Machine translation - Will it Blend?

Anyone who has seen the eye-watering iPad destruction on the geek site Will it Blend? (link at the end of this article) will know how destructive blending something can be.

The concept of mixing or 'blending' machine and human translation, ostensibly to save on costs, is not a new one, but it hasn’t exactly taken off either. While the concept seems to make sense, the practice, for the most part, does not.

So why can't you just 'repair' a machine translation?

We set out to discover if editing or 'brushing up' a machine translation could save any time at all (or money) and to find out if the results would be any good.

We chose a short, straight-forward English source text of a general-nature that included some technical terms. Specifically, an article on sports for youngsters with heart disease. 

We pre-translated the text into German using Google. The results at first glace seemed promising. Single medical terms for instance were quite good, but then, any conscientious translator would double-check these terms so this is not such a great time-saver in reality.

Going through the Google-translated sentences, however, it quickly became clear that virtually none of the sentences made sense. In fact, in many cases they were quite unintelligible because the software is incapable of making logical connections between the individual words.

To cite just one example, a heading 'Sport matters' was rendered as 'Sport affairs' (in German) as Google hadn't recognised that 'matters' was a verb. While it was possible to glean some of the meaning from the machine translation, there were no three words in a row that could be used in a professional context. The main reason being that machines consistently fail to correctly identify the word types (verbs, nouns, adjectives etc.). 

Our translator found that the strange 'German' results were actually harder to work with, the frequent contextual errors caused confusion and when less incorrect parts did appear, the temptation was to work around them instead of revising the structure as a whole.

This tendency to work around the 'less bad' machine results had the inevitable effect of producing a translation that was mediocre at best. If our translator had worked from the English source text instead of a machine translation, the translation produced would have been more natural sounding and a lot easier to read.

Most worryingly, the amount of time and effort put into creating this mediocre blended translation was equal to if not greater than that of producing a translation in the normal way.

In short, we believe that attempting to blend machine translation with human translation to save money is a false economy. A lot of work can go into producing mediocre results. If less effort is put in, then the copy is basically unpublishable nonsense.

So what you save in translation costs, you will spend in endless editing and proofreading. 

Machine translation, will it blend? No, but it will definitely shred...

Those who want to know if an iPad will blend, click here