Will Machine Translation Displace Only Those Humans Who Translate Like Machines?

Here is the text of the popular Internet meme, coined by dr Arle Richard Lommel of the German Research Centre for Artificial Intelligence, himself a freelance translator:

Machine translation will displace only those humans who translate like machines

There is also a less known second sentence, which I don't see in those memes, and which reads:

Humans will focus on tasks that require intelligence.

I'm not sure if I actually disagree with dr Lommel, but I disagree with what I think the meme is likely to be typically understood to mean. Which is that only low-skill unimaginative human translators will be displaced by MT.

In the world around you, do you or do you not see MT used in the wrong place, for the wrong purpose or reasons or both?

In real life MT is used where it's not up to the task and where a human translator would do a better job.

The longer explanation is that people don't really know how translation works, human, machine or otherwise. They aren't necessarily motivated to learn, either, or even capable.

But what everybody can see — and what they are guided by in their decision-making — is that MT is cheap or free of charge, fast, or rather instantaneous, and readily available.

Choices made by clients in real life are not made the way the meme seems to imply that they are, i.e. objective reality + poetic justice. Objective reality is something they may or may not be aware of, and poetic justice is, well, poetic. So don't allow sentiments to get in the way of your logic.

Clients makes their decisions rather on the basis of what they know and what they want to do and what they ultimately do do with that knowledge.

Hence, rather, MT can displace human translators:

(i) for those clients who:

• can't afford the higher cost of human translation (if it is higher), or
• don't know the difference and don't have some other reason to stick with human translation (whatever that might be), or
• know the difference but don't see it as worth paying for, or
• know the difference, see the value but still go for MT for other reasons (whatever those might be) 

(ii) in that type of translation in which:

• MT produces results that are fit for the purpose, and
• human translation doesn't offer any benefit the client is interested in (or, more precisely, willing to pay for), whether that's more-than-fit quality or something else

This could be understood to mean that in at least some of those cases we all translate like machines. But that's not necessarily true — it only means we achieve the same results that a machine can achieve.

And machines can do amazing things with enough processing power and the ingenuity of, surprise, surprise, human engineers.

Next, 'even' structurally bound rule-based translation that relies on substitution as a method is not necessarily a task that doesn't require intelligence. It's only a task where artificial intelligence may be sufficient and in some cases superior. Just because Kasparov lost to Deep Blue doesn't mean chess doesn't require intelligence, does it?

Finally, MT with human post-editing (PEMTing) is not the same as MT without it, in terms of risks. Human-assisted MT is more of a threat than pure MT.

Make no mistake:

• Just because you don't see yourself as a low-skill translator, or just because you objectively aren't one, doesn't mean that rules-based MT with good input data, good rules, a lot of processing power (millions of operations per minute etc.) and especially human assistance (PEMTing or otherwise) can't produce results that are x% as good as yours in x% of cases, for example 80% as good in 80% of the cases. Don't kid yourself, not everybody wants or even needs[1] the double 100% figure, nor can you actually achieve it (it's humanly impossible). It may be helpful to read up on bounded rationality, i.e. full optimization versus satisficing, and maximizers versus satisficers. As a translator, you are probably more likely to be a maximizer than a satisficer (100% pure maximizers probably don't exist, so it's not a 0/1 matter, more like, say, 0.9/0.1). An economist, business manager or IT person — or PM for that matter — is more likely to be a satisficer than you are (there are exceptions, some of them notably quite annoying, e.g. in the literal reading of some translation agency contracts only 100% of the best theoretically possible result will do, else they'll reject your translation; how does that feel?). Most people use simplified heuristics in their decision-making. This means they make 'approximate decisions' and accept certain risks in exchange for time and cost savings. You aren't immune to the 'like machine' just because you're good, let alone if you should be merely not too bad at what you do.
• In real life MT displaces human translation just like bad human translators displace good human translators. Read Paul Sulzberger's interview with Luigi Muzii to get a better understanding why (Gresham's law: bad money drives out good), though it's not just the price. Some of the people who would otherwise be your clients will go for MT just like some of them go for cheaper translators or DIY translation, for all sorts of reasons.

Get proper context. You're only immune if a machine can't translate like you. And if you can convince your clients that the difference is worth paying for.

---

[1] It's important to note that fit-for-purpose quality — one of the core concepts behind MT, or at least behind its more responsible incarnation — is not fundamentally flawed. Simply put, it's not true that nothing is good enough unless it's absolute best. We can argue how good is good enough, or how beneficial it would be to go beyond the enough. We can — and should — protest when the bar is set too low. But we can't pretend that only the absolute best is ever good enough. Reductio ad absurdum: Most of us would be unemployed if only the best translators and translations were good enough for any client and any job. And most of us would be spending n times more time editing and QA'ing.