Unreasonable Expectations for Machine Translation Are Based on Blind Faith in the Infallibility of Technology

Even the best, almost miraculous modern technology, has its limitations.

I love the way the GPS system can guide me to my destination through my iPhone because I no longer have to study a paper map before I start my car.

But the GPS sometime picks the wrong route for me, a much longer route than a local would choose because it generally picks the main roads, even if I try to exclude highways. It took me and left me near an airport, but not at the airport where I wanted to go, while insisting that the destination was on my left. It wasn’t.

Once when I was looking for a UPS (United Parcel Service) store, it directed me to a parking lot of a high school and insisted again that the destination was on my left. It wasn’t. On my left was a parent directing other parents to the few remaining free parking spots left in a high school’s parking lot. After I managed to find another UPS store, 15 minutes away from that school, the store’s clerk explained to me that the other UPS store was in a shopping mall that was quite familiar to me, but not to GPS, about a mile south from the high school. Somehow I never noticed that the store was there.

“Proceed to the route, proceed to the route, proceed to the route”, the voice in the machine keeps admonishing me if I dare to ignore its instructions because I know better that to trust a machine.

Yet, many people have an almost fanatical belief in technology, they trust it completely and unquestioningly, the way hardline communists used to believe in the righteousness of their own system, or the way religious nuts believe that their version of religion is the answer to everything.

“The translation industry” is shamelessly exploiting this belief in omnipotence of technology.

The industry posits as an irrefutable fact the highly nonsensical proposition that as machine pseudo-translation is constantly being improved, human translators will eventually, perhaps quite soon, become an unnecessary, expensive appendage that can be simply removed completely with the scalpel of cutting-edge “language technology”. According to the industry’s credo, the only way forward for translators is “to embrace technology”, by which they mean that translators must incorporate machine pseudo-translation into their own business model, preferably by becoming post-processing factotums ready and willing to be exploited by “the translation industry”.

The problem is, this line of reasoning simply ignores reality.

The reality is that machine translation has been around for more than half a century and that it can now be accessed for free with a few mouse clicks by about a billion people as I am writing these words.

Reasonably good machine translation, within certain limits, has been available to this patent translator at official websites of the of the European Patent Office, Japan Patent Office and other websites containing digital libraries with millions of patent applications in dozens of languages for at least the last 15 years. When I was translating a patent ten or fifteen years ago, I only had a single file on my computer’s desktop. I now need to have a folder on my desktop for each patent that I am translating, not just a file, and one of the files in that folder, which contains a number of reference files, is usually a machine-translated version of the same patent that I am translating.

All of my clients have had access to the same machine-translated reference files for at least a decade. But although machine translation, or more precisely speaking, machine-pseudo translation, has been available for free on the Internet for more than a decade, it has not been able to replace me and thousands of other human patent translators. Fifteen years ago, I was afraid of it, and I thought it possible that the day might in fact come sooner or later.

About five years ago, a troll on my blog told me that machine translation would put me out of business within five years. He kept repeating the same thing over and over like a mantra, evidently drawing comfort and some kind of pleasure from a statement that was based again on nothing but wishful thinking. After a while I got tired of trying to reason with him and instead started deleting his comments. He went away for a while, only to come back under a new pseudonym, but after I pointed out to him that I can see through his most recent sock puppet reincarnation, he finally went away, hopefully for good.

Well, I am still here, doing the same thing that I was doing 15 years ago, 10 years ago, and 5 years ago, and I will still be here doing the same thing 5 or 10 or 15 years from now, unless I die or decide to retire.

All of my clients have had access to machine-translated files for at least a decade now. And yet, they still keep me busy, even in the midst of the worst economic depression that has been with us for at least the last 8 years.

Machine pseudo-translation is not going to change my line of work that much. It is useful for many things, for example for identifying patents that do not need to be translated by a human. But its capabilities seem to have reached yet another a plateau. The statistical method developed by Google is much better than the decades-old, rule-based attempts to use software to simulate human thinking. But why is Google’s approach so much better? Because it is based on human thinking. A sentence that is so cleverly translated by software sounds perfect only because it was originally translated by a clever human.

An algorithm will sometime find a perfect or almost perfect match for another sentence in another language. But often, perhaps due to a difference in one word, or a misplaced comma, it will say the opposite of what was in fact meant in that other language. How do you overcome this dilemma?

The same way that I found the location of a UPS store that was invisible to the GPS in my iPhone when the wrong location was for some reason associated with its address by the software.

When the software is wrong, you need to ask a knowledgeable human being for directions. That is why “the translation industry” needs translators to fix errors created by software. The truth is that it needs them to retranslate the whole thing, but not as translators. Once translators are reclassified as post-processors, they will be inevitably paid a low hourly rate to help to drive the profits of “the translation industry” into the stratosphere.

Some translators, the most vulnerable among us, will no doubt fall into this trap, without realizing that they are digging their own grave in this manner if they accept this greedy scheme as their own, unavoidable fate.

I could be wrong, but something is telling me that most of us are not that stupid.