Man and the machine

As the AI bubble continues to inflate (if it hasn’t already burst by the time you read this), many businesses and organisations are trying or have already tried machine translation for their documentation required in another language. This is highly understandable – if you can have a machine produce your translation quickly at (almost) no cost, what’s not to like? Artificial intelligence is unquestionably offering our world some exciting opportunities such as medical applications including designing new antibiotics, right through to the world of physics, predicting potentially disruptive solar flares, and so on. But language is another thing.

A language and a well-written text have dimensions, nuance, depth, suggestion, feeling and more such traits. These are phenomena beyond the capabilities and functioning of a machine. The output of a machine for a translation can only be as good as its input, which is data (legally or illegally) scraped from anywhere and everywhere on the internet. Any quality control of the input is there not. Add the abstract nature of language to this, and it is wise to proceed with the utmost caution when experimenting with machine translation. Another major stumbling block for machine translation is the goodly synonym. Most words have synonyms – the English word “run” has well over 600 meanings listed in the Oxford English Dictionary. The runner up is the word “set” with no fewer than 450 potential interpretations.

When a machine comes up against a synonym for translation it tries to find strings containing the word in an attempt to gain some context, before outputting its best guess without any fear of reprimand. The internet is indeed brimming with mistranslations of synonyms, some highly comical, some downright dangerous. A machine translation can suffice in a small number of cases such as when you just want to know what a few words mean for yourself, or for a simple menu maybe. But when the quality of documentation is of importance, not a few businesses and organisations are apparently also returning to human translation having seen machine translation’s best shot, in the knowledge that if something is worth doing, it’s worth doing right.