Home » Blog » Machine translation and human clangers

Machine translation and human clangers

The New York Times has recently run a comparison of Web translation tools.  Over the last few weeks I have had reason to look at some Italian sites, and the existence of these tools has made me realise that there cannot be much money left in the business of “gist” translations. What has reassured me, however, as someone who makes a living from translation, is that even now these tools often fail even to give a “gist”.  Of course, we can expect the results to improve, and possibly to do so quite fast, but in many cases, especially if the source is at all complex, the result is near gibberish.  Consider, for instance, the word “provanti”, which does indeed seem to be some kind of Italian word.  But what does it mean?  Attempts to use the Google translator on phrases containing this word yield gibberish: “Casa Provanti”, for instance, is yielded as “home hard to deal with”, and other trials show that at the moment the translation engine believes that “provanti” means “hard to deal with”. No, no, no, no, no.  It is a participle associated with a verb for attempting or trying, as in “I am trying to please my guests”.  The engine seems to have fallen for what I recall as a schoolboy joke:

“Can’t you do that with a bit more effort?”
“Sorry, I’m trying.”
“Yes you are, very!”
(Boom, boom!)

But humans make mistakes too. Consider, for instance, a site to which I was recently referred, in the “languagering”, offering ” Training in writting”, and telling us that “writing can be a tedious tasks”.  It would appear that for that writer, checking spelling and proofreading were just too tedious by far!

And here is another, although the author of this one can be perhaps forgiven, as the text was submitted for proofreading and correction.  I will not name the source, as it is a client who pays me. It illustrates the mistake that can be made by some Germans (and no doubt those with other native tongues) who plan to save money by doing the “translation” themselves, then paying a native of the target for the proofreading only.  The problem is, of course, that the difficulties created by this process mean that the “proofreading and correction” may demand more time and money than simply translating in the first place.  The source text was “Bei Werkstattmontage gebohrt”. For those who don’t know German, “bei” is related to the English “by”, and carries meanings like “in association with” or “at the same time as”, as well as “next to” and so on.  It is not used, however, to convey agency in the same way as in “I was knocked down by a car”. “Werkstattmontage” is a simple example of a German compound noun, and can uncontroversially be translated as “workshop assembly”. “Bohren” is to make a hole, as into “bore into a piece of leather”; in engineering contexts it is most often translatable as to “drill”, and a “Bohrung” is a drilled hole.  So our phrase is a comment on a hole, and tells us that it is “Drilled during workshop assembly“. Our German “translator”, however, had rendered it as (wait for it…): “Bored by workshop assembly“.  They probably were!

Disclaimer: having criticised the spelling and command of language of others it is a cosmic law (WIP – Wilding’s Inevitability Principle) that I have made at least one silly mistake in this article.  Don’t blame me!


  1. A machine translator produces the translation, using rules that transfer the grammatical structure from the origin language to the destination language. A good translation depends on the quality of the analysis of the translation and of the size of the used dictionary.

  2. Alex says:

    But, “Proofreader”, the reality is that in most cases they don’t actually do that. You’d think they would, and maybe one day they will, but at the moment *real* grammar, as opposed to some specially restricted style, cannot reliably yet be understood by machines.

Leave a Reply

Your email address will not be published. Required fields are marked *