Post-Editing of Machine Translations Is a Good Definition of the Term “A Fool’s Errand”

Fool’s errand:

 1. A foolish undertaking, especially one that is purposeless, fruitless, nonsensical or certain to fail.

 2. Such an undertaking assigned as a prank.

 (Source: The Viktionary, open-content dictionary).

I happen to know something about machine translation (MT) as I have been using it in my line of work, translation of patents from Japanese and other languages, for about the last 15 years.

And yet, although more often than not these days I do have a machine-translated version of a patent application in Japanese, German, French or another language available to me, I have never post-edited the MT product. I only use MT more or less as a pretty good, context-based dictionary. Plus it’s free. You can’t beat that price.

Would I save time if  I could simply edit the MT product instead of having to retranslate everything from scratch? I am sure that I would. Then why don’t I do it?

Because I am an irrational, intransigent, conceited, self-absorbed, mad patent translator who can’t be reasoned with?

It’s not really up to me to say, I may well be all of the above (although I would take exception to the word “irrational”), but that’s not why I don’t post-edit MT. I don’t do it for the following two reasons:

1. I want my clients to continue sending work to me, and most people are smart enough to be able to tell post-edited MT from real translation. My clients pay me for translating, not post-editing.

2. If I tried to massage MT with my editing to lick it into a shape that would resemble human translation, it would take me at least as long as retranslating because ….. MT post-editing that can add real value is in fact retranslating. Even when you have most of the right words and some of the sentences that have been translated correctly in the MT product, you still have to rearrange everything and change a lot of things, which is very time consuming.

If you start pulling out bricks from a house that was poorly built, it will fall down. The same is true about post-editing of machine translations. It is much better and also usually faster to build the house of translation based on a blueprint that can be designed only in a fallible translator’s head, rather than building it based on algorithms of infallible and quite arrogant mathematical geniuses who don’t seem to have any understanding of how languages work. If they did, they would be working on something that should be easier to achieve, perpetuum mobile for instance, or how to turn water into gold.

It does not really matter how large the “corpus”, “database”, or “content tsunami” (the vocabulary available to the hardware) is and how fast the microprocessors are these days.

The problem is, instead of working with the meaning, which is how the human mind works, machines work with algorithms.

Algorithms are and always will be an insufficient substitute for that precious blueprint formed by billions of neurons and synapses firing seemingly at random in human brain. This is a problem that can be solved only if somebody can design an algorithm that will replace the role that meaning plays in human mind and human languages.

The way I see it, it should easier to bring the dead back to life, which is something that happens only in scary movies like The Flatliners, and in a few books like the Bible or Mary Shelly’s Frankenstein.

Post-editing of machine translations could be used to improve texts that are relatively unimportant. But I don’t see a bright future for post-editing of machine translations of texts that nobody would pay any money to have them written or translated in the first place, such as this, hopefully somewhat entertaining, but otherwise unimportant post. Why pay any money at all to translate something that is of no consequence?

Nor do I see promising future for application of post-editing by humans to patents that were originally translated by machines.

So far I have been asked to perform this task twice: once about 12 years ago when MT was still kind of new, by a patent lawyer who spoke with authority and had a deep voice. He must have been a partner. I mentioned it in this article that I wrote back then if you want to read it (but it’s pretty long).

The second time was a couple of months ago. Some guy, a private individual who found my website, asked me for a quote for translating a long Japanese patent. When I sent him my quote for over two thousand dollars, he replied that he already had in his possession a machine translation of the document and that he would be willing to pay me 400 dollars for post-editing.

I declined both job offers, the first one politely, and the second one rather rudely. I don’t like people who try to play the bait-and-switch con game with me.

The thing is, the patents that I translate are kind of important. A lot of money is often at stake, which is why the translation must be as accurate as possible. Post-edited machine translations simply won’t do.

Patent lawyers use my translations to file new patent applications for their clients, or to argue about fine (and sometime nonexistent) technical differences and details for months or years before one or the other corporation finally loses a lawsuit and has to pay royalties.

Most often, though, lawsuits involving patents with complicated technology have the potential of going on for such a long time and being so expensive that both parties find it easier and cheaper to settle out of court.

I am trying now to think of good examples of promising applications for post-editing of machine translations. Something that is not important enough to pay humans to translate it, but important enough to pay humans to try to edit mechanical attempts at translation.

Perhaps there are, perhaps there must be things like that out there.

But I can’t think of any.

Maybe you can.