Why human-in-the-loop (HITL) doesn't work for translation

We keep hearing about HITL, or human in the loop, but how many of us have asked ourselves what it means for translation?

Let’s take this definition and break it down in the context of MTPE:

“Human-in-the-Loop aims to achieve what neither a human being nor a machine can achieve on their own. When a machine isn’t able to solve a problem, humans need to step in and intervene. This process results in the creation of a continuous feedback loop. With constant feedback, the algorithm learns and produces better results every time.” (faculty.ai)

Human-in-the-Loop aims to achieve what neither a human being nor a machine can achieve on their own.

Professional translators are quite capable of translating on their own without machines. Last time I checked, MT wasn’t a thing in Mesopotamia. So we are the sine qua non here, not the machine. Can we do it at the same speed without machines? No, but all that means is that machines are tools. Our tools. Nobody else’s.

When a machine isn't able to solve a problem, humans need to step in and intervene.

To know whether we need to step in and intervene, we must spend time reading and often rereading the machine output.

That time is uncompensated under MTPE rates because that’s a business model created to line someone else’s pocket.

The work doesn’t start when we start changing the output; it starts when we start reading it.

And sometimes, the output doesn’t need changing at all.

Sometimes it’s as good as—or even better than—we would have come up with ourselves. Yes, it happens. We need to leave our egos at the door.

So, what do we do? Tweak it because we don’t want to be accused of retrieving instead of translating? Or do what a good proofreader does? Don’t mess with what doesn’t need fixing. I say the latter.

Of course, the next segment may contain omissions, hallucinations, and clunky phrasing.

The point is, it all takes time!

With constant feedback, the algorithm learns and produces better results every time.

No, this assumes that specialist translators are post-editing the machine translation. But they’re not—the rates are too low.

MTPE work is mainly outsourced to younger translators trying to build their careers. Specialised translators largely refuse the work; some even plan to leave the profession or diversify.

Plus, people get bored working with MTPE, lulled into a false sense of security. They miss lots of mistakes.

There’s little to no quality control at these large MTPE-only agencies.

The proverbial dog is chasing its tail.

Takeaway

View the loop as our workflow and decide if and when to use MT as one tool in it.