The Mechanical Turk, or Chess Turk, was a fake machine that played chess against human players who almost always got beaten by the ghostly machine. The ingenious contraption, constructed in the late 18th century, was very popular in Europe for about eight decades until people finally figured out that the desk behind which the formidable Chess Turk automaton was sitting had enough empty space in it, camouflaged by useless gear, to hide a human chess player there. If you click on the introductory Youtube video, it will tell you the story of the Mechanical Turk in dramatic and authoritative German accent.
But if you Google the words Mechanical Turk two centuries later in early 21st century, the first few entries that you will find will not be for the old fake machine that had a sly chess master hidden in its entrails. Instead, you will see several entries for the term “Amazon Mechanical Turk”, described in Wikipedia as “A crowdsourcing Internet marketplace that enables individuals and businesses (known as Requesters) to coordinate the use of human intelligence to perform tasks that computers are currently unable to do” …. “Employers are able to post jobs known as HITs (Human Intelligence Tasks), such as choosing the best among several photographs of a storefront, writing product descriptions, or identifying performers on music CDs. Workers (called Providers in Mechanical Turk’s Terms of Service, or, more colloquially, Turkers) can then browse among existing jobs and complete them for a monetary payment set by the employer“ (emphasis mine).
The concept of Turkers is similar to the concept of online portals for translators who can on a good day (but is it really a good day?) find work on these portals, mostly at incredibly low rates, although these rates are still higher than what crowd workers who work for companies such as Amazon, called Turkers, are being paid. Turkers earn on average about 2 dollars an hour, typically about 0.001 dollars per task, but unlike translators who want to find work on poorly paying translation portals, Amazon Turkers do not need to pay a membership fee to a portal. All they have to do is create an Amazon Turker account and they can start bringing home the bacon immediately, although it will be only a very tiny piece of bacon that will still leave them hungry for more food.
There are many very simple things that machines, no matter how incredibly fast they may run their calculations, cannot understand. That is why we always have to prove online that we are humans and not just robots, called web crawlers, looking for information. Our genuine humanness is now mostly tested when we are asked to identify a string of numbers or letters in which some of them may be in a different font or askew. Even the dumbest human can notice something like that right away, while even the fastest and most powerful computer will fail at this easy task.
The tasks that Turkers perform are very simple. Provided that you are in fact a human rather than a machine, you will be able to quickly conclude that green grass looks better on a real estate ad than yellow grass, you will be able to easily find contact information hiding somewhere in a web page, determine that a telephone number or zip code has too many or too few digits, or tell which girl is pretty and which one is a dog (although different humans will have differing opinions when it comes to the last task).
There are many people who are willing to work for such a pitifully low remuneration in this world, as they have plenty of time on their hands, typically because they cannot find a better job.
And in quite a few countries in the this world, 2 dollars an hour is nothing to sneeze at.
According to BusinessInsider.com, the minimum hourly wage is below 1 US dollar in the following countries: Sierra Leone ($0.03), India ($0.28), Afghanistan ($0.57), the Philippines ($0.61), Mexico ($0.66), China ($0.80) and Russia ($0.98), while the minimum wage in Brazil ($1.98), corresponds roughly to the average hourly wage of a Turker who may be working for Amazon, Target, or Walmart or another corporation from anywhere in the world (the numbers are from 2013, but I don’t think they budged much since then).
The way translation industry, or at least a certain segment of it, sees the arrangement of the natural order in the world, the task that translators perform, for example during post-processing of machine translations, is not very different from and typically not much more difficult than what non-translating Turkers are doing, when certain tools, called language technology tools, are employed.
That is why the translation industry is so excited about what it calls language technology!
Language technology, or processing of human language with computer tools such as spell checkers and word counters, computer assisted translation (CAT) tools, optical character recognition (OCR), machine translation, speech to text conversion and many other computerized tools has been with us for many years, some of them for decades.
One relatively recent language technology tool that I absolutely adore is called telephone voice caller ID. I bought two Panasonic phone systems with several extensions, a black one for my business and a white one for my home, both with the same phone voice caller ID system. Unlike last year when I still had to get up in order to look at the call ID displayed on the phone, now I don’t have to get up from my chair or sofa when telemarketers in vain attempt to disturb the serenity of my day.
But the translation industry, for lack of a better term, is not interested that much in language technology tools such as spell checkers, or speech to text conversion. Maybe a little bit, but not that much.
The translation industry is mostly interested in the one language technology tool that looks the most like the Mechanical Turk that was invented at the end of the 18th century, namely machine translation combined with post-processing of the result of machine pseudo-translation, once it has been fixed and straightened up by translating Turkers.
But there is a problem with the concept of replacing a human translator by machine translation, so that a fast but still not quite human-like machine is then assisted by humans having the function of translating Turkers. While Turkers who work for close to nothing on very simply tasks as they fix computer errors for large corporations do not need to know much about anything as long as they have a pulse, a computer and Internet access, translator-Turkers would need to know something to be able to fix machine translation errors.
A lot, in fact, because they would need to know basically as much as a human translator must need to do good work.
We keep hearing from what is called the translation industry that machine translation is being constantly improved, which is true as far as that goes. The way machine translation, or pseudo-translation to be more precise, is described by merchants of language technology, the only, relatively minor problem with machine translation is that “it is not perfect yet, or “not as elegant” as human translation.
The way results of machine translation are described by merchants who are so excited about their new language tools, all that is needed is to hook up online with a bunch of idle translator-Turkers, who will then be able to fix little details that machines don’t understand yet.
Post-processing of machine translation is not that different, according to this theory, from determining the correct number of digits in a telephone number, or whether something is in green or yellow color, or which girl is pretty, and which is not.
That is the version of post-processing operations that is being sold to translators who might be interested in becoming post-processing Turkers in the current version of what is called the translation industry.
The reality, however, is something else altogether. The real job of post-processors of machine translation is to identify and fix mistranslation, not just to look for careless, stupid, but relatively minor computer mistakes in order to make the text more elegant, more idiomatic, or just slightly better.
Their job could be compared to what in the home remodeling industry is called “a gut job”. If you ever saw one of the reality TV shows about home buying and remodeling, you know that a gut job is what you are left with when you buy a house because of the famous location, location, location, because it has “good bones”, and because it is really cheap.
Let’s say that you buy an old and decrepit house for a hundred thousand, you remove everything from the kitchen, the bathroom and the bedrooms and invest another forty thousand in upgrading everything that was taken out. After a few weeks or months during which you are kept very busy working on your gut job, you may end up with a house that is now worth two hundred thousand, at least according to what they tell us on teevee – if you know what you’re doing.
That, rather than just fixing minor errors, is also how post-processing of machine translation looks in reality.
Coming back to our original analogy of Mechanical Turk, or Chess Turk, there would need to be many invisible translating Turkers hidden in the magical box that is being built for translators by the translation industry, in which the translation industry would love to marry computer technology with post-processing humans. And most of these translating Turkers would need to be as good as the sly chess master who, hidden in a desk behind which a scary figurine of a Turk was pretending to blink, shake his head and move his hands, was busy moving chess pieces with magnets under the desktop.
Since the magical machine of the translation tools will not work unless great multitudes of translating Turkers can be hidden inside the box that the translation industry is busy building for them now, the interesting question is: how many translators will fit into this box?