have some advantages that other terms do not.

A much more common term is “machine translation,” which harks back to the 1950s and 1960s, when researchers at various universities in the U.S., Japan, and Russia began in earnest to work on systems that would automate translation. Back then, it made sense to use such a term. After all, the systems used truly qualified as “machines” – as bulky and clunky as we would imagine them to be. (For proof, see this video from a prior blog post on the history of machine translation).

However, there’s something I’ve never liked about the term “machine translation.” It conjures up the outdated idea of man versus machine, when society has long since moved past that notion and accepted that machines are only as useful as humans make them. To me, “machine translation” sounds a bit antiquated, not in keeping with modern times. Worse yet, when we talk to others about our field, this term does us a disservice. The idea of machines completely taking over a process – any process – can sound a little scary to people. It can lead people to believe that humans are not in control whatsoever, even though, of course, humans are behind the building and refinement of all such tools.

Another somewhat common term in the industry is “automated translation” or “automatic translation.” The problem with this term, in my opinion, is that it oversimplifies the process it describes. While the translation itself may happen in an automated way, again, there are humans that enable that automation. And, unlike an automatic door, which opens the same way time and time again via a simple process, even when translation is automated, it’s rarely as repetitive, simple, or reliable. “Automated translation” makes things sound a little too easy. People should be warned that while this process is valuable and has important applications, it is actually quite complex.

When I heard the term “computer-generated translation,” my ears perked up in a positive way, because it reminded me of a much more common term, CGI, or computer-generated imagery. When the average person thinks of CGI, they think of the kinds of images and simulations that are found in video games and movies. It’s rare for someone to think that CGI is “as good as the real thing.” Everyone understands that it’s only an approximation, a simulation, and will never be a replacement for the beauty and art found in real life. In fact, even when people compliment CGI, they tend to add a caveat, in full acknowledgment of the limitations of computers, such as “pretty impressive effects, considering they were all CGI.”

The term “computer-generated translation” conjures up similar connotations. The word “computer” is much friendlier and more accessible than “machine,” and the word “generated” implies a process that is longer and more complex than “automatic” does. To me, the idea of “computer-generated translation” brings to mind the idea of something that is not a replacement for human work, just like CGI is not a replacement for the world’s greatest cinematographers. With a term like “computer-generated translation,” it becomes clearer to people that computers are merely attempting to do something that humans can do better.

Of course, as with CGI, computer-generated translation also fits into areas and offers certain advantages that humans cannot. And, because it is seen as something so vastly different from human work, we don’t blink when human beings are given an Academy Award for their breathtaking applications of CGI in films. That doesn’t mean that human beings no longer receive Oscars for non-computer-related work, or that it’s any less important. People understand that these skills fall into separate categories.

It also strikes me as interesting that we so commonly use the term “computer-assisted translation” to refer to translations that benefit from computer software. Is it a stretch to refer to translations that rely almost entirely on software as “computer-generated?” Actually, it seems like a pretty logical linguistic leap.

In our industry’s ongoing quest to educate others on the differences between humans who translate website or app content and translations that are generated by computers, let’s ask ourselves whether the words we use to describe these two very different processes are helping us or hurting us in achieving our objective – and, whether there might be better alternatives out there.

About Nataly Kelly

Nataly brings nearly two decades of translation industry experience to Smartling, most recently as Chief Research Officer at industry research firm Common Sense Advisory. Previously, she held positions at AT&T Language Line and NetworkOmni (acquired by Language Line), where she oversaw product development. A veteran translator and certified court interpreter for Spanish, she has formally studied seven languages, and is currently learning Irish. A former Fulbright scholar in sociolinguistics, Nataly lives in the Boston area with her husband and daughter. When she isn’t working, you’ll usually find her translating Ecuadorian poetry, writing books, and exploring the world (36 countries and counting!).