MACHINE TRANSLATING


I was curious about something I read: that there existed some specimens, self-proclaimed translators, that translated jobs using machine translation, as a way to sell very cheap services without really doing any work. This curiosity took me to do an experiment with the kind of texts that might be easier to translate by a machine, that is, texts that are straightforward or don’t have practically any optional vocabulary or varieties that might confuse a machine. I must say I’m not a specialist in linguistics or grammar, things that in school I always found hideous and a pain wherever. Maybe a specialist would be able to perform a much more analytical experiment, but the results I have obtained I believe are enough to put any translator off. The difference between translating or proofreading a well written text from beginning to end engaging with the feelings of who wrote it, understanding and preserving in your mental hard disk new knowledge useful for future assignments or even for one’s own life, has no comparison with the painful gamoogaboofpigswillhogwash that entails revising something full of accidents and usually incomprehensible. And that is what one would have to do if one had a minimum of respect for oneself and the profession. But the other specimens don’t have that, they just send it off. Then the client, if he’s clever, he’ll have it read by someone who knows the language, and he won’t pay. Obvious. He’ll have to pay a good translator the value of a new translation by what it’s worth. Oh, sure, he’ll say: “but it’s only a proofreading/editing/reorganizing/gap discovering-filling in/cutting/trimming/deleting job, you should charge less”. In the end it would be cheaper to translate it all over again, and leave out the pain.

I have translated the following texts:
- A cover letter for a job application. I haven’t included a Human Translation to compare because the results of the Machine Translation aren’t worth comparing but with the contents of my trash can.
- Parts of Data Sheets for fertilizers. Technical texts of this type usually follow a language that is a standard in the industry so one shouldn’t find very different translations of a given concept.

RESULTS

I have put the different translations from three Machine Translators (XXX; YYY and ZZZ) in a table side-by-side with the English original. And in the technical text I included my own translation as, would-be, “control sample” or “Human Translation” with which to compare. I have eliminated from the technical texts those sections that were “correctly” translated by the Machines, because this experiment was designed to be skewed in the sense that I am not interested in promoting Machine Translation, not because I’m afraid of it but because it’s incompetent. In any case these sections don’t represent more than 29% of the total text translated. These correct sections usually had no more than one, two or three words each, except for two cases that surprisingly had 8 and 20 words in the sentence.
The Machine Translators are named anonymously due to the skewed design, but whoever has time to waste can discover their names repeating my experiment, although I don’t recommend it, it could be bad for your digestion.

You can find the tables at www.muestraweb.webs.com


CONCLUSION

Some people, mostly engineers and stakeholders, believe that Machine Translation, some day (soon they say), will take over technical translation. Maybe they are right. Let’s leave it to a reasonable doubt. At present it has nothing or very little to offer. Just turn the page and call a translator, a human being. Someone that can adapt to any circumstance.