Translating when you’re in the zone is like driving a Maserati through the Alps. It’s not always like that, but I know I know I’m not the only who has experienced times when words zip into place without effort and hours fly by in minutes. In his book on work satisfaction, psychologist Mihaly Csikszentmihalyi refers to this experience as flow: a state of relaxed, enjoyable engagement when you’re immersed in a task that is neither boring nor too difficult.

Now imagine you’re trying to get to an appointment during rush hour: it’s stop-start traffic all the way, you hit nothing but red lights and other drivers keep switching lanes and cutting you off.

That’s what post-editing raw machine translation output is like.

If the topic of machine translation tends to make tempers flare, this is why. Nobody wants to give up mountain roads for rush-hour hell, and you can’t blame people for getting a wee bit defensive when they’re told that resistance is futile and the last one on the MT bandwagon is a rotten egg. It reminds me of this “Fry and Laurie” skit where a minister deflects any critical questioning of his proposal by shouting that it’s “a GOOD deal for Britain! A GOOD DEAL!”

As in politics, any rational discussion of MT tends to degenerate into two extremes (Hell No I Won’t Go versus MT or Die) trying to shout each other down. Unfortunately this heartfelt frustration on both sides also obscures some interesting ideas that might be worth pondering.

As you have probably discerned by now I am not a fan of MT, so I decided it was time to take a closer look at the arguments of the pro-MT camp. I’m happy to use technology but reading about it puts me to sleep, so my impressions of “the other side” had always been limited to whatever I managed to pick up through osmosis. I figured Jost Zetzsche’s GeekSpeak column in the ATA Chronicle would be a good place to start since I knew he had written extensively on MT, so, fortified with extra-strong coffee, I delved in.

The first thing I discovered is that Zetzsche is a translator first and foremost, that he promotes technology only to the extent that it benefits translators and that he is not, in fact, an MT-industry flunky. So Jost, you have no idea who I am but please accept my apologies for misjudging you.

I also learned that there are actually two ways to use machine translation.

The obvious one we all know and hate is post-editing, i.e. fixing the raw MT output produced by someone else’s MT engine. In this scenario, the MT is the active agent and the translator cleans up the mess. I did quite a bit of post-editing during my first year as a translator, as I was unaware at the time of any controversy surrounding this issue and as a new translator I was happy to get any work at all. After a few months I stopped accepting these assignments, though, because the work was so mind-numbingly boring and frustrating, plus it dawned on me that I was getting paid less money for more work that was exponentially less enjoyable.

What was new to me is that it is also possible to use MT technology for our own purposes in combination with our own CAT tools. Two of the possibilities suggested by Zetzsche are:

• Using MT to correct fuzzy TM matches, thus increasing the match percentage of the segment
• Using MT as an autosuggest tool, where you do your own translating but the MT offers suggestions as you type.

In this scenario, the translator is the driving force and the MT just one of many tools at his or her disposal. As Zetzsche reiterates in his August 2014 column, “There really is no place for post-editing in that kind of environment”.

He also emphasizes that even this personalized use of MT is not for everyone and not suitable for every type of document. Software strings obviously lend themselves much more to tools that take advantage of repetition than, say, a marketing text or a case history in a medical journal. By the same token, translators who specialize in IT and technology are much more likely to enjoy spending lots of time customizing their CAT tools and tinkering with fuzzy matches. Their “flow” experience is the challenge of modifying their tools to create ever closer matches.

There are valid uses for MT, for example in informal settings (Facebook) or emergencies (“Please call a doctor!”). It can also be useful, as Zetszche shows, as one of many tools at the translator’s disposal.

But MT engines are still very far from the utopian ideal expressed in the 2009 White House policy paper of “automatic, highly accurate and real-time translation”, hence the need for human post-editing. I can’t blame MT coders for coding; it’s what they do and they are no doubt motivated by a communication-related higher purpose that is very similar to ours. But that doesn’t mean we have to do their job for them.

So what can we do?

Refuse post-editing jobs. There is no basis for claims that increasing numbers of translators are turning to post-editing and that eventually we’ll all have to bite the bullet if we want to survive. First of all, there are no data on how many of us currently accept post-editing work and whether or not this is an upward trend. I also read somewhere that it’s becoming harder and harder for PMs to find takers for post-editing projects because everyone hates doing it. This is certainly closer to my own experience, because PMs who offer me post-editing jobs usually come back several times to try and renegotiate rather than give this prized job to one of the other eager takers. The point is that we are under no obligation to turn doomsday claims into self-fulfilling prophecies.

Advocate for our interests. In many of his articles, Zetzsche expresses the need for ongoing dialogue with MT developers. They need to make a living as well, and if the only market they can see consists of non-translators, the product will meet the needs of non-translators only. Translators who are into technology should talk to developers and suggest features they would like to see. This would indirectly benefit all of us, as every translator-friendly feature is one more step towards transforming a post-editing tool into a translation environment tool.

This is closely related to PR, another hot-button topic. In his article on machine translation (NY Times, June 2015), Gideon Lewis-Kraus relates a conversation with a computational linguist at an MT conference about the tension between MT developers and translators. “`Go to the American Translators Association convention’, one marathon attendee told me, ‘and you’ll see — they hate us.’” This is probably pretty accurate, to be honest, but it’s not a good thing. Being known for what we’re against rather than what we’re for is a good way to relegate ourselves to the “haters gonna hate” corner, where we’ll have exactly zero influence on the very issues that affect us most. We can do better than that.